Compare commits
52
Commits
| Author | SHA1 | Date | |
|---|---|---|---|
|
|
2f05e70734 | ||
|
|
b5f812270e | ||
|
|
3cce15fa81 | ||
|
|
86ba54b06e | ||
|
|
af91a7e660 | ||
|
|
0270027bf9 | ||
|
|
835bf22b7d | ||
|
|
76a0a726f0 | ||
|
|
df821c8a4a | ||
|
|
f40269caec | ||
|
|
2149d8f3e7 | ||
|
|
e65c8c8961 | ||
|
|
ea8e0def6e | ||
|
|
f8ac498d64 | ||
|
|
276422a85b | ||
|
|
28ef6e9e27 | ||
|
|
f339d0aab7 | ||
|
|
43fccb843c | ||
|
|
f550f1c6b2 | ||
|
|
71f47c3291 | ||
|
|
dc703d1511 | ||
|
|
d70135bf08 | ||
|
|
ff0f43ea2f | ||
|
|
2b2517e4a5 | ||
|
|
b7a4a1a6b0 | ||
|
|
c1a9ad2abc | ||
|
|
c93f77ac19 | ||
|
|
75f3f52589 | ||
|
|
900a2da3e3 | ||
|
|
b7f3664839 | ||
|
|
9511074f40 | ||
|
|
9651a8b2da | ||
|
|
32c66a98c1 | ||
|
|
62437f8964 | ||
|
|
ac4d156cb7 | ||
|
|
95e69276c3 | ||
|
|
56b285d3e8 | ||
|
|
5cc401683c | ||
|
|
79ce72a7b7 | ||
|
|
f40fa6e8e0 | ||
|
|
b015576a13 | ||
|
|
e4fb4cc793 | ||
|
|
1de9a903c8 | ||
|
|
0bb6584437 | ||
|
|
c0bc0d4826 | ||
|
|
efac1b4856 | ||
|
|
bc77afa6bb | ||
|
|
e4373768af | ||
|
|
2cb2e27cdf | ||
|
|
0f4e631cdd | ||
|
|
802c72e4ed | ||
|
|
eb717cc514 |
+9
-15
@@ -22,18 +22,16 @@ Every new head of an open pull request is announced as `pull.updated`, once, and
|
||||
minutes, or when the search fails.
|
||||
- **A pull request nobody checks is said — by the delivery's owner, not here.** `gitea_open_pulls` answers every
|
||||
open pull request from the forge itself, with the statuses its base requires and the ones its head has. The
|
||||
delivery gate asks it every five minutes, and a head on a base requiring `mesh/check` (or hq's
|
||||
`mesh/repo-check`) with none of them for ten minutes since it first saw it so is a delivery held past its bound
|
||||
`mesh-delivery` module asks it every five minutes, and a head on a base requiring `mesh/merge-gate` or
|
||||
`mesh/repo-check` with neither for ten minutes since it first saw it so is a delivery held past its bound
|
||||
(`delivery.<id>.stalled`, state `unannounced`), which the controller raises like any other (issue 347).
|
||||
Each of the head's statuses comes with its state and when it was set, so a head whose check started and never
|
||||
answered — a required check pending, or never set beside one that is — is said as a stalled line in the state
|
||||
`unanswered` once that has lasted the delivery table's bound for a check asked and not answered, an hour, counted
|
||||
from its first sight and again only when a check is asked anew (issue 438).
|
||||
- **How to see it work.** A pull request opened more than a minute after its branch was pushed — this fix's own
|
||||
was one, and the announcer it fixes never saw it — is announced within a pass of the next sweep at the latest, its
|
||||
`mesh/check` is the checker's to set, and `mesh-controller.conditions` names none for it. (Since issue 1301 this
|
||||
module writes `mesh/merge-gate` only where a base still requires it, and none does once the forge requires
|
||||
`mesh/check`.)
|
||||
was one, and the announcer it fixes never saw it — has `mesh/merge-gate` pending within a pass of the next sweep
|
||||
at the latest, and `mesh-controller.conditions` names none for it.
|
||||
- **A pull request or issue number that is not one is refused** before the forge is asked: a tool called with a
|
||||
misnamed argument sent `pulls/NaN`.
|
||||
|
||||
@@ -156,19 +154,15 @@ listed repository:
|
||||
naming the description, and a `merge-check.sh` declaring `# mesh-check-toolchain: go` that vets and runs the unit
|
||||
tests, then the integration tests (`-tags integration`), of every Go module in the tree, and says so and passes on
|
||||
a tree with none. The rule is then narrowed to main's rule of ADR 0398 §2 (no push, only `mesh-tools` merges),
|
||||
requiring `mesh/check` (ADR 0447 §1, issue 1301), and only then is the repository moved into the organisation.
|
||||
requiring `mesh/repo-check` and `mesh/merge-gate`, and only then is the repository moved into the organisation.
|
||||
There main's rule is read back and put right once more (the step `rule-after-move`): the forge drops from a merge
|
||||
list a user without access to the repository, and under the setup account `mesh-tools` has none;
|
||||
- one found still under the setup account is finished: the steps it lacks are done, main is never written twice;
|
||||
- one in the organisation is only put right: main's rule narrowed, `mesh/check` required in place of
|
||||
`mesh/merge-gate` and `mesh/repo-check`, and every other status already there kept. A main that requires
|
||||
`mesh/repo-check` without `mesh/merge-gate` is judged by its own script alone (hq, ADR 0503 §5, until issue 1296)
|
||||
and keeps it; `mesh/check` is never dropped.
|
||||
- one in the organisation is only put right: main's rule narrowed, a missing one of the two statuses added and every
|
||||
status already there kept.
|
||||
|
||||
Every other repository of the organisation whose main requires `mesh/merge-gate` has the two old statuses swapped
|
||||
for `mesh/check`, and only its statuses (issue 1301): the switch reaches every repository the merge gate judged, with
|
||||
no list kept by hand. A step the forge refuses stops that repository, and the others go on. Beyond that swap, a
|
||||
repository taken out of the value is never changed: nothing is deleted, renamed, moved out or unprotected. The pass keeps what it did per repository in
|
||||
A step the forge refuses stops that repository, and the others go on. A repository taken out of the value is never
|
||||
asked about: nothing is deleted, renamed, moved out or unprotected. The pass keeps what it did per repository in
|
||||
`repositories-outcome.json`.
|
||||
|
||||
**How a refusal is seen.** `repositories sync` is a run-once process, declared as the module's last resource, so
|
||||
|
||||
@@ -1,215 +0,0 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"strings"
|
||||
"testing"
|
||||
)
|
||||
|
||||
// The tests of issue 1301 (ADR 0447 §1, §5; ADR 0503 §3, §5; lead decisions L4 and L11): main requires `mesh/check`
|
||||
// alone in place of `mesh/merge-gate` and `mesh/repo-check`; a main that requires `mesh/repo-check` without
|
||||
// `mesh/merge-gate` (hq) keeps it; every repository of the organisation whose main requires `mesh/merge-gate` is
|
||||
// switched, listed in the value or not; no other status is dropped.
|
||||
|
||||
func narrowedRule(contexts ...any) map[string]any {
|
||||
return map[string]any{
|
||||
"enable_push": false, "enable_push_whitelist": false, "push_whitelist_usernames": []any{},
|
||||
"enable_merge_whitelist": true, "merge_whitelist_usernames": []any{tools},
|
||||
"enable_status_check": true, "status_check_contexts": contexts,
|
||||
"block_admin_merge_override": true,
|
||||
}
|
||||
}
|
||||
|
||||
// 1301-S1: a listed repository whose main requires the two old statuses is put right to `mesh/check` in their place,
|
||||
// every other status kept; a second pass changes nothing.
|
||||
func TestListedRepositorySwapsTheTwoStatusesForCheck(t *testing.T) {
|
||||
f := newFake()
|
||||
repo := f.add("novox/mesh-delivery", map[string]map[string]any{mainRule: narrowedRule("mesh/merge-gate", "mesh/repo-check", "other/x")}, "main")
|
||||
s, _ := syncer(t, f)
|
||||
o := run(t, s, Value{"mesh-delivery": {}})
|
||||
got := strs(repo.Rules[mainRule]["status_check_contexts"])
|
||||
if strings.Join(got, ",") != "other/x,mesh/check" {
|
||||
t.Errorf("main requires %v, want other/x and mesh/check", got)
|
||||
}
|
||||
if r := result(o, "mesh-delivery"); r.Did != DidPutRight {
|
||||
t.Errorf("outcome %+v, want put right", r)
|
||||
}
|
||||
before := len(f.writes())
|
||||
run(t, s, Value{"mesh-delivery": {}})
|
||||
if w := f.writes(); len(w) != before {
|
||||
t.Errorf("a second pass wrote %v", w[before:])
|
||||
}
|
||||
}
|
||||
|
||||
// 1301-S1: a repository the program makes requires `mesh/check` alone from its first rule in the organisation on.
|
||||
func TestNewRepositoryRequiresCheckAlone(t *testing.T) {
|
||||
f := newFake()
|
||||
s, _ := syncer(t, f)
|
||||
run(t, s, Value{"fresh": {Description: "d"}})
|
||||
rule := f.repos["novox/fresh"].Rules[mainRule]
|
||||
if got := strs(rule["status_check_contexts"]); rule["enable_status_check"] != true || strings.Join(got, ",") != CheckStatus {
|
||||
t.Errorf("a new repository's main requires %v, want %s alone", got, CheckStatus)
|
||||
}
|
||||
if readme := f.repos["novox/fresh"].Files["README.md"]; !strings.Contains(readme, CheckStatus) ||
|
||||
strings.Contains(readme, "mesh/merge-gate") || strings.Contains(readme, "mesh/repo-check") {
|
||||
t.Errorf("the README names the statuses of before: %q", readme)
|
||||
}
|
||||
}
|
||||
|
||||
// 1301-S1: mesh/check is never dropped, and a rule that requires it beside a retired status loses the retired one.
|
||||
func TestCheckIsKeptAndARetiredStatusBesideItGoes(t *testing.T) {
|
||||
f := newFake()
|
||||
both := f.add("novox/both", map[string]map[string]any{mainRule: narrowedRule("mesh/check", "mesh/repo-check")}, "main")
|
||||
right := f.add("novox/right", map[string]map[string]any{mainRule: narrowedRule("mesh/check")}, "main")
|
||||
s, _ := syncer(t, f)
|
||||
o := run(t, s, Value{"both": {}, "right": {}})
|
||||
if got := strs(both.Rules[mainRule]["status_check_contexts"]); strings.Join(got, ",") != CheckStatus {
|
||||
t.Errorf("both requires %v, want %s alone", got, CheckStatus)
|
||||
}
|
||||
if got := strs(right.Rules[mainRule]["status_check_contexts"]); strings.Join(got, ",") != CheckStatus {
|
||||
t.Errorf("right requires %v", got)
|
||||
}
|
||||
if result(o, "right").Did != DidUnchanged || result(o, "both").Did != DidPutRight {
|
||||
t.Errorf("outcome %+v", o)
|
||||
}
|
||||
}
|
||||
|
||||
// 1301-S1: a listed repository whose main requires nothing is given mesh/check, not the old two.
|
||||
func TestListedRuleRequiringNothingGetsCheck(t *testing.T) {
|
||||
f := newFake()
|
||||
none := f.add("novox/none", map[string]map[string]any{mainRule: {"enable_push": false}}, "main")
|
||||
s, _ := syncer(t, f)
|
||||
run(t, s, Value{"none": {}})
|
||||
if got := strs(none.Rules[mainRule]["status_check_contexts"]); strings.Join(got, ",") != CheckStatus {
|
||||
t.Errorf("none requires %v, want %s", got, CheckStatus)
|
||||
}
|
||||
}
|
||||
|
||||
// 1301-S2 (ADR 0503 §5): a main that requires mesh/repo-check without mesh/merge-gate is judged by its own script
|
||||
// alone (hq): it keeps mesh/repo-check and is not given mesh/check, listed or not.
|
||||
func TestOwnCheckAloneIsKept(t *testing.T) {
|
||||
f := newFake()
|
||||
hq := f.add("novox/hq", map[string]map[string]any{mainRule: narrowedRule("mesh/repo-check")}, "main")
|
||||
lab := f.add("novox/mesh-lab", map[string]map[string]any{mainRule: narrowedRule("mesh/repo-check")}, "main")
|
||||
s, _ := syncer(t, f)
|
||||
o := run(t, s, Value{"hq": {}})
|
||||
for name, repo := range map[string]*fRepo{"hq": hq, "mesh-lab": lab} {
|
||||
if got := strs(repo.Rules[mainRule]["status_check_contexts"]); strings.Join(got, ",") != "mesh/repo-check" {
|
||||
t.Errorf("%s requires %v, want mesh/repo-check alone", name, got)
|
||||
}
|
||||
}
|
||||
if result(o, "hq").Did != DidUnchanged {
|
||||
t.Errorf("hq outcome %+v, want unchanged", result(o, "hq"))
|
||||
}
|
||||
for _, w := range f.writes() {
|
||||
t.Errorf("a repository judged by its own check alone was written: %s", w)
|
||||
}
|
||||
}
|
||||
|
||||
// 1301-S3 (L4, L11): a repository the value does not list whose main requires mesh/merge-gate has the two old
|
||||
// statuses swapped for mesh/check, and only its statuses: nothing else of its rule is sent. One that requires
|
||||
// neither, or mesh/repo-check alone, is not written to.
|
||||
func TestUnlistedRepositoryRequiringTheGateIsSwitched(t *testing.T) {
|
||||
f := newFake()
|
||||
catalog := f.add("novox/mesh-catalog", map[string]map[string]any{mainRule: {
|
||||
"enable_push": true, "enable_status_check": true, "status_check_contexts": []any{"mesh/merge-gate", "mesh/repo-check"},
|
||||
}}, "main")
|
||||
toolsGo := f.add("novox/mesh-tools-go", map[string]map[string]any{mainRule: narrowedRule("mesh/merge-gate")}, "main")
|
||||
f.add("novox/plain", map[string]map[string]any{mainRule: {"enable_push": false}}, "main")
|
||||
f.add("novox/hq", map[string]map[string]any{mainRule: narrowedRule("mesh/repo-check")}, "main")
|
||||
f.add("novox/norule", nil, "main")
|
||||
f.add("mesh-admin/elsewhere", map[string]map[string]any{mainRule: narrowedRule("mesh/merge-gate")}, "main")
|
||||
s, _ := syncer(t, f)
|
||||
o := run(t, s, Value{})
|
||||
|
||||
if got := strs(catalog.Rules[mainRule]["status_check_contexts"]); strings.Join(got, ",") != CheckStatus {
|
||||
t.Errorf("mesh-catalog requires %v, want %s", got, CheckStatus)
|
||||
}
|
||||
if catalog.Rules[mainRule]["enable_push"] != true {
|
||||
t.Errorf("an unlisted repository's rule was narrowed beyond its statuses: %v", catalog.Rules[mainRule])
|
||||
}
|
||||
if got := strs(toolsGo.Rules[mainRule]["status_check_contexts"]); strings.Join(got, ",") != CheckStatus {
|
||||
t.Errorf("mesh-tools-go requires %v, want %s", got, CheckStatus)
|
||||
}
|
||||
want := []string{
|
||||
"PATCH /api/v1/repos/novox/mesh-catalog/branch_protections/main",
|
||||
"PATCH /api/v1/repos/novox/mesh-tools-go/branch_protections/main",
|
||||
}
|
||||
if got := f.writes(); strings.Join(got, "\n") != strings.Join(want, "\n") {
|
||||
t.Errorf("writes:\n%s\nwant:\n%s", strings.Join(got, "\n"), strings.Join(want, "\n"))
|
||||
}
|
||||
for _, n := range []string{"mesh-catalog", "mesh-tools-go"} {
|
||||
if r := result(o, n); r.Did != DidPutRight {
|
||||
t.Errorf("%s outcome %+v, want put right", n, r)
|
||||
}
|
||||
}
|
||||
for _, n := range []string{"plain", "hq", "norule", "elsewhere"} {
|
||||
if r := result(o, n); r.Name != "" {
|
||||
t.Errorf("%s is in the outcome though nothing was done: %+v", n, r)
|
||||
}
|
||||
}
|
||||
before := len(f.writes())
|
||||
run(t, s, Value{})
|
||||
if w := f.writes(); len(w) != before {
|
||||
t.Errorf("a second pass wrote %v", w[before:])
|
||||
}
|
||||
}
|
||||
|
||||
// 1301-S3: the organisation's repositories are read page by page, so one past the first page is switched too.
|
||||
func TestUnlistedRepositoriesAreReadPastTheFirstPage(t *testing.T) {
|
||||
f := newFake()
|
||||
f.pageSize = 2
|
||||
for _, n := range []string{"a", "b", "c", "d", "e"} {
|
||||
f.add("novox/"+n, map[string]map[string]any{mainRule: narrowedRule("mesh/merge-gate", "mesh/repo-check")}, "main")
|
||||
}
|
||||
s, _ := syncer(t, f)
|
||||
run(t, s, Value{})
|
||||
for _, n := range []string{"a", "b", "c", "d", "e"} {
|
||||
if got := strs(f.repos["novox/"+n].Rules[mainRule]["status_check_contexts"]); strings.Join(got, ",") != CheckStatus {
|
||||
t.Errorf("%s requires %v", n, got)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 1301-S3: a refused swap is said with the repository and its step, and the pass goes on to the next.
|
||||
func TestRefusedSwapIsSaid(t *testing.T) {
|
||||
f := newFake()
|
||||
f.add("novox/a", map[string]map[string]any{mainRule: narrowedRule("mesh/merge-gate")}, "main")
|
||||
f.add("novox/b", map[string]map[string]any{mainRule: narrowedRule("mesh/merge-gate")}, "main")
|
||||
f.refuse = func(method, path string) (int, string) {
|
||||
if method == "PATCH" && strings.Contains(path, "/novox/a/") {
|
||||
return 403, "no"
|
||||
}
|
||||
return 0, ""
|
||||
}
|
||||
s, _ := syncer(t, f)
|
||||
o := run(t, s, Value{})
|
||||
if r := result(o, "a"); r.Did != DidRefused || r.Step != StepSwap {
|
||||
t.Errorf("a: %+v, want refused at %s", r, StepSwap)
|
||||
}
|
||||
if r := result(o, "b"); r.Did != DidPutRight {
|
||||
t.Errorf("b: %+v, want put right", r)
|
||||
}
|
||||
}
|
||||
|
||||
// 1301-S3: the organisation's list refused is said, and the listed repositories are still done.
|
||||
func TestOrganisationListRefusedIsSaid(t *testing.T) {
|
||||
f := newFake()
|
||||
f.refuse = func(method, path string) (int, string) {
|
||||
if strings.HasPrefix(path, "/api/v1/orgs/") {
|
||||
return 500, "down"
|
||||
}
|
||||
return 0, ""
|
||||
}
|
||||
s, _ := syncer(t, f)
|
||||
o := run(t, s, Value{"x": {}})
|
||||
if result(o, "x").Did != DidMade {
|
||||
t.Errorf("x: %+v", result(o, "x"))
|
||||
}
|
||||
refusedList := false
|
||||
for _, r := range o.Refused() {
|
||||
refusedList = refusedList || r.Step == StepList
|
||||
}
|
||||
if !refusedList {
|
||||
t.Errorf("the refused list is not in the outcome: %+v", o)
|
||||
}
|
||||
}
|
||||
@@ -11,7 +11,6 @@ import (
|
||||
"path/filepath"
|
||||
"regexp"
|
||||
"sort"
|
||||
"strconv"
|
||||
"strings"
|
||||
"sync"
|
||||
"testing"
|
||||
@@ -44,8 +43,6 @@ type fakeForge struct {
|
||||
calls []string // "METHOD /path"
|
||||
// refuse answers a call with a status and a message when it returns a status above zero.
|
||||
refuse func(method, path string) (int, string)
|
||||
// pageSize is how many repositories one page of the organisation's list holds; 50 when zero.
|
||||
pageSize int
|
||||
}
|
||||
|
||||
func newFake() *fakeForge { return &fakeForge{repos: map[string]*fRepo{}} }
|
||||
@@ -87,7 +84,6 @@ var (
|
||||
branchPath = regexp.MustCompile(`^/api/v1/repos/([^/]+)/([^/]+)/branches/([^/]+)$`)
|
||||
filesPath = regexp.MustCompile(`^/api/v1/repos/([^/]+)/([^/]+)/contents$`)
|
||||
movePath = regexp.MustCompile(`^/api/v1/repos/([^/]+)/([^/]+)/transfer$`)
|
||||
orgRepos = regexp.MustCompile(`^/api/v1/orgs/([^/]+)/repos$`)
|
||||
)
|
||||
|
||||
func strs(v any) []string {
|
||||
@@ -171,31 +167,6 @@ func (f *fakeForge) ServeHTTP(w http.ResponseWriter, r *http.Request) {
|
||||
switch {
|
||||
case p == "/api/v1/version":
|
||||
reply(200, map[string]string{"version": "1.27"})
|
||||
case orgRepos.MatchString(p) && r.Method == http.MethodGet:
|
||||
owner := orgRepos.FindStringSubmatch(p)[1]
|
||||
var names []string
|
||||
for full := range f.repos {
|
||||
if o, n, _ := strings.Cut(full, "/"); o == owner {
|
||||
names = append(names, n)
|
||||
}
|
||||
}
|
||||
sort.Strings(names)
|
||||
size := f.pageSize
|
||||
if size == 0 {
|
||||
size = 50
|
||||
}
|
||||
if l, _ := strconv.Atoi(r.URL.Query().Get("limit")); l > 0 && l < size {
|
||||
size = l
|
||||
}
|
||||
page, _ := strconv.Atoi(r.URL.Query().Get("page"))
|
||||
if page < 1 {
|
||||
page = 1
|
||||
}
|
||||
out := []map[string]any{}
|
||||
for i := (page - 1) * size; i < len(names) && i < page*size; i++ {
|
||||
out = append(out, map[string]any{"name": names[i], "full_name": owner + "/" + names[i]})
|
||||
}
|
||||
reply(200, out)
|
||||
case p == "/api/v1/user/repos" && r.Method == http.MethodPost:
|
||||
name := fmt.Sprint(in["name"])
|
||||
if f.repos[admin+"/"+name] != nil {
|
||||
@@ -341,3 +312,12 @@ func result(o Outcome, name string) Result {
|
||||
}
|
||||
return Result{}
|
||||
}
|
||||
|
||||
func has(list []string, want string) bool {
|
||||
for _, s := range list {
|
||||
if s == want {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
@@ -8,7 +8,7 @@ import "strings"
|
||||
const MergeCheck = `#!/bin/sh
|
||||
# mesh-check-toolchain: go
|
||||
#
|
||||
# This repository's own check (novox/hq ADR 0238 §4), run by the checker within mesh/check (ADR 0447), written with its first commit (ADR 0428 §4):
|
||||
# This repository's own check, mesh/repo-check (novox/hq ADR 0238 §4), written with its first commit (ADR 0428 §4):
|
||||
# every Go module in the tree is vetted, its unit tests run, and then its integration tests (-tags integration).
|
||||
# A tree with no Go module has nothing to test, and the check says so and passes.
|
||||
set -eu
|
||||
@@ -37,7 +37,7 @@ func FirstCommit(name string, r Repo) map[string]string {
|
||||
}
|
||||
readme := "# " + name + "\n\n" + purpose + "\n\n" +
|
||||
"Made by the mesh from the gitea module's config value `repositories` (novox/hq ADR 0428). Main takes no push: " +
|
||||
"a change reaches it by a pull request whose `mesh/check` passed: the checker's verdict on its head merged onto " +
|
||||
"main, this repository's `merge-check.sh` among its tests (novox/hq ADR 0447).\n"
|
||||
"a change reaches it by a pull request whose `mesh/repo-check` (this repository's `merge-check.sh`) and " +
|
||||
"`mesh/merge-gate` passed.\n"
|
||||
return map[string]string{"README.md": readme, "merge-check.sh": MergeCheck}
|
||||
}
|
||||
|
||||
@@ -3,8 +3,7 @@
|
||||
// again whenever the value changes: it reads the file the mesh renders from that value and, with the setup login,
|
||||
// makes every listed repository that does not exist under that login's own account, protects its main, writes
|
||||
// main's first commit, narrows main's rule and only then moves it into the organisation; one that exists only has
|
||||
// its main rule put right, requiring mesh/check in place of the two statuses it replaces (issue 1301). A repository
|
||||
// the value does not list is touched only to swap those two for mesh/check when its main requires mesh/merge-gate. It keeps what it did, per
|
||||
// its main rule put right. A repository the value does not list is never touched. It keeps what it did, per
|
||||
// repository, in the outcome file, says each change on stdout in one line and never a secret, and exits 1 when a
|
||||
// step was refused, its last stderr lines naming each (outcome.go Report): the node-engine's gate on a run-once
|
||||
// step (ADR 0136) reports it and runs it again on every apply until it passes.
|
||||
|
||||
@@ -18,7 +18,7 @@ func TestNewRepositoryRefusesTheAdministratorsOverride(t *testing.T) {
|
||||
}
|
||||
|
||||
// 1141/AC2: a narrowed main rule that still lets an administrator override is not right: it is put right, its
|
||||
// statuses kept (the two old ones as mesh/check, issue 1301), and a second pass changes nothing.
|
||||
// statuses kept, and a second pass changes nothing.
|
||||
func TestNarrowedRuleLettingTheAdministratorOverrideIsPutRight(t *testing.T) {
|
||||
f := newFake()
|
||||
repo := f.add("novox/mesh-delivery", map[string]map[string]any{mainRule: {
|
||||
@@ -33,8 +33,10 @@ func TestNarrowedRuleLettingTheAdministratorOverrideIsPutRight(t *testing.T) {
|
||||
if rule["block_admin_merge_override"] != true {
|
||||
t.Errorf("the administrator's override is still allowed: %v", rule)
|
||||
}
|
||||
if c := strs(rule["status_check_contexts"]); strings.Join(c, ",") != "other/x,mesh/check" {
|
||||
t.Errorf("statuses %v, want other/x kept and mesh/check for the two it replaces (issue 1301)", c)
|
||||
for _, c := range []string{"mesh/merge-gate", "mesh/repo-check", "other/x"} {
|
||||
if !has(strs(rule["status_check_contexts"]), c) {
|
||||
t.Errorf("a status was dropped: %v", rule["status_check_contexts"])
|
||||
}
|
||||
}
|
||||
if r := result(o, "mesh-delivery"); r.Did != DidPutRight {
|
||||
t.Errorf("outcome %+v, want put right", r)
|
||||
|
||||
@@ -26,47 +26,9 @@ const (
|
||||
MainBranch = "main"
|
||||
)
|
||||
|
||||
// CheckStatus is the one status main requires (ADR 0447 §1, issue 1301): the checker's verdict on a pull request's
|
||||
// head merged onto its trunk's head. It replaces GateStatus and OwnCheckStatus, and is never dropped.
|
||||
const CheckStatus = "mesh/check"
|
||||
|
||||
// GateStatus and OwnCheckStatus are the two statuses CheckStatus replaces: the controller's merge gate and the
|
||||
// repository's own merge-check.sh (ADR 0238 §4, ADR 0428 §5 before ADR 0447 §1).
|
||||
const (
|
||||
GateStatus = "mesh/merge-gate"
|
||||
OwnCheckStatus = "mesh/repo-check"
|
||||
)
|
||||
|
||||
// wanted is what a main rule requiring have should require: have with GateStatus and OwnCheckStatus replaced by
|
||||
// CheckStatus, every other status kept in its order (ADR 0447 §1). A rule requiring OwnCheckStatus with neither
|
||||
// GateStatus nor CheckStatus is judged by its own script alone, as hq is until issue 1296 settles how hq is
|
||||
// checked (ADR 0503 §5): the checker writes no status there (ADR 0484 §8), so it is left as it is. This reads the
|
||||
// rule, not a list of names (lead decision L11 of R4e).
|
||||
func wanted(have []string) []string {
|
||||
if !has(have, GateStatus) && !has(have, CheckStatus) && has(have, OwnCheckStatus) {
|
||||
return have
|
||||
}
|
||||
out := []string{}
|
||||
for _, s := range have {
|
||||
if s != GateStatus && s != OwnCheckStatus && s != CheckStatus {
|
||||
out = append(out, s)
|
||||
}
|
||||
}
|
||||
return append(out, CheckStatus)
|
||||
}
|
||||
|
||||
func has(list []string, want string) bool {
|
||||
for _, s := range list {
|
||||
if s == want {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
func sameList(a, b []string) bool {
|
||||
return strings.Join(a, "\x00") == strings.Join(b, "\x00")
|
||||
}
|
||||
// RequiredStatuses are what main requires in every listed repository from its first commit (ADR 0428 §5): the
|
||||
// repository's own check and the module graph's gate. A status already on the rule is kept beside them.
|
||||
var RequiredStatuses = []string{"mesh/merge-gate", "mesh/repo-check"}
|
||||
|
||||
// The steps of one repository, as an outcome names a refused one (ADR 0428 §6).
|
||||
const (
|
||||
@@ -80,10 +42,6 @@ const (
|
||||
// StepRuleAfterMove reads main's rule back once the repository is in the organisation and puts it right: the
|
||||
// forge drops from a merge whitelist a user without access, and under the setup account the tools' have none.
|
||||
StepRuleAfterMove = "rule-after-move"
|
||||
// StepList reads the organisation's repositories, for those the value does not list (issue 1301).
|
||||
StepList = "list"
|
||||
// StepSwap swaps the two old statuses for CheckStatus on a repository the value does not list (issue 1301).
|
||||
StepSwap = "swap"
|
||||
)
|
||||
|
||||
// What a pass did with one repository.
|
||||
@@ -199,73 +157,9 @@ func (s *Syncer) Sync(ctx context.Context, v Value) (Outcome, error) {
|
||||
}
|
||||
out.Repositories = append(out.Repositories, r)
|
||||
}
|
||||
for _, r := range s.switchUnlisted(ctx, v) {
|
||||
if r.Did == DidRefused {
|
||||
s.say("refused: %s/%s at the step %s: %s", s.Org, r.Name, r.Step, r.Said)
|
||||
} else {
|
||||
s.say("%s: %s/%s (%s in place of %s and %s)", r.Did, s.Org, r.Name, CheckStatus, GateStatus, OwnCheckStatus)
|
||||
}
|
||||
out.Repositories = append(out.Repositories, r)
|
||||
}
|
||||
return out, nil
|
||||
}
|
||||
|
||||
// orgPage is how many repositories one read of the organisation's list asks for.
|
||||
const orgPage = 50
|
||||
|
||||
// switchUnlisted swaps the two old statuses for CheckStatus on main's rule of every repository of the organisation
|
||||
// the value does not list whose main requires GateStatus (issue 1301, lead decisions L4 and L11 of R4e): the
|
||||
// repositories the merge gate judges are switched in the delivery that has the checker write mesh/check (ADR 0447
|
||||
// §5), with no list kept by hand. Only the statuses of a rule covering main are sent; nothing else of a rule the
|
||||
// value does not hold is changed (ADR 0428 §7 for all of it but the statuses the mesh itself requires). A main that
|
||||
// does not require GateStatus gets no call. It answers only what it did: a repository switched, or refused.
|
||||
func (s *Syncer) switchUnlisted(ctx context.Context, v Value) []Result {
|
||||
var names []string
|
||||
for page := 1; ; page++ {
|
||||
var repos []struct {
|
||||
Name string `json:"name"`
|
||||
}
|
||||
q := "?limit=" + fmt.Sprint(orgPage) + "&page=" + fmt.Sprint(page)
|
||||
if err := s.Forge.GetJSON(ctx, "/api/v1/orgs/"+url.PathEscape(s.Org)+"/repos"+q, &repos); err != nil {
|
||||
return []Result{refused("(the organisation's repositories)", StepList, err)}
|
||||
}
|
||||
for _, r := range repos {
|
||||
if _, listed := v[r.Name]; !listed && r.Name != "" {
|
||||
names = append(names, r.Name)
|
||||
}
|
||||
}
|
||||
if len(repos) == 0 || page > 1000 {
|
||||
break
|
||||
}
|
||||
}
|
||||
sort.Strings(names)
|
||||
var out []Result
|
||||
for _, n := range names {
|
||||
rules, err := s.rules(ctx, s.Org, n)
|
||||
if err != nil {
|
||||
out = append(out, refused(n, StepRead, err))
|
||||
continue
|
||||
}
|
||||
var did *Result
|
||||
for _, r := range rules {
|
||||
if ok, _ := path.Match(ruleName(r), MainBranch); !ok || !has(required(r), GateStatus) {
|
||||
continue
|
||||
}
|
||||
body := map[string]any{"enable_status_check": true, "status_check_contexts": wanted(required(r))}
|
||||
if err := s.Forge.SendJSON(ctx, http.MethodPatch,
|
||||
repoAPI(s.Org, n)+"/branch_protections/"+url.PathEscape(ruleName(r)), body, nil); err != nil {
|
||||
did = &Result{Name: n, Did: DidRefused, Step: StepSwap, Said: err.Error()}
|
||||
break
|
||||
}
|
||||
did = &Result{Name: n, Did: DidPutRight}
|
||||
}
|
||||
if did != nil {
|
||||
out = append(out, *did)
|
||||
}
|
||||
}
|
||||
return out
|
||||
}
|
||||
|
||||
// waitForForge is cmd/agent-teams' (teams.go waitForForge): the forge starting with the module answers late.
|
||||
func (s *Syncer) waitForForge(ctx context.Context) error {
|
||||
deadline := time.Now().Add(s.WaitFor)
|
||||
@@ -415,7 +309,7 @@ func strList(v any) []string {
|
||||
return out
|
||||
}
|
||||
|
||||
// statuses is what the rules require together, as wanted makes it.
|
||||
// statuses is what a rule requires, with RequiredStatuses added and none dropped.
|
||||
func statuses(rules ...rule) []string {
|
||||
seen := map[string]bool{}
|
||||
var out []string
|
||||
@@ -426,19 +320,16 @@ func statuses(rules ...rule) []string {
|
||||
}
|
||||
}
|
||||
for _, r := range rules {
|
||||
for _, s := range required(r) {
|
||||
add(s)
|
||||
if r != nil && r["enable_status_check"] == true {
|
||||
for _, s := range strList(r["status_check_contexts"]) {
|
||||
add(s)
|
||||
}
|
||||
}
|
||||
}
|
||||
return wanted(out)
|
||||
}
|
||||
|
||||
// required is what a rule requires: its statuses when it checks them, else none.
|
||||
func required(r rule) []string {
|
||||
if r == nil || r["enable_status_check"] != true {
|
||||
return nil
|
||||
for _, s := range RequiredStatuses {
|
||||
add(s)
|
||||
}
|
||||
return strList(r["status_check_contexts"])
|
||||
return out
|
||||
}
|
||||
|
||||
// narrowed is main's rule of ADR 0398 §2 (protection.ts mainRuleBody): no push, only the tools' account merges,
|
||||
@@ -453,7 +344,7 @@ func narrowed(r rule, toolsAccount string, also ...rule) map[string]any {
|
||||
}
|
||||
}
|
||||
|
||||
// isRight says whether a rule already is narrowed and requires what wanted makes of it.
|
||||
// isRight says whether a rule already is narrowed and requires both statuses.
|
||||
func isRight(r rule, toolsAccount string) bool {
|
||||
if r["enable_push"] != false || r["enable_push_whitelist"] == true || r["enable_merge_whitelist"] != true {
|
||||
return false
|
||||
@@ -461,15 +352,26 @@ func isRight(r rule, toolsAccount string) bool {
|
||||
if m := strList(r["merge_whitelist_usernames"]); len(m) != 1 || m[0] != toolsAccount {
|
||||
return false
|
||||
}
|
||||
have := strList(r["status_check_contexts"])
|
||||
return r["enable_status_check"] == true && len(have) > 0 && sameList(have, wanted(have)) &&
|
||||
return r["enable_status_check"] == true && hasAll(strList(r["status_check_contexts"])) &&
|
||||
r["block_admin_merge_override"] == true
|
||||
}
|
||||
|
||||
// putRight makes main's rule in an organisation's repository the narrowed one, keeping every status it requires but
|
||||
// the two CheckStatus replaces (ADR 0447 §1).
|
||||
func hasAll(have []string) bool {
|
||||
for _, want := range RequiredStatuses {
|
||||
found := false
|
||||
for _, h := range have {
|
||||
found = found || h == want
|
||||
}
|
||||
if !found {
|
||||
return false
|
||||
}
|
||||
}
|
||||
return true
|
||||
}
|
||||
|
||||
// putRight makes main's rule in an organisation's repository the narrowed one, keeping every status it requires.
|
||||
// A repository with no rule named main gets one, carrying what a glob rule covering main asks (client.ts
|
||||
// ensureMainMerges, issue 948). It takes away no other status and never unprotects (ADR 0428 §5, §7).
|
||||
// ensureMainMerges, issue 948). It never takes a status away and never unprotects (ADR 0428 §5, §7).
|
||||
func (s *Syncer) putRight(ctx context.Context, owner, name string) (bool, error) {
|
||||
rules, err := s.rules(ctx, owner, name)
|
||||
if err != nil {
|
||||
|
||||
@@ -17,8 +17,8 @@ import (
|
||||
const mainRule = "main"
|
||||
|
||||
// 1124/AC2: a listed repository that does not exist is made under the setup account, given main's rule letting that
|
||||
// login alone push, given its first commit, narrowed to no push and the tools' merge requiring mesh/check (1124/AC5
|
||||
// as issue 1301 changes it), and only then moved into the organisation. Nothing reaches the organisation before the
|
||||
// login alone push, given its first commit, narrowed to no push and the tools' merge requiring mesh/repo-check (and
|
||||
// mesh/merge-gate, 1124/AC5), and only then moved into the organisation. Nothing reaches the organisation before the
|
||||
// narrowed rule, and nothing else is pushed to main.
|
||||
func TestNewRepositoryIsMadeInOrder(t *testing.T) {
|
||||
f := newFake()
|
||||
@@ -67,8 +67,8 @@ func TestNewRepositoryIsMadeInOrder(t *testing.T) {
|
||||
t.Errorf("main's merge whitelist is not the tools' account alone: %v", rule)
|
||||
}
|
||||
checks := strs(rule["status_check_contexts"])
|
||||
if rule["enable_status_check"] != true || strings.Join(checks, ",") != CheckStatus {
|
||||
t.Errorf("main does not require mesh/check alone (1124/AC5, issue 1301): %v", rule)
|
||||
if rule["enable_status_check"] != true || !has(checks, "mesh/repo-check") || !has(checks, "mesh/merge-gate") {
|
||||
t.Errorf("main does not require mesh/repo-check and mesh/merge-gate (1124/AC5): %v", rule)
|
||||
}
|
||||
if r := result(o, "mesh-delivery"); r.Did != DidMade || r.Step != "" {
|
||||
t.Errorf("outcome %+v, want made", r)
|
||||
@@ -87,7 +87,7 @@ func TestNewRepositoryIsMadeInOrder(t *testing.T) {
|
||||
|
||||
// 1124/AC2: the forge drops from a merge whitelist a user without access to the repository, and under the setup
|
||||
// account the tools' account has none; so once the repository is in the organisation its main rule is read back and
|
||||
// put right: the tools' account the only one that merges, mesh/check required.
|
||||
// put right: the tools' account the only one that merges, both statuses required.
|
||||
func TestOrganisationRuleListsTheToolsAfterTheMove(t *testing.T) {
|
||||
f := newFake()
|
||||
s, _ := syncer(t, f)
|
||||
@@ -104,8 +104,8 @@ func TestOrganisationRuleListsTheToolsAfterTheMove(t *testing.T) {
|
||||
t.Errorf("after the move main takes a push: %v", rule)
|
||||
}
|
||||
checks := strs(rule["status_check_contexts"])
|
||||
if rule["enable_status_check"] != true || strings.Join(checks, ",") != CheckStatus {
|
||||
t.Errorf("after the move main does not require mesh/check alone: %v", rule)
|
||||
if rule["enable_status_check"] != true || !has(checks, "mesh/repo-check") || !has(checks, "mesh/merge-gate") {
|
||||
t.Errorf("after the move main does not require both statuses: %v", rule)
|
||||
}
|
||||
if r := result(o, "mesh-delivery"); r.Did != DidMade {
|
||||
t.Errorf("outcome %+v, want made", r)
|
||||
@@ -136,8 +136,7 @@ func TestAdminRuleLetsTheSetupLoginAlonePush(t *testing.T) {
|
||||
}
|
||||
|
||||
// 1124/AC3: a listed repository that exists in the organisation is not made again, only its main rule put right;
|
||||
// a rule still letting the setup login push is narrowed, its statuses kept but the two mesh/check replaces (1124/AC5,
|
||||
// issue 1301).
|
||||
// a rule still letting the setup login push is narrowed, its statuses kept (1124/AC5).
|
||||
func TestExistingOrganisationRepositoryIsOnlyPutRight(t *testing.T) {
|
||||
f := newFake()
|
||||
hq := f.add("novox/hq", map[string]map[string]any{mainRule: {
|
||||
@@ -158,11 +157,14 @@ func TestExistingOrganisationRepositoryIsOnlyPutRight(t *testing.T) {
|
||||
t.Errorf("hq's rule not narrowed: %v", rule)
|
||||
}
|
||||
checks := strs(rule["status_check_contexts"])
|
||||
if strings.Join(checks, ",") != "other/x,mesh/check" {
|
||||
t.Errorf("hq's rule requires %v, want other/x and mesh/check", checks)
|
||||
for _, c := range []string{"other/x", "mesh/merge-gate", "mesh/repo-check"} {
|
||||
if !has(checks, c) {
|
||||
t.Errorf("hq's rule lacks %s: %v", c, checks)
|
||||
}
|
||||
}
|
||||
made := bare.Rules[mainRule]
|
||||
if made == nil || made["enable_push"] != false || strings.Join(strs(made["status_check_contexts"]), ",") != CheckStatus {
|
||||
if made == nil || made["enable_push"] != false || !has(strs(made["status_check_contexts"]), "mesh/repo-check") ||
|
||||
!has(strs(made["status_check_contexts"]), "mesh/merge-gate") {
|
||||
t.Errorf("bare's rule: %v", made)
|
||||
}
|
||||
if hq.Commits != 0 || bare.Commits != 0 {
|
||||
@@ -212,8 +214,8 @@ func TestRepositoryUnderTheSetupAccountIsFinished(t *testing.T) {
|
||||
}
|
||||
}
|
||||
|
||||
// 1124/AC5 as issue 1301 changes it: putting a rule right adds mesh/check when the rule requires mesh/merge-gate,
|
||||
// and keeps every other status there; a rule requiring mesh/repo-check alone keeps it (ADR 0503 §5).
|
||||
// 1124/AC5: putting a rule right adds a missing one of mesh/repo-check and mesh/merge-gate and keeps every status
|
||||
// already there; no pass drops one.
|
||||
func TestStatusesAreAddedAndNeverDropped(t *testing.T) {
|
||||
f := newFake()
|
||||
only := f.add("novox/only", map[string]map[string]any{mainRule: {
|
||||
@@ -222,15 +224,15 @@ func TestStatusesAreAddedAndNeverDropped(t *testing.T) {
|
||||
}}, "main")
|
||||
more := f.add("novox/more", map[string]map[string]any{mainRule: {
|
||||
"enable_push": false, "enable_merge_whitelist": true, "merge_whitelist_usernames": []any{tools},
|
||||
"enable_status_check": true, "status_check_contexts": []any{"mesh/delivery", "mesh/check"},
|
||||
"enable_status_check": true, "status_check_contexts": []any{"mesh/merge-gate", "mesh/repo-check", "mesh/delivery"},
|
||||
"block_admin_merge_override": true, // right since issue 1141
|
||||
}}, "main")
|
||||
s, _ := syncer(t, f)
|
||||
o := run(t, s, Value{"only": {}, "more": {}})
|
||||
if c := strs(only.Rules[mainRule]["status_check_contexts"]); strings.Join(c, ",") != "mesh/repo-check" {
|
||||
if c := strs(only.Rules[mainRule]["status_check_contexts"]); !has(c, "mesh/merge-gate") || !has(c, "mesh/repo-check") {
|
||||
t.Errorf("only: %v", c)
|
||||
}
|
||||
if c := strs(more.Rules[mainRule]["status_check_contexts"]); len(c) != 2 || !has(c, "mesh/delivery") {
|
||||
if c := strs(more.Rules[mainRule]["status_check_contexts"]); len(c) != 3 || !has(c, "mesh/delivery") {
|
||||
t.Errorf("more lost a status: %v", c)
|
||||
}
|
||||
for _, w := range f.writes() {
|
||||
@@ -312,8 +314,7 @@ func TestRefusedStepStopsThatRepositoryOnly(t *testing.T) {
|
||||
}
|
||||
}
|
||||
|
||||
// 1124/AC7: a repository taken out of the value is not changed: its rules are read (issue 1301 switches one whose
|
||||
// main requires mesh/merge-gate), nothing is written, and one still under the setup account is not asked about.
|
||||
// 1124/AC7: a repository taken out of the value gets no call at all.
|
||||
func TestRepositoryOutOfTheValueIsNotTouched(t *testing.T) {
|
||||
f := newFake()
|
||||
f.add("novox/old", map[string]map[string]any{mainRule: {"enable_push": true}}, "main")
|
||||
@@ -322,7 +323,7 @@ func TestRepositoryOutOfTheValueIsNotTouched(t *testing.T) {
|
||||
run(t, s, Value{})
|
||||
run(t, s, Value{"new": {}})
|
||||
for _, c := range f.all() {
|
||||
if strings.Contains(c, "older") || (strings.Contains(c, "old") && !strings.HasPrefix(c, "GET ")) {
|
||||
if strings.Contains(c, "old") {
|
||||
t.Errorf("a repository out of the value was asked: %s", c)
|
||||
}
|
||||
}
|
||||
|
||||
@@ -16,9 +16,6 @@ import { promisify } from "node:util";
|
||||
import type { ToolDefinition } from "@novox/mesh-sdk/tools";
|
||||
import type { CommitStatus } from "./client.js";
|
||||
import { checkName } from "./guard.js";
|
||||
import { CHECK_STATUS } from "./pulls.js";
|
||||
|
||||
export { CHECK_STATUS };
|
||||
|
||||
/** How a command is run: docker on the forge's machine, or a test's. */
|
||||
export type Runner = (file: string, args: string[]) => Promise<{ stdout: string; code: number }>;
|
||||
@@ -94,18 +91,11 @@ const states = new Set(["pending", "success", "error", "failure", "warning"]);
|
||||
/** The merge check's contexts (pulls.ts): a branch's protection may require them. */
|
||||
const mergeCheckContexts = new Set(["mesh/merge-gate", "mesh/repo-check"]);
|
||||
|
||||
|
||||
/** A status mesh-delivery asks for, checked: one of the forge's states, a context of the mesh's own, a
|
||||
* bounded description. */
|
||||
export function deliveryStatus(context: string, state: string, description: string, target?: string) {
|
||||
if (!/^mesh\/[a-z-]+$/.test(context)) throw new Error(`${context} is not a status of the mesh's own`);
|
||||
if (!states.has(state)) throw new Error(`${state} is not a status the forge keeps`);
|
||||
// mesh/check is the checker's verdict and the one status main requires (ADR 0447 §1): no hand sets it in any
|
||||
// state, as issue 293 closed for the two before it. The forge bridge writes it through gitea_check_status, a tool
|
||||
// only its account may call (issue 1301).
|
||||
if (context === CHECK_STATUS) {
|
||||
throw new Error(`${CHECK_STATUS} is the checker's status: the forge bridge writes it through gitea_check_status, never this tool`);
|
||||
}
|
||||
// The merge check's statuses are its verdict's, set when the controller says it, and a branch may require
|
||||
// them: never set to a verdict by hand, so nothing passes — or fails — a merge past the check (novox/hq
|
||||
// issue 293). Pending is no verdict: it is what this holder says on every new head until the verdict comes,
|
||||
@@ -136,7 +126,7 @@ export interface StatusWriters {
|
||||
export function commitStatusTool(writers: StatusWriters): ToolDefinition {
|
||||
return {
|
||||
name: "gitea_commit_status",
|
||||
description: "Set one of the mesh's statuses on a commit (mesh/delivery, mesh/delivery-group): pending, success, error, failure or warning, a short description, and the page it links to. mesh/check, the status main requires, is refused: the checker's forge bridge writes it through gitea_check_status. hq's mesh/repo-check (and mesh/merge-gate, kept for a revert) take pending only — a recheck putting them back until the fresh verdict sets them — on the head's whole sha, in their turn after the check's earlier writes.",
|
||||
description: "Set one of the mesh's statuses on a commit (mesh/delivery, mesh/delivery-group): pending, success, error, failure or warning, a short description, and the page it links to. The merge check's statuses (mesh/merge-gate, mesh/repo-check) take pending only — a recheck putting them back until the fresh verdict sets them — on the head's whole sha, in their turn after the check's earlier writes.",
|
||||
input: {
|
||||
owner: { type: "string", description: "the repository owner" },
|
||||
repo: { type: "string", description: "the repository name" },
|
||||
@@ -163,48 +153,3 @@ export function commitStatusTool(writers: StatusWriters): ToolDefinition {
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
/** What the check's status tool writes with: the forge. */
|
||||
export interface CheckStatusWriter {
|
||||
setCommitStatus(owner: string, repo: string, sha: string, status: CommitStatus): Promise<void>;
|
||||
}
|
||||
|
||||
/**
|
||||
* The `gitea_check_status` tool (novox/hq issue 1301, ADR 0447 §1): the one way `mesh/check` is written. Its arguments
|
||||
* are gitea_commit_status's, so the forge bridge sends an owed status's arguments unchanged to it: owner, repo, sha
|
||||
* (the head's whole sha), state, description, target_url, and context, which may be left out and is mesh/check when
|
||||
* given. **Who may write it is the bus's grant**: the runtime tells a tool no caller, so only a module that lists
|
||||
* `gitea.gitea_check_status` in its manifest's `invokes` may call it, and the forge bridge alone does. Served beside
|
||||
* gitea_commit_status, from the watcher's bundle.
|
||||
*/
|
||||
export function checkStatusTool(writer: CheckStatusWriter): ToolDefinition {
|
||||
return {
|
||||
name: "gitea_check_status",
|
||||
description: "Set mesh/check, the checker's verdict and the one status main requires, on a pull request's head: pending, success, error, failure or warning, a short description, and the page it links to. The checker's forge bridge alone may call it (its manifest's invokes); no other status is written here, and no other tool writes this one.",
|
||||
input: {
|
||||
owner: { type: "string", description: "the repository owner" },
|
||||
repo: { type: "string", description: "the repository name" },
|
||||
sha: { type: "string", description: "the head's whole sha" },
|
||||
context: { type: "string", description: "mesh/check, or left out" },
|
||||
state: { type: "string", description: "pending | success | error | failure | warning" },
|
||||
description: { type: "string", description: "one line" },
|
||||
target_url: { type: "string", description: "the page it links to (the check's report)" },
|
||||
},
|
||||
run: async (args) => {
|
||||
const owner = checkName("owner", args.owner), repo = checkName("name", args.repo), sha = String(args.sha ?? "");
|
||||
if (args.context !== undefined && args.context !== null && String(args.context) !== CHECK_STATUS) {
|
||||
throw new Error(`${JSON.stringify(args.context)}: this tool writes ${CHECK_STATUS} alone. Nothing was sent`);
|
||||
}
|
||||
if (!/^([0-9a-f]{40}|[0-9a-f]{64})$/i.test(sha)) {
|
||||
throw new Error(`${JSON.stringify(args.sha)} is not a head's whole sha: ${CHECK_STATUS} names the head whole. Nothing was sent`);
|
||||
}
|
||||
// deliveryStatus checks the state and cuts the description; it refuses mesh/check, so it is asked for a
|
||||
// context of the mesh's own and the status is named mesh/check after.
|
||||
const checked = deliveryStatus("mesh/delivery", String(args.state), String(args.description ?? ""),
|
||||
args.target_url ? String(args.target_url) : undefined);
|
||||
const status = { ...checked, context: CHECK_STATUS };
|
||||
await writer.setCommitStatus(owner, repo, sha, status);
|
||||
return { set: status };
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
@@ -212,8 +212,7 @@ export class GateWriter {
|
||||
let since = h.pendingAt;
|
||||
if (since === undefined) {
|
||||
const gate = await this.forge.gateOf(owner, repo, sha);
|
||||
// No gate on the forge: none was written (its base does not require it, issue 1301), and nothing is said.
|
||||
if (!gate || gate.state !== "pending") return;
|
||||
if (gate && gate.state !== "pending") return;
|
||||
since = gate?.updated_at ? Date.parse(gate.updated_at) : undefined;
|
||||
}
|
||||
if (since !== undefined && at.getTime() < since - CLOCK_SLACK_MS) return;
|
||||
|
||||
+11
-22
@@ -12,9 +12,7 @@
|
||||
// Consumes mesh-controller.checked — a pull request's merge check, judged — and sets it as the head
|
||||
// commit's statuses: `mesh/merge-gate`, the modules of the mesh's graph the change touches, and
|
||||
// `mesh/repo-check`, the repository's own merge-check.sh (novox/hq ADR 0237 as amended); with a comment
|
||||
// saying why when the gate is not a pass or the repository's own check failed. Of the two, only those the base
|
||||
// still requires are written (issue 1301): main requires `mesh/check` — the checker's, written by its forge bridge
|
||||
// through gitea_check_status — everywhere but hq, whose main requires `mesh/repo-check` alone (ADR 0503 §5).
|
||||
// saying why when the gate is not a pass or the repository's own check failed.
|
||||
//
|
||||
// Consumes node-build-agent.started — a build machine took an ask — and says, on the head a pending check is of,
|
||||
// which machine builds it since when (novox/hq issue 446). This module is the one writer of mesh/merge-gate.
|
||||
@@ -40,12 +38,12 @@
|
||||
|
||||
import { emit, on } from "@novox/mesh-sdk/events";
|
||||
import { registerModuleTools } from "@novox/mesh-sdk/tools";
|
||||
import { checkStatusTool, commitStatusTool } from "./delivery.js";
|
||||
import { commitStatusTool } from "./delivery.js";
|
||||
import { announce, keepWatermark, readWatermark } from "./announce.js";
|
||||
import { mergeTool } from "./merge.js";
|
||||
import { GiteaAdmin, GiteaClient, movedSince } from "./client.js";
|
||||
import { CHECK_CONTEXT, CHECK_STATUS, REPO_CHECK_CONTEXT, commentFor, composedGate, groupGateStatus, headsToAnnounce, maintained,
|
||||
requiredOf, statusesFor, writesGate, type Announced, type Checked, repoCheckFacts, reposToLook } from "./pulls.js";
|
||||
import { CHECK_CONTEXT, REPO_CHECK_CONTEXT, commentFor, composedGate, groupGateStatus, headsToAnnounce, statusesFor, type Announced, type Checked,
|
||||
repoCheckFacts, reposToLook } from "./pulls.js";
|
||||
import { GateWriter, clonedAs, type BuildStarted } from "./gate.js";
|
||||
import { HOOK_PATH, HOOK_PORT, Pace, claimHead, ensureSystemHook, eventsTool, keptSecret, receiver, type ForgeEvent } from "./webhook.js";
|
||||
import { keepSiblings, noteComposed, noteVerdict, readSiblings, siblingsFile, siblingsOf, weighSiblings } from "./siblings.js";
|
||||
@@ -249,11 +247,8 @@ async function announceHead(client: GiteaClient, repo: RepoOf, pull: OpenOf): Pr
|
||||
// Said, so an operator knows the mesh was asked to check it.
|
||||
console.log(`[gitea] announced ${key} at ${sha.slice(0, 8)} to be checked before it merges`);
|
||||
// Pending until the verdict comes, so the pull request says a check is running rather than nothing — and
|
||||
// where it is: in the build queue from now, until a build machine takes it (novox/hq issue 446). Only where the base
|
||||
// still requires mesh/merge-gate (issue 1301): a main that requires mesh/check is the checker's, and hq's requires
|
||||
// mesh/repo-check alone; a rule that cannot be read is said pending, as before.
|
||||
const rule = await client.appliedProtection(repo.owner, repo.name, String(pull.base || "main")).catch(() => undefined);
|
||||
if (writesGate(requiredOf(rule))) await gates.pending(repo.owner, repo.name, sha);
|
||||
// where it is: in the build queue from now, until a build machine takes it (novox/hq issue 446).
|
||||
await gates.pending(repo.owner, repo.name, sha);
|
||||
return true;
|
||||
}
|
||||
type OpenOf = Awaited<ReturnType<GiteaClient["listOpenPullsSince"]>>[number];
|
||||
@@ -262,7 +257,7 @@ type OpenOf = Awaited<ReturnType<GiteaClient["listOpenPullsSince"]>>[number];
|
||||
async function alreadyChecked(client: GiteaClient, repo: RepoOf, pull: OpenOf): Promise<boolean> {
|
||||
try {
|
||||
const checks = await client.headChecks(repo.owner, repo.name, String(pull.head_sha));
|
||||
return CHECK_CONTEXT in checks || REPO_CHECK_CONTEXT in checks || CHECK_STATUS in checks;
|
||||
return CHECK_CONTEXT in checks || REPO_CHECK_CONTEXT in checks;
|
||||
} catch {
|
||||
return false; // the forge cannot say: announced, and the bus's message id collapses it if it was
|
||||
}
|
||||
@@ -399,20 +394,16 @@ async function setVerdict(client: GiteaClient, event: { body: unknown }): Promis
|
||||
// (novox/hq ADR 0323 §4): a check asked before the trunk moved. It is said so, pending with the reason, in place of
|
||||
// its pass or failure; a trunk that cannot be read lets the verdict stand, and the merge tool compares the commits.
|
||||
const trunk = base ?? c.plan?.base;
|
||||
// Only the statuses the base still requires are written (issue 1301): on hq, mesh/repo-check alone.
|
||||
const required = trunk ? requiredOf(await client.appliedProtection(c.owner, c.repo, trunk).catch(() => undefined)) : undefined;
|
||||
const kept = maintained(required);
|
||||
if (c.trunk && trunk) {
|
||||
const head = await client.branchHead(c.owner, c.repo, trunk).catch(() => undefined);
|
||||
if (head && !sameCommit(c.trunk, head)) {
|
||||
const description = outOfDateWords(c.trunk, head);
|
||||
for (const context of [...(composedGate(c) ? [] : [CHECK_CONTEXT]), ...(c["repo-check"] ? [REPO_CHECK_CONTEXT] : [])]
|
||||
.filter((x) => kept.includes(x))) {
|
||||
for (const context of [...(composedGate(c) ? [] : [CHECK_CONTEXT]), ...(c["repo-check"] ? [REPO_CHECK_CONTEXT] : [])]) {
|
||||
await client.setCommitStatus(c.owner, c.repo, c.commit, { state: "pending", context, description,
|
||||
...(target ? { target_url: target } : {}) });
|
||||
}
|
||||
// Its account is kept and shown all the same (ADR 0323 §7): what it judged, and that it is out of date.
|
||||
const said = commentFor(c, await repoCheckFacts(client, c, trunk), required);
|
||||
const said = commentFor(c, await repoCheckFacts(client, c, trunk));
|
||||
if (c.number && !composedGate(c)) await client.addComment(c.owner, c.repo, c.number, lateVerdictComment(c, said, description));
|
||||
console.log(`[gitea] ${c.owner}/${c.repo}#${c.number ?? "?"} at ${c.commit.slice(0, 8)}: merge check ${c.verdict}, ` +
|
||||
`${description}`);
|
||||
@@ -421,10 +412,10 @@ async function setVerdict(client: GiteaClient, event: { body: unknown }): Promis
|
||||
}
|
||||
const facts = await repoCheckFacts(client, c, base ?? c.plan?.base);
|
||||
const gateNow = composedGate(c) ? (await client.headChecks(c.owner, c.repo, c.commit).catch(() => ({} as Record<string, { description?: string }>)))[CHECK_CONTEXT]?.description : undefined;
|
||||
for (const status of statusesFor(c, facts, gateNow, required)) {
|
||||
for (const status of statusesFor(c, facts, gateNow)) {
|
||||
await client.setCommitStatus(c.owner, c.repo, c.commit, target ? { ...status, target_url: target } : status);
|
||||
}
|
||||
const comment = composedGate(c) ? null : commentFor(c, facts, required);
|
||||
const comment = composedGate(c) ? null : commentFor(c, facts);
|
||||
if (comment && c.number) await client.addComment(c.owner, c.repo, c.number, comment);
|
||||
console.log(`[gitea] ${c.owner}/${c.repo}#${c.number ?? "?"} at ${c.commit.slice(0, 8)}: merge check ${c.verdict}`);
|
||||
}
|
||||
@@ -486,8 +477,6 @@ if (!gitea) pace.unheard("the forge is not watched: no way to a token", false);
|
||||
registerModuleTools("gitea", () => [eventsTool(pace), ...(gitea ? [commitStatusTool({
|
||||
setCommitStatus: (owner, repo, sha, status) => (gitea as GiteaClient).setCommitStatus(owner, repo, sha, status),
|
||||
gates,
|
||||
}), checkStatusTool({
|
||||
setCommitStatus: (owner, repo, sha, status) => (gitea as GiteaClient).setCommitStatus(owner, repo, sha, status),
|
||||
}), mergeTool(gitea, { gates })] : [])]);
|
||||
|
||||
// **A build machine took an ask** (the build seat's `started`, novox/hq ADR 0157): said on the head it checks, when
|
||||
|
||||
+10
-41
@@ -4,18 +4,13 @@
|
||||
// when a verdict is judged against an earlier trunk commit than the trunk's head, the tool asks the head's check and
|
||||
// puts its statuses back to pending through that writer, so where the check runs is said and no second check is asked.
|
||||
// Each bundle the runtime serves is its own process; from the tools' bundle the pending bypassed the writer's chain.
|
||||
//
|
||||
// **It judges on the status the base requires** (issue 1301; ADR 0447 §1, §4; ADR 0503 §5): mesh/check, the checker's,
|
||||
// whose line names the trunk it judged ("pass <head12> on <trunk12>"); where a base does not require it, the controller's
|
||||
// status it still requires (hq's mesh/repo-check), as before. A mesh/check of another trunk is refused and nothing is
|
||||
// asked: the checker sets it out of date on the trunk's move and checks it again.
|
||||
|
||||
import { announce } from "./announce.js";
|
||||
import type { ToolDefinition } from "@novox/mesh-sdk/tools";
|
||||
import type { CommitStatus, GiteaClient } from "./client.js";
|
||||
import { mergeableUnder, newestByContext } from "./checks.js";
|
||||
import { MERGE_METHODS, Refused, checkName } from "./guard.js";
|
||||
import { CHECK_CONTEXT, CHECK_STATUS, REPO_CHECK_CONTEXT, maintained, requiredOf } from "./pulls.js";
|
||||
import { CHECK_CONTEXT, REPO_CHECK_CONTEXT } from "./pulls.js";
|
||||
import { approvedHeadsOf, groupRefusal, memberKey, namesSiblings, notApproved, readSiblings, siblingOutOfDate, siblingsFile } from "./siblings.js";
|
||||
import { announcementOf, askedWords, outOfDateWords, sameCommit, tookFile, tookWords, twelve, verdictAtTrunk, waitsForComposed } from "./trunk.js";
|
||||
|
||||
@@ -37,7 +32,7 @@ export function mergeTool(gitea: GiteaClient, given: MergeDeps): ToolDefinition
|
||||
return {
|
||||
name: "gitea_merge_pull_request",
|
||||
description: "Merge a pull request at the head you name — the commit you reviewed — when that is still its head " +
|
||||
"and every status its base branch's protection requires succeeded there (mesh/check; hq's mesh/repo-check), " +
|
||||
"and every status its base branch's protection requires succeeded there (mesh/merge-gate, mesh/repo-check), " +
|
||||
"judged against the trunk's head as it is now, optionally deleting the source branch afterwards. Refused " +
|
||||
"otherwise, with nothing merged: no tool merges past a check, and none merges a head that moved after it was " +
|
||||
"reviewed. A verdict judged against an earlier trunk commit is refused, naming both, and the head's check against " +
|
||||
@@ -113,33 +108,11 @@ export function mergeTool(gitea: GiteaClient, given: MergeDeps): ToolDefinition
|
||||
// head makes — never again while one is on its way. The caller asks again at the reviewed head once it passes.
|
||||
const trunkHead = await gitea.branchHead(owner, repo, pull.base);
|
||||
const checks = await gitea.headChecks(owner, repo, pull.head_sha);
|
||||
const required = requiredOf(rule) ?? [];
|
||||
if (required.includes(CHECK_STATUS)) {
|
||||
// **mesh/check is the checker's** (ADR 0447 §4): it is judged here against the trunk's head, and refused when it
|
||||
// is not current; nothing is written or asked — the checker checks a head again when its trunk moves.
|
||||
const check = checks[CHECK_STATUS];
|
||||
const at = verdictAtTrunk(checks, trunkHead, CHECK_STATUS);
|
||||
if (!check || check.state === "pending") {
|
||||
throw new Refused(`${owner}/${repo}#${number} is not merged: its ${CHECK_STATUS} is ` +
|
||||
`${check ? `pending (${check.description ?? ""})` : "not set yet"}; the checker answers it. Ask again at ` +
|
||||
`${twelve(pull.head_sha)} once it passes against ${pull.base}'s head ${twelve(trunkHead)}. Nothing was merged`);
|
||||
}
|
||||
if (!at.current) {
|
||||
throw new Refused(`${owner}/${repo}#${number} is not merged: its ${CHECK_STATUS} ` +
|
||||
`${at.judged ? `was judged on trunk ${twelve(at.judged)}` : "names no trunk commit it was judged on"}, and ` +
|
||||
`${pull.base} is now ${twelve(trunkHead)}. The checker checks it again against the trunk's head (novox/hq ` +
|
||||
`ADR 0447); this tool asks nothing. Ask again at ${twelve(pull.head_sha)} once it passes. Nothing was merged`);
|
||||
}
|
||||
}
|
||||
// Where the base still requires the controller's statuses (hq's mesh/repo-check; mesh/merge-gate only on a
|
||||
// revert), the first it requires judges, as before (ADR 0323 §5).
|
||||
const kept = maintained(required);
|
||||
const judge = kept[0] ?? CHECK_CONTEXT;
|
||||
const at = kept.length ? verdictAtTrunk(checks, trunkHead, judge) : { current: true, onItsWay: false };
|
||||
const at = verdictAtTrunk(checks, trunkHead);
|
||||
if (!at.current) {
|
||||
// A verdict put out of date by a sibling's move says so in its own words (ADR 0382 §5); it is asked again as a
|
||||
// trunk's move is, and a failed ask puts those words back, not the trunk's.
|
||||
const siblingWords = siblingOutOfDate(checks[judge]?.description) ? checks[judge]?.description : undefined;
|
||||
const siblingWords = siblingOutOfDate(checks[CHECK_CONTEXT]?.description) ? checks[CHECK_CONTEXT]?.description : undefined;
|
||||
const judged = siblingWords ? `was put out of date by a sibling's move (${siblingWords})`
|
||||
: at.judged ? `was judged against trunk ${twelve(at.judged)}` : "names no trunk commit it was judged against";
|
||||
const took = tookWords(tookFile(deps.stateDir), `${owner}/${repo}`);
|
||||
@@ -147,22 +120,21 @@ export function mergeTool(gitea: GiteaClient, given: MergeDeps): ToolDefinition
|
||||
// **A member that waits for its group's composed check is not announced again** (novox/hq issue 1106): its own
|
||||
// check can only say "composed" once more, so an ask here brought a verdict naming no trunk and the next merge was
|
||||
// refused the same way for ever. The composed check is mesh-delivery's to ask; this says whose verdict is awaited.
|
||||
if (waitsForComposed(checks[judge])) {
|
||||
throw new Refused(`${owner}/${repo}#${number} is not merged: its merge gate ${checks[judge]?.description}; the ` +
|
||||
if (waitsForComposed(checks[CHECK_CONTEXT])) {
|
||||
throw new Refused(`${owner}/${repo}#${number} is not merged: its merge gate ${checks[CHECK_CONTEXT]?.description}; the ` +
|
||||
`verdict that names its trunk commit is its group's composed check's, and mesh-delivery asks it, this tool asks no check ` +
|
||||
`(main is now ${now}). Ask again once the group's composed check has passed. Nothing was merged`);
|
||||
}
|
||||
if (at.onItsWay) {
|
||||
throw new Refused(`${owner}/${repo}#${number} is not merged: its verdict ${judged}, and ${pull.base} is now ` +
|
||||
`${now}. Its check is already asked and not yet answered (${checks[judge]?.description}), so it is not ` +
|
||||
`${now}. Its check is already asked and not yet answered (${checks[CHECK_CONTEXT]?.description}), so it is not ` +
|
||||
`asked again; ${took}. Ask again at ${head} once both statuses pass: this tool never merges by itself ` +
|
||||
`(novox/hq ADR 0323). A check that never answers is asked again with mesh-delivery.recheck. Nothing was merged`);
|
||||
}
|
||||
// Pending first, so a second ask finds the check on its way; then the head is announced as a new head is. An
|
||||
// announcement that cannot be made puts the statuses back to out of date, so the next ask asks again rather
|
||||
// than wait for ever on a check nobody asked.
|
||||
// The statuses the base requires of the controller's (issue 1301): hq's mesh/repo-check alone.
|
||||
const contexts = kept.filter((c) => c === judge || (c === REPO_CHECK_CONTEXT && checks[REPO_CHECK_CONTEXT]));
|
||||
const contexts = [CHECK_CONTEXT, ...(checks[REPO_CHECK_CONTEXT] ? [REPO_CHECK_CONTEXT] : [])];
|
||||
// Through the module's one writer of the merge check (gate.ts, novox/hq issue 456): the pending opens a new
|
||||
// check of the head, so where it runs is said ("building on …") and its verdict is not weighed as older.
|
||||
const say = async (description: string) => {
|
||||
@@ -195,12 +167,9 @@ export function mergeTool(gitea: GiteaClient, given: MergeDeps): ToolDefinition
|
||||
// **A member judged beside its siblings merges only with its group passed and reviewed** (novox/hq ADR 0382 §6): the
|
||||
// siblings its verdict names are the holder's record (siblings.ts); a status that says it read siblings with no record
|
||||
// of them is no verdict to merge on, never read as none.
|
||||
// Under mesh/check (issue 1301) the group is the checker's one verdict on every member's head (ADR 0447 §2) and the
|
||||
// Approve's heads above; a record the controller's verdicts left before the switch judges nothing there.
|
||||
const heard = required.includes(CHECK_STATUS) ? undefined : readSiblings(siblingsFile(deps.stateDir))[memberKey(owner, repo, number)];
|
||||
const heard = readSiblings(siblingsFile(deps.stateDir))[memberKey(owner, repo, number)];
|
||||
const siblings = heard && sameCommit(heard.head, pull.head_sha) ? heard.siblings : [];
|
||||
if (!siblings.length && !required.includes(CHECK_STATUS) &&
|
||||
namesSiblings(statuses.find((s) => s.context === CHECK_CONTEXT)?.description)) {
|
||||
if (!siblings.length && namesSiblings(statuses.find((s) => s.context === CHECK_CONTEXT)?.description)) {
|
||||
throw new Refused(`${owner}/${repo}#${number} is not merged: its verdict read siblings, and which ones is not held ` +
|
||||
`here, so the group cannot be weighed. Ask its check again at ${twelve(pull.head_sha)}. Nothing was merged`);
|
||||
}
|
||||
|
||||
@@ -172,7 +172,7 @@
|
||||
},
|
||||
"repositories": {
|
||||
"type": "object",
|
||||
"description": "The repositories of the organisation novox the mesh holds (ADR 0428): each by name, {\"<name>\": {\"private\": true, \"description\": \"...\"}}. One that does not exist is made by the module's setup login under its own account, its main protected and given a first commit (a README naming the description and a merge-check.sh), its rule narrowed to no push and the tools' merge requiring mesh/check, and only then moved into the organisation; one that exists only has main's rule put right, mesh/check in place of mesh/merge-gate and mesh/repo-check and its other statuses kept (a main requiring mesh/repo-check alone keeps it). Every other repository whose main requires mesh/merge-gate has the two swapped for mesh/check; nothing else of a repository out of the value is touched.",
|
||||
"description": "The repositories of the organisation novox the mesh holds (ADR 0428): each by name, {\"<name>\": {\"private\": true, \"description\": \"...\"}}. One that does not exist is made by the module's setup login under its own account, its main protected and given a first commit (a README naming the description and a merge-check.sh), its rule narrowed to no push and the tools' merge requiring mesh/repo-check and mesh/merge-gate, and only then moved into the organisation; one that exists only has main's rule put right, its statuses kept. A repository taken out of the value is never touched.",
|
||||
"trusted": true,
|
||||
"default": {}
|
||||
}
|
||||
|
||||
@@ -1,6 +1,6 @@
|
||||
// A branch's protection (novox/hq ADR 0237): what makes a pull request wait for the mesh's check — `mesh/check`,
|
||||
// the checker's verdict on the head merged onto its trunk (ADR 0447 §1; hq's `mesh/repo-check` until issue 1296) —
|
||||
// before it may merge. Pure, so it is tested without a forge; the client does the asking (client.ts).
|
||||
// A branch's protection (novox/hq ADR 0237): what makes a pull request wait for the mesh's merge check —
|
||||
// `mesh/merge-gate`, the module graph's gate, and `mesh/repo-check`, the repository's own tests — before it
|
||||
// may merge. Pure, so it is tested without a forge; the client does the asking (client.ts).
|
||||
|
||||
import { TOOLS_ACCOUNT } from "./tools-account.js";
|
||||
|
||||
@@ -24,7 +24,7 @@ export interface BranchProtection {
|
||||
/** What a branch's protection is set to. A field not given is left as the rule has it (or the forge's
|
||||
* default on a new rule), except pushes, which a new rule refuses unless `push` says otherwise. */
|
||||
export interface ProtectionWanted {
|
||||
/** The statuses a pull request must have succeeded before it merges, e.g. mesh/check. Empty: none. */
|
||||
/** The statuses a pull request must have succeeded before it merges, e.g. mesh/merge-gate. Empty: none. */
|
||||
statusChecks: string[];
|
||||
/** Whether a person may push to the branch directly. */
|
||||
push?: boolean;
|
||||
|
||||
+7
-45
@@ -19,34 +19,6 @@ export const CHECK_CONTEXT = "mesh/merge-gate";
|
||||
/** The status context of the repository's own merge-check.sh, the check's second layer (novox/hq ADR 0237). */
|
||||
export const REPO_CHECK_CONTEXT = "mesh/repo-check";
|
||||
|
||||
/** The one status main requires (novox/hq ADR 0447 §1, issue 1301): the checker's verdict, written by its forge bridge
|
||||
* through gitea_check_status. Its line names the head and the trunk commit it judged: "pass <head12> on <trunk12> …". */
|
||||
export const CHECK_STATUS = "mesh/check";
|
||||
|
||||
/** The statuses a branch's rule requires, from the rule the forge applies to it: none when it is unprotected or checks
|
||||
* no status; undefined when the rule could not be read. */
|
||||
export function requiredOf(rule: { enable_status_check?: boolean; status_check_contexts?: string[] } | null | undefined): string[] | undefined {
|
||||
if (rule === undefined) return undefined;
|
||||
if (!rule || !rule.enable_status_check) return [];
|
||||
return (rule.status_check_contexts ?? []).filter(Boolean);
|
||||
}
|
||||
|
||||
/**
|
||||
* **The controller's statuses are written only where a base still requires them** (issue 1301; ADR 0503 §3, §5): of
|
||||
* mesh/merge-gate and mesh/repo-check, those the base requires — hq's mesh/repo-check alone once the forge requires
|
||||
* mesh/check everywhere else, so no mesh/merge-gate stands on a head that nothing will set. A base whose rule could not
|
||||
* be read keeps both, as before: a status nothing requires blocks nothing, and one that is required is never left out.
|
||||
*/
|
||||
export function maintained(required: string[] | undefined): string[] {
|
||||
const both = [CHECK_CONTEXT, REPO_CHECK_CONTEXT];
|
||||
return required === undefined ? both : both.filter((c) => required.includes(c));
|
||||
}
|
||||
|
||||
/** Whether mesh/merge-gate is written on a head of this base (maintained). */
|
||||
export function writesGate(required: string[] | undefined): boolean {
|
||||
return maintained(required).includes(CHECK_CONTEXT);
|
||||
}
|
||||
|
||||
/**
|
||||
* The pending words a head's merge check is announced with (novox/hq issue 446): where the check is, and since when,
|
||||
* in the time the operator reads. They stood as "the mesh is mapping this head onto its module graph" for the whole
|
||||
@@ -285,14 +257,8 @@ export function composedGate(c: Checked): boolean {
|
||||
return c.gate?.verdict === "composed";
|
||||
}
|
||||
|
||||
/** Every status a verdict sets: the gate's, and the repository's own check's when it was said; of them, only those the
|
||||
* base requires (`required`, the base's rule: maintained). */
|
||||
export function statusesFor(c: Checked, facts: RepoCheckFacts = {}, gateNow?: string, required?: string[]): CommitStatus[] {
|
||||
const kept = maintained(required);
|
||||
return allStatusesFor(c, facts, gateNow).filter((s) => kept.includes(s.context));
|
||||
}
|
||||
|
||||
function allStatusesFor(c: Checked, facts: RepoCheckFacts = {}, gateNow?: string): CommitStatus[] {
|
||||
/** Every status a verdict sets: the gate's, and the repository's own check's when it was said. */
|
||||
export function statusesFor(c: Checked, facts: RepoCheckFacts = {}, gateNow?: string): CommitStatus[] {
|
||||
if (composedGate(c)) {
|
||||
// A member whose gate the group's composed check judges (ADR 0403 §1): only its repository's check is its own to
|
||||
// set. The gate stays pending, saying it waits for the group, unless a composed verdict already stands on it.
|
||||
@@ -326,13 +292,11 @@ function judgedFirst(c: Checked, status: CommitStatus): CommitStatus {
|
||||
* change that builds something (novox/hq ADR 0238), and why, when the gate is not a pass or the
|
||||
* repository's own check failed or could not run. A repository with no merge-check.sh, touching nothing,
|
||||
* is said by its statuses alone, not by a comment on every push. */
|
||||
export function commentFor(c: Checked, facts: RepoCheckFacts = {}, required?: string[]): string | null {
|
||||
export function commentFor(c: Checked, facts: RepoCheckFacts = {}): string | null {
|
||||
const gate = c.gate ?? { verdict: c.verdict, summary: c.summary };
|
||||
// A base that no longer requires the gate (issue 1301) is told nothing of it: its verdict is the repository's own check.
|
||||
const withGate = writesGate(required);
|
||||
const repo = c["repo-check"];
|
||||
const wrong = repoWrong(repo, facts);
|
||||
if ((!withGate || gate.verdict === "pass") && !wrong && !builds(c.plan)) return null;
|
||||
if (gate.verdict === "pass" && !wrong && !builds(c.plan)) return null;
|
||||
const lines = [`**Merge check** at \`${c.commit.slice(0, 8)}\``, ""];
|
||||
if (c.trunk) {
|
||||
// What was judged (novox/hq ADR 0323 §3): the head merged into the trunk commit, and what was read beside it.
|
||||
@@ -348,11 +312,9 @@ export function commentFor(c: Checked, facts: RepoCheckFacts = {}, required?: st
|
||||
} else if (c.siblings_unknown) {
|
||||
lines.splice(c.trunk ? 2 : 1, 0, `Siblings could not be learned: ${c.siblings_unknown}`);
|
||||
}
|
||||
if (withGate) {
|
||||
lines.push(`- \`${CHECK_CONTEXT}\`: **${saidAs(gate.verdict).toUpperCase()}** — ${gate.summary}` +
|
||||
(gate.modules?.length ? ` (modules: ${gate.modules.join(", ")}` +
|
||||
(gate.dependents?.length ? `; built after them: ${gate.dependents.join(", ")}` : "") + ")" : ""));
|
||||
}
|
||||
lines.push(`- \`${CHECK_CONTEXT}\`: **${saidAs(gate.verdict).toUpperCase()}** — ${gate.summary}` +
|
||||
(gate.modules?.length ? ` (modules: ${gate.modules.join(", ")}` +
|
||||
(gate.dependents?.length ? `; built after them: ${gate.dependents.join(", ")}` : "") + ")" : ""));
|
||||
if (repo) {
|
||||
const st = repoStatus(repo, facts);
|
||||
lines.push(`- \`${REPO_CHECK_CONTEXT}\`: ` + (repo.verdict === "warning" && facts.defined === false
|
||||
|
||||
@@ -61,38 +61,3 @@ test("one view per pull request, found by its marker; only the mesh's statuses",
|
||||
assert.equal(deliveryStatus("mesh/repo-check", "pending", "checking again: main moved on").context, "mesh/repo-check");
|
||||
assert.equal(deliveryStatus("mesh/delivery", "success", "x", "javascript:alert(1)").target_url, undefined);
|
||||
});
|
||||
|
||||
// Issue 1301-S4 (ADR 0447 §1; the lead's choice A): main requires mesh/check, so no hand may set it. The status tool
|
||||
// every caller may use refuses it in every state; it is written only by gitea_check_status, which the forge bridge
|
||||
// alone lists in its manifest's `invokes`, so only its account may call it.
|
||||
test("mesh/check is refused by the status tool and written by the check's own tool alone", async () => {
|
||||
const { commitStatusTool, checkStatusTool, deliveryStatus, CHECK_STATUS } = await import("../delivery.ts");
|
||||
assert.equal(CHECK_STATUS, "mesh/check");
|
||||
const sha = "0123456789abcdef0123456789abcdef01234567";
|
||||
const sent: string[] = [];
|
||||
const setCommitStatus = async (o: string, r: string, s: string, st: any) =>
|
||||
void sent.push(`${o}/${r}@${s.slice(0, 8)} ${st.context} ${st.state} ${st.description}${st.target_url ? " " + st.target_url : ""}`);
|
||||
const gates = { rechecked: async () => { throw new Error("not the gate writer's"); } };
|
||||
for (const state of ["pending", "success", "failure", "error", "warning"]) {
|
||||
assert.throws(() => deliveryStatus("mesh/check", state, "x"), /mesh\/check is the checker's status/);
|
||||
await assert.rejects(commitStatusTool({ setCommitStatus, gates } as any).run({ owner: "novox", repo: "mesh-sdk", sha,
|
||||
context: "mesh/check", state, description: "by hand" }), /gitea_check_status/);
|
||||
}
|
||||
assert.deepEqual(sent, [], "nothing was sent for a hand-set mesh/check");
|
||||
|
||||
const tool = checkStatusTool({ setCommitStatus });
|
||||
assert.equal(tool.name, "gitea_check_status");
|
||||
for (const k of ["owner", "repo", "sha", "state", "description", "target_url", "context"]) assert.ok(k in tool.input, k);
|
||||
// The bridge's arguments as it sends them for gitea_commit_status: the context may be given, and is mesh/check.
|
||||
await tool.run({ owner: "novox", repo: "mesh-sdk", sha, context: "mesh/check", state: "success", description: "pass",
|
||||
target_url: "https://forge.invalid/novox/mesh-sdk/pulls/1" });
|
||||
await tool.run({ owner: "novox", repo: "mesh-sdk", sha, state: "pending", description: "y".repeat(300) });
|
||||
assert.equal(sent[0], "novox/mesh-sdk@01234567 mesh/check success pass https://forge.invalid/novox/mesh-sdk/pulls/1");
|
||||
assert.ok(sent[1].startsWith("novox/mesh-sdk@01234567 mesh/check pending yyyy"), sent[1]);
|
||||
assert.ok(sent[1].length < 200, "the description is cut as every status's is");
|
||||
await assert.rejects(tool.run({ owner: "novox", repo: "mesh-sdk", sha, context: "mesh/delivery", state: "success" }), /mesh\/check alone/);
|
||||
await assert.rejects(tool.run({ owner: "novox", repo: "mesh-sdk", sha: sha.slice(0, 12), state: "success" }), /whole sha/);
|
||||
await assert.rejects(tool.run({ owner: "novox", repo: "mesh-sdk", sha, state: "green" }), /not a status/);
|
||||
await assert.rejects(tool.run({ owner: "../x", repo: "mesh-sdk", sha, state: "success" }));
|
||||
assert.equal(sent.length, 2, "nothing more was sent");
|
||||
});
|
||||
|
||||
@@ -322,13 +322,3 @@ test("a repository as the mesh clones it is told from another", async () => {
|
||||
assert.equal(clonedAs("http://127.0.0.1:20000/novox/mesh-hq.git", "novox", "hq"), false);
|
||||
assert.equal(clonedAs("http://127.0.0.1:20000/novox/mesh-controller.git", "novox", "hq"), false);
|
||||
});
|
||||
|
||||
// Issue 1301: a build machine's start on a head that holds no mesh/merge-gate (its base requires none, so none was written)
|
||||
// says nothing: no "building on" status appears on a main that does not require the gate.
|
||||
test("a start on a head with no gate on the forge writes nothing", async () => {
|
||||
const { GateWriter } = await import("../gate.ts");
|
||||
const forge = aForge();
|
||||
const gates = new GateWriter(forge, announcedAt);
|
||||
assert.equal(await gates.started("novox", "hq", SHA, startedOnShanks), false);
|
||||
assert.equal(forge.kept["mesh/merge-gate"], undefined);
|
||||
});
|
||||
|
||||
@@ -15,7 +15,7 @@ import { GiteaClient } from "../dist/client.js";
|
||||
import { ConfiguredToken } from "../dist/token.js";
|
||||
import { getGiteaTools } from "../dist/tools/index.js";
|
||||
import { mergeTool } from "../dist/merge.js";
|
||||
import { checkStatusTool, commitStatusTool } from "../dist/delivery.js";
|
||||
import { commitStatusTool } from "../dist/delivery.js";
|
||||
import { GateWriter } from "../dist/gate.js";
|
||||
import { judgeApiCall, judgeProtectionChange, ruleCovers, routedPath } from "../dist/guard.js";
|
||||
|
||||
@@ -168,7 +168,7 @@ const gates = { rechecked: (o: string, r: string, sha: string, s: any) => {
|
||||
} };
|
||||
const tools = [...getGiteaTools(client), commitStatusTool({
|
||||
setCommitStatus: (o, r, sha, s) => client.setCommitStatus(o, r, sha, s), gates: writer,
|
||||
}), checkStatusTool({ setCommitStatus: (o, r, sha, s) => client.setCommitStatus(o, r, sha, s) }), mergeTool(client, { gates, stateDir, emit: async (type, body) => { state.emitted.push({ type, body }); },
|
||||
}), mergeTool(client, { gates, stateDir, emit: async (type, body) => { state.emitted.push({ type, body }); },
|
||||
now: () => new Date(2026, 9, 12, 15, 1) })];
|
||||
const tool = (name: string) => tools.find((t) => t.name === name)!;
|
||||
const writes = () => state.sent.map((s) => `${s.method} ${s.path}`);
|
||||
@@ -568,10 +568,6 @@ test("no tool's caller text becomes another path or another user: `..` and `?sud
|
||||
["gitea_get_file", { ...repo, path: "README.md?sudo=jschoubben" }],
|
||||
["gitea_commit_status", { ...repo, sha: "abc1234?sudo=jschoubben", context: "mesh/delivery", state: "success" }],
|
||||
["gitea_commit_status", { ...repo, sha: "../../../admin/users", context: "mesh/delivery", state: "success" }],
|
||||
["gitea_check_status", { ...repo, sha: "abc1234?sudo=jschoubben", state: "success" }],
|
||||
["gitea_check_status", { ...repo, sha: "../../../admin/users", state: "success" }],
|
||||
["gitea_check_status", { owner: "novox", repo: "..", sha: "0123456789abcdef0123456789abcdef01234567", state: "success" }],
|
||||
["gitea_commit_status", { ...repo, sha: "0123456789abcdef0123456789abcdef01234567", context: "mesh/check", state: "success" }],
|
||||
["gitea_commit_statuses", { ...repo, sha: ".." }],
|
||||
["gitea_commit_statuses", { ...repo, sha: "main?sudo=jschoubben" }],
|
||||
["gitea_branch_protection_set", { ...repo, branch: "..", status_checks: "mesh/merge-gate" }],
|
||||
@@ -708,53 +704,3 @@ test("a group of two merges in one pass: the sibling's merge of the composed hea
|
||||
assert.equal(done.merged, true);
|
||||
assert.equal(state.emitted.length, 0, "no check was asked");
|
||||
});
|
||||
|
||||
// **Issue 1301 (ADR 0447 §1, §4; ADR 0503 §3, §5): the merge tool judges on the status the base requires.** A main that
|
||||
// requires mesh/check is judged by the checker's line ("pass <head12> on <trunk12> …"): it merges on a pass that names the
|
||||
// trunk's head, and refuses one that names an older trunk, naming both, writing nothing and asking nothing — a re-check on
|
||||
// a trunk's move is the checker's. No mesh/merge-gate is read or written there.
|
||||
test("gitea_merge_pull_request judges a main requiring mesh/check on mesh/check alone, and asks nothing of it", async () => {
|
||||
const merge = () => tool("gitea_merge_pull_request").run({ owner: "novox", repo: "mesh-catalog", number: 7, head: REVIEWED });
|
||||
state.rules = [{ ...protectedMain(), status_check_contexts: ["mesh/check"] }];
|
||||
const line = (trunk: string, s = "success", words = "pass") =>
|
||||
[{ context: "mesh/check", state: s, description: `${words} ${REVIEWED.slice(0, 12)} on ${trunk.slice(0, 12)} | unit 12 run` }];
|
||||
const before = state.emitted.length;
|
||||
|
||||
state.statuses = line(MOVED);
|
||||
await assert.rejects(merge(), (err: Error) => {
|
||||
assert.match(err.message, /mesh\/check/);
|
||||
assert.match(err.message, new RegExp(`${MOVED.slice(0, 12)}.*${TRUNK.slice(0, 12)}`));
|
||||
assert.match(err.message, /checker checks it again/i);
|
||||
assert.match(err.message, /Nothing was merged/);
|
||||
return true;
|
||||
});
|
||||
state.statuses = line(TRUNK, "pending", "checking");
|
||||
await assert.rejects(merge(), /mesh\/check is pending/);
|
||||
state.statuses = judgedAt(TRUNK); // the old two, current: not what this main requires
|
||||
await assert.rejects(merge(), /mesh\/check/);
|
||||
assert.deepEqual(writes(), [], "nothing is written: mesh/check is the checker's");
|
||||
assert.equal(state.emitted.length, before, "no check is asked");
|
||||
assert.equal(state.merged, false);
|
||||
|
||||
state.statuses = line(TRUNK);
|
||||
assert.equal(((await merge()) as Record<string, unknown>).merged, true);
|
||||
});
|
||||
|
||||
// hq's main requires mesh/repo-check alone (ADR 0503 §5): judged on its words, and an out-of-date one is asked again on
|
||||
// mesh/repo-check alone; no mesh/merge-gate is written on it.
|
||||
test("gitea_merge_pull_request judges a main requiring mesh/repo-check alone on it, and writes no mesh/merge-gate", async () => {
|
||||
const { keepSiblings, siblingsFile } = await import("../dist/siblings.js");
|
||||
keepSiblings(siblingsFile(stateDir), {}); // no sibling record left by the tests before
|
||||
const merge = () => tool("gitea_merge_pull_request").run({ owner: "novox", repo: "mesh-catalog", number: 7, head: REVIEWED });
|
||||
state.rules = [{ ...protectedMain(), status_check_contexts: ["mesh/repo-check"] }];
|
||||
state.statuses = [judgedAt(MOVED)[1]];
|
||||
const before = state.emitted.length;
|
||||
await assert.rejects(merge(), /judged against trunk/);
|
||||
assert.deepEqual(state.sent.filter((s) => s.path.includes("/statuses/")).map((s) => `${s.body.context} ${s.body.state}`),
|
||||
["mesh/repo-check pending"]);
|
||||
assert.equal(state.emitted.length, before + 1, "its check is asked once");
|
||||
await assert.rejects(merge(), /already asked/);
|
||||
state.sent = [];
|
||||
state.statuses = [judgedAt(TRUNK)[1]];
|
||||
assert.equal(((await merge()) as Record<string, unknown>).merged, true);
|
||||
});
|
||||
|
||||
@@ -192,27 +192,3 @@ test("a member's own verdict with a composed gate layer leaves the gate to the c
|
||||
// A verdict with an ordinary gate is as before.
|
||||
assert.equal(composedGate({ ...c, gate: { verdict: "pass", summary: "x" } }), false);
|
||||
});
|
||||
|
||||
// Issue 1301 (ADR 0503 §3, §5): gitea writes the controller's statuses only where the base still requires them. hq's main
|
||||
// requires mesh/repo-check alone: its verdict sets mesh/repo-check and no mesh/merge-gate, and its comment names no gate.
|
||||
test("a base's required statuses decide which of the controller's statuses are written", async () => {
|
||||
const { maintained, writesGate, statusesFor, commentFor, CHECK_STATUS } = await import("../pulls.ts");
|
||||
assert.equal(CHECK_STATUS, "mesh/check");
|
||||
assert.deepEqual(maintained(undefined), ["mesh/merge-gate", "mesh/repo-check"], "a rule not read: as before");
|
||||
assert.deepEqual(maintained(["mesh/repo-check"]), ["mesh/repo-check"]);
|
||||
assert.deepEqual(maintained(["mesh/check", "other/x"]), []);
|
||||
assert.deepEqual(maintained([]), []);
|
||||
assert.equal(writesGate(undefined), true);
|
||||
assert.equal(writesGate(["mesh/repo-check"]), false);
|
||||
assert.equal(writesGate(["mesh/check"]), false);
|
||||
assert.equal(writesGate(["mesh/merge-gate", "mesh/repo-check"]), true);
|
||||
const c = { owner: "novox", repo: "hq", number: 5, commit: "a".repeat(40), id: "x", verdict: "warning", summary: "no script",
|
||||
trunk: "b".repeat(12), "repo-check": { verdict: "pass", summary: "ok" } } as any;
|
||||
const hq = statusesFor(c, {}, undefined, ["mesh/repo-check"]);
|
||||
assert.deepEqual(hq.map((s: any) => s.context), ["mesh/repo-check"]);
|
||||
assert.match(hq[0].description, /merged into trunk bbbbbbbbbbbb/);
|
||||
assert.deepEqual(statusesFor(c).map((s: any) => s.context), ["mesh/merge-gate", "mesh/repo-check"], "unread: as before");
|
||||
assert.deepEqual(statusesFor(c, {}, undefined, ["mesh/check"]), []);
|
||||
const comment = commentFor(c, {}, ["mesh/repo-check"]);
|
||||
assert.ok(comment === null || !comment.includes("mesh/merge-gate"), comment ?? "");
|
||||
});
|
||||
|
||||
@@ -199,21 +199,3 @@ test("a release branch's move puts every open pull request into it out of date,
|
||||
assert.equal(verdictAtTrunk({ "mesh/merge-gate": { state: "success",
|
||||
description: "head x merged into trunk eeeeeeeeeeee — pass" } }, MOVED).current, false);
|
||||
});
|
||||
|
||||
// Issue 1301: a trunk's move puts out of date only the controller's statuses the base still requires; on hq that is
|
||||
// mesh/repo-check alone, judged by its own words; on a main requiring mesh/check, none (the checker's, ADR 0447 §4).
|
||||
test("a trunk's move writes only the statuses the base still requires", async () => {
|
||||
const { outOfDateStatuses, trunkNamed } = await import("../trunk.ts");
|
||||
const old = "1".repeat(40), now = "2".repeat(40);
|
||||
const repo = { state: "success", description: `head ${"a".repeat(12)} merged into trunk ${old.slice(0, 12)} — pass: y` };
|
||||
const gate = { state: "success", description: `head ${"a".repeat(12)} merged into trunk ${old.slice(0, 12)} — pass: x` };
|
||||
assert.deepEqual(outOfDateStatuses({ "mesh/repo-check": repo }, now, ["mesh/repo-check"]).map((s: any) => s.context), ["mesh/repo-check"]);
|
||||
assert.deepEqual(outOfDateStatuses({ "mesh/merge-gate": gate, "mesh/repo-check": repo }, now, ["mesh/repo-check"]).map((s: any) => s.context),
|
||||
["mesh/repo-check"]);
|
||||
assert.deepEqual(outOfDateStatuses({ "mesh/merge-gate": gate, "mesh/repo-check": repo }, now, ["mesh/check"]), []);
|
||||
assert.deepEqual(outOfDateStatuses({ "mesh/merge-gate": gate, "mesh/repo-check": repo }, now).map((s: any) => s.context),
|
||||
["mesh/merge-gate", "mesh/repo-check"], "a rule not read: as before");
|
||||
// The checker's line names its trunk as "<head12> on <trunk12>".
|
||||
assert.equal(trunkNamed(`pass ${"a".repeat(12)} on ${old.slice(0, 12)} | unit 3 run`), old.slice(0, 12));
|
||||
assert.equal(trunkNamed("building on shanks since 10:00"), undefined);
|
||||
});
|
||||
|
||||
@@ -350,14 +350,3 @@ test("a pull request's edit event carries its description, and no other pull req
|
||||
assert.equal(eventOf("pull_request", { action, pull_request: pull, repository }).body.body, undefined, action);
|
||||
}
|
||||
});
|
||||
|
||||
// Issue 1302: a status event carries the status's words, so the delivery gate reads mesh/check's line (the head and the trunk
|
||||
// it judged) from the event itself.
|
||||
test("a status event carries its context, state and words", async () => {
|
||||
const { eventOf } = await import("../webhook.ts");
|
||||
const e = eventOf("status", { sha: "a".repeat(40), context: "mesh/check", state: "success",
|
||||
description: "pass aaaaaaaaaaaa on bbbbbbbbbbbb | unit 3 run, 0 kept", repository: { name: "x", owner: { login: "novox" } } });
|
||||
assert.equal(e.name, "forge.status");
|
||||
assert.equal((e.body as any).context, "mesh/check");
|
||||
assert.equal((e.body as any).description, "pass aaaaaaaaaaaa on bbbbbbbbbbbb | unit 3 run, 0 kept");
|
||||
});
|
||||
|
||||
@@ -395,7 +395,7 @@ function allGiteaTools(gitea: GiteaClient, admin?: () => GiteaAdmin): ToolDefini
|
||||
// ---- Branch protection (novox/hq ADR 0237) ----
|
||||
{
|
||||
name: "gitea_branch_protection_get",
|
||||
description: "A repository's branch protection: the rule for one branch (null when it has none) or every rule — whether direct pushes are refused, which commit statuses a pull request must have succeeded to merge (e.g. mesh/check), approvals, and whether an administrator may merge past them.",
|
||||
description: "A repository's branch protection: the rule for one branch (null when it has none) or every rule — whether direct pushes are refused, which commit statuses a pull request must have succeeded to merge (e.g. mesh/merge-gate), approvals, and whether an administrator may merge past them.",
|
||||
input: {
|
||||
owner: { type: "string", description: "the repository owner" },
|
||||
repo: { type: "string", description: "the repository name" },
|
||||
@@ -413,12 +413,12 @@ function allGiteaTools(gitea: GiteaClient, admin?: () => GiteaAdmin): ToolDefini
|
||||
},
|
||||
{
|
||||
name: "gitea_branch_protection_set",
|
||||
description: "Only ever stronger (novox/hq ADR 0266): a change that drops a required status, lets a person push directly or lets an administrator merge past a status is refused, and on main so is a change that drops the merge whitelist (only mesh-tools merges there) or widens it (novox/hq ADR 0398 §2, issue 948) — that is the operator's, at the forge. Make a branch require commit statuses before a pull request merges into it — the mesh's checker sets `mesh/check` (ADR 0447), and the repositories program requires it on every main. Edits the branch's rule (its required statuses replaced by these, everything else kept unless given) or creates one, which refuses direct pushes unless push=true. Answers the rule before and after.",
|
||||
description: "Only ever stronger (novox/hq ADR 0266): a change that drops a required status, lets a person push directly or lets an administrator merge past a status is refused, and on main so is a change that drops the merge whitelist (only mesh-tools merges there) or widens it (novox/hq ADR 0398 §2, issue 948) — that is the operator's, at the forge. Make a branch require commit statuses before a pull request merges into it — the mesh's merge check sets `mesh/merge-gate` (the module graph's gate) and `mesh/repo-check` (the repository's own tests). Edits the branch's rule (its required statuses replaced by these, everything else kept unless given) or creates one, which refuses direct pushes unless push=true. Answers the rule before and after.",
|
||||
input: {
|
||||
owner: { type: "string", description: "the repository owner" },
|
||||
repo: { type: "string", description: "the repository name" },
|
||||
branch: { type: "string", description: "the branch to protect, e.g. main" },
|
||||
status_checks: { type: "string", description: "comma-separated statuses required to merge, e.g. mesh/check; empty requires none" },
|
||||
status_checks: { type: "string", description: "comma-separated statuses required to merge, e.g. mesh/merge-gate,mesh/repo-check; empty requires none" },
|
||||
push: { type: "boolean", description: "whether a person may push to the branch directly (a new rule refuses it when not given)" },
|
||||
block_admin_override: { type: "boolean", description: "stop an administrator merging past a status that has not succeeded (left as it is when not given)" },
|
||||
},
|
||||
@@ -478,7 +478,7 @@ function allGiteaTools(gitea: GiteaClient, admin?: () => GiteaAdmin): ToolDefini
|
||||
},
|
||||
{
|
||||
name: "gitea_commit_statuses",
|
||||
description: "Read a commit's statuses — or a pull request's head's — as the forge keeps them: the newest of each context (mesh/check, mesh/delivery, …) with its state, description, link, when it was set and by whom; and whether the branch it merges into would let it merge, every status that branch's protection requires being a success. Reads only.",
|
||||
description: "Read a commit's statuses — or a pull request's head's — as the forge keeps them: the newest of each context (mesh/merge-gate, mesh/repo-check, mesh/delivery, …) with its state, description, link, when it was set and by whom; and whether the branch it merges into would let it merge, every status that branch's protection requires being a success. Reads only.",
|
||||
input: {
|
||||
owner: { type: "string", description: "the repository owner" },
|
||||
repo: { type: "string", description: "the repository name" },
|
||||
|
||||
+12
-24
@@ -12,17 +12,12 @@
|
||||
// the moment it is asked, merges nothing when they differ, asks the head's check once — the same announcement a
|
||||
// new head makes — and says how long the last check of that repository took. It never merges by itself after.
|
||||
//
|
||||
// Since issue 1301 (ADR 0447 §4) a main that requires mesh/check is the checker's: its line names the trunk commit it judged
|
||||
// ("pass <head12> on <trunk12>"), the checker sets it out of date when the trunk moves, and the merge tool refuses a verdict
|
||||
// of another trunk and asks nothing. The two statuses above are kept only where a base still requires them (hq's
|
||||
// mesh/repo-check, pulls.ts maintained).
|
||||
//
|
||||
// Pure functions and two small files of the module's own state, so this is tested without a forge.
|
||||
|
||||
import { existsSync, mkdirSync, readFileSync, renameSync, writeFileSync } from "node:fs";
|
||||
import { dirname } from "node:path";
|
||||
import type { CommitStatus, GiteaClient, GiteaPull, HeadCheck } from "./client.js";
|
||||
import { CHECK_CONTEXT, WAITS_FOR_COMPOSED, maintained, requiredOf } from "./pulls.js";
|
||||
import { CHECK_CONTEXT, REPO_CHECK_CONTEXT, WAITS_FOR_COMPOSED } from "./pulls.js";
|
||||
|
||||
/** A commit as a verdict names it: its first twelve characters. */
|
||||
export function twelve(sha: string | undefined): string {
|
||||
@@ -45,9 +40,7 @@ export function judgedWords(head: string, trunk: string): string {
|
||||
* verdict put out of date, the one it had judged ("judged against trunk …"). Undefined when it names none, as a
|
||||
* verdict of a build seat from before ADR 0323 does. */
|
||||
export function trunkNamed(description: string | undefined): string | undefined {
|
||||
const m = /(?:merged into|judged against) trunk ([0-9a-f]{7,64})\b/i.exec(description ?? "")
|
||||
// The checker's mesh/check line (ADR 0447 §3): "<head12> on <trunk12>".
|
||||
?? /\b[0-9a-f]{12} on ([0-9a-f]{12})\b/i.exec(description ?? "");
|
||||
const m = /(?:merged into|judged against) trunk ([0-9a-f]{7,64})\b/i.exec(description ?? "");
|
||||
return m ? m[1].toLowerCase() : undefined;
|
||||
}
|
||||
|
||||
@@ -77,15 +70,15 @@ export function waitsForComposed(gate: HeadCheck | undefined): boolean {
|
||||
* gate, and the repository's own check when the head has one — a pending repo-check a verdict will not set would
|
||||
* stand for ever (novox/hq issue 305's rule).
|
||||
*/
|
||||
export function outOfDateStatuses(checks: Record<string, HeadCheck>, trunkHead: string, required?: string[]): CommitStatus[] {
|
||||
// Only the statuses the base still requires (issue 1301): the first of them the head holds judges.
|
||||
const kept = maintained(required).filter((c) => checks[c]);
|
||||
const judge = kept.length ? checks[kept[0]] : undefined;
|
||||
if (!judge || judge.state === "pending") return [];
|
||||
const judged = trunkNamed(judge.description);
|
||||
export function outOfDateStatuses(checks: Record<string, HeadCheck>, trunkHead: string): CommitStatus[] {
|
||||
const gate = checks[CHECK_CONTEXT];
|
||||
if (!gate || gate.state === "pending") return [];
|
||||
const judged = trunkNamed(gate.description);
|
||||
if (judged && sameCommit(judged, trunkHead)) return [];
|
||||
const description = outOfDateWords(judged, trunkHead);
|
||||
return kept.map((context) => ({ state: "pending" as const, context, description }));
|
||||
const out: CommitStatus[] = [{ state: "pending", context: CHECK_CONTEXT, description }];
|
||||
if (checks[REPO_CHECK_CONTEXT]) out.push({ state: "pending", context: REPO_CHECK_CONTEXT, description });
|
||||
return out;
|
||||
}
|
||||
|
||||
/** Where the merge tool stands with a head's verdict and the trunk's head (ADR 0323 §5). */
|
||||
@@ -98,8 +91,8 @@ export interface VerdictAtTrunk {
|
||||
judged?: string;
|
||||
}
|
||||
|
||||
export function verdictAtTrunk(checks: Record<string, HeadCheck>, trunkHead: string, context = CHECK_CONTEXT): VerdictAtTrunk {
|
||||
const gate = checks[context];
|
||||
export function verdictAtTrunk(checks: Record<string, HeadCheck>, trunkHead: string): VerdictAtTrunk {
|
||||
const gate = checks[CHECK_CONTEXT];
|
||||
const judged = trunkNamed(gate?.description);
|
||||
if (checkOnItsWay(gate)) return { current: false, onItsWay: true, judged };
|
||||
return { current: !!gate && gate.state !== "pending" && !!judged && sameCommit(judged, trunkHead), onItsWay: false, judged };
|
||||
@@ -126,9 +119,6 @@ export function lateVerdictComment(c: { commit: string; trunk?: string; verdict:
|
||||
export interface TrunkForge {
|
||||
listOpenPulls(owner: string, repo: string): Promise<GiteaPull[]>;
|
||||
branchHead(owner: string, repo: string, branch: string): Promise<string>;
|
||||
/** The rule the forge applies to a branch: which of the controller's statuses are kept there (issue 1301). Not given,
|
||||
* or not read: both, as before. */
|
||||
appliedProtection?(owner: string, repo: string, branch: string): Promise<{ enable_status_check?: boolean; status_check_contexts?: string[] } | null>;
|
||||
}
|
||||
|
||||
/** What it writes with: the merge check's one writer (gate.ts), which decides in the head's own turn. */
|
||||
@@ -157,13 +147,11 @@ export async function weighTrunks(forge: TrunkForge, gates: OutOfDateWriter, tru
|
||||
if (before && sameCommit(before, head)) continue;
|
||||
if (before) {
|
||||
let put = 0;
|
||||
const required = forge.appliedProtection
|
||||
? requiredOf(await forge.appliedProtection(owner, name, base).catch(() => undefined)) : undefined;
|
||||
for (const pull of open.filter((p) => p.base === base)) {
|
||||
// The trunk is read again in the head's own turn: a verdict judged against a head newer than the pass's first
|
||||
// read — the trunk moved again meanwhile, and a check answered — is not put out of date.
|
||||
const written = await gates.outOfDate(owner, name, String(pull.head_sha),
|
||||
async (checks) => outOfDateStatuses(checks, await forge.branchHead(owner, name, base), required), pull.html_url || undefined);
|
||||
async (checks) => outOfDateStatuses(checks, await forge.branchHead(owner, name, base)), pull.html_url || undefined);
|
||||
if (written.length) put++;
|
||||
}
|
||||
console.log(`[gitea] ${owner}/${name}'s ${base} moved from ${twelve(before)} to ${twelve(head)}: ${put} open pull ` +
|
||||
|
||||
@@ -127,9 +127,7 @@ function forgeEventOf(said: string, payload: any): ForgeEvent {
|
||||
return { name: kind, body: { ...base, ref: str(payload.ref), ref_type: str(payload.ref_type), sha: str(payload.sha) } };
|
||||
}
|
||||
if (kind === "status") {
|
||||
// Its words too (issue 1302): mesh/check's line names the head and the trunk it judged, which the delivery gate reads.
|
||||
return { name: "status", body: { ...base, sha: str(payload.sha), context: str(payload.context), state: str(payload.state),
|
||||
description: str(payload.description) } };
|
||||
return { name: "status", body: { ...base, sha: str(payload.sha), context: str(payload.context), state: str(payload.state) } };
|
||||
}
|
||||
const name = family === "other" ? `other.${kind.replace(/[^a-z0-9_-]/gi, "_") || "unknown"}` :
|
||||
["fork"].includes(family) ? family : `${family}.${action ?? "changed"}`;
|
||||
|
||||
@@ -1,8 +1,9 @@
|
||||
# kernel
|
||||
|
||||
The kernel's run-time settings on a machine (novox/hq to-be 42 Phase 1, row 11). Today it holds one:
|
||||
**transparent huge pages**, set per machine. The rest of the row (the kernel package, microcode, boot
|
||||
entries) is not built yet.
|
||||
The kernel's run-time settings on a machine (novox/hq to-be 42 Phase 1, row 11). It holds two things:
|
||||
**transparent huge pages**, set per machine, and **the fallback kernel** (`linux-lts` with its headers) and
|
||||
its place on the boot menu (novox/hq ADR 0508 §9 and §10). The default kernel's package, and microcode, are
|
||||
not built yet.
|
||||
|
||||
## Why (novox/hq issue 746)
|
||||
|
||||
@@ -35,16 +36,34 @@ under the new mode. A failed rotation is said in the unit's journal and does not
|
||||
rotation is made while the machine is still starting (`systemctl is-system-running` says `initializing`
|
||||
or `starting`): nothing is in huge pages yet, and the persistent journal opens after this unit.
|
||||
|
||||
## The fallback kernel (novox/hq ADR 0508, issue 1841)
|
||||
|
||||
Every machine keeps a second, older kernel to start from, so a kernel upgrade that does not boot is not the
|
||||
end of the machine. The module declares the packages `linux-lts` and `linux-lts-headers` (so every DKMS module is
|
||||
built for the fallback kernel too) after its boot-entries service, so the entry is in place before the kernel is.
|
||||
|
||||
| | what | notes |
|
||||
|---|---|---|
|
||||
| file | `${dir:state}/boot-entries` | the script. It reads from the machine which loader booted it: the EFI variable `LoaderEntrySelected` present means systemd-boot, a `/boot/grub/grub.cfg` without it means GRUB, neither fails the unit loudly |
|
||||
| | systemd-boot | writes `/boot/loader/entries/mesh-linux-lts.conf` from `/proc/cmdline` without `initrd=` and `BOOT_IMAGE=`, with the microcode images present in `/boot` and the fallback initramfs as initrd lines; `loader.conf` and the default entry are never touched |
|
||||
| | GRUB | runs `grub-mkconfig -o /boot/grub/grub.cfg` only when the set of `/boot/vmlinuz-*` differs from the kernels the menu names, with `GRUB_TOP_LEVEL` naming the kernel the machine booted (`BOOT_IMAGE`; else the non-fallback kernel), since GRUB lists kernels in reverse version order and would otherwise make the fallback kernel the default; `/etc/default/grub` still overrides it when it sets one |
|
||||
| | room | while `/boot/vmlinuz-linux-lts` is not there yet, less than 96 MiB free on `/boot` fails the unit and says so (the fallback kernel writes about 72 MB, issue 1836) |
|
||||
| | a kernel-copying hook | a hook in `/etc/pacman.d/hooks` that the mesh did not write, triggered by kernel images and copying to one fixed file name in `/boot`, would put the fallback kernel in the default kernel's place: it is moved into `${dir:state}/taken-over/` and named in the unit's journal. Each kernel is then placed under its own package's name by the distribution's hooks |
|
||||
| file | `/etc/systemd/system/mesh-kernel-boot-entries.service` | oneshot, sandboxed: the file system read-only but for `/boot`, `/etc/pacman.d/hooks` and the module's state; no home, no network |
|
||||
| file | `/etc/pacman.d/hooks/90-mesh-kernel-boot-entries.hook` | runs the script after a kernel image is installed, upgraded or removed |
|
||||
| packages | `linux-lts`, `linux-lts-headers` | declared after the service and the hook |
|
||||
|
||||
## Tools
|
||||
|
||||
| tool | | answers |
|
||||
|---|---|---|
|
||||
| `kernel_boot` | r | the loader (`systemd-boot`, `grub` or `none`), the menu's entries, the kernels in `/boot`, the entry booted (systemd-boot) or the kernel image booted (GRUB), whether the fallback kernel is on the menu with its kernel in `/boot`, and the room free on `/boot` in KiB |
|
||||
| `kernel_huge_pages` | r | the mode, defrag and shared-memory mode; each per-size control and what `inherit` resolves to; huge pages by kind from `/proc/meminfo` (MiB); the units holding the most file pages in huge folios (`memory.stat` `file_thp`) with their dirty and writeback bytes, khugepaged's collapses and the bytes their cgroup wrote (`io.stat`, the most any one device saw: a mirror and its members each count the same write) |
|
||||
|
||||
## Tests
|
||||
|
||||
`go test ./...`: the tool over a stand-in tree holding the control node's own files of 2026-10-14; the
|
||||
manifest's shape (the default, the order, the unit and its sandbox, reload, restart and health); and the script as the mesh
|
||||
manifest's shape (the default, the order, the unit and its sandbox, reload, restart and health); the boot-entries script run with `sh` over stand-in roots (a systemd-boot entries directory, a GRUB menu, a hook that copies kernels to one fixed name): the entry from the booted command line, a menu generated again only when the kernels changed, no loader, no room, the taken-over hook; `kernel_boot` over the same roots; and the huge-pages script as the mesh
|
||||
writes it, run with `sh` against a stand-in directory: it sets and reads back, keeps what it found once,
|
||||
rotates the journal once and not for `always` nor while the machine starts, goes straight to a new value on a
|
||||
reload without rewriting what it found, gives back what it found and then keeps nothing, refuses a value the kernel
|
||||
|
||||
@@ -0,0 +1,89 @@
|
||||
package main
|
||||
|
||||
// kernel_boot: which loader booted the machine, its entries, the kernels in /boot, the entry it booted and
|
||||
// whether the fallback kernel is on the menu (novox/hq ADR 0508 §9, issue 1841). Reads only.
|
||||
|
||||
import (
|
||||
"os"
|
||||
"path/filepath"
|
||||
"regexp"
|
||||
"sort"
|
||||
"strings"
|
||||
"syscall"
|
||||
"unicode/utf16"
|
||||
)
|
||||
|
||||
const loaderSelected = "sys/firmware/efi/efivars/LoaderEntrySelected-4a67b082-0a4c-41cf-b6c7-440b29bb8c4f"
|
||||
|
||||
// BootState is what kernel_boot answers.
|
||||
type BootState struct {
|
||||
Loader string `json:"loader"`
|
||||
Entries []string `json:"entries"`
|
||||
Kernels []string `json:"kernels"`
|
||||
Booted string `json:"booted"`
|
||||
FallbackPresent bool `json:"fallback_present"`
|
||||
BootFreeKiB int64 `json:"boot_free_kib"`
|
||||
}
|
||||
|
||||
var grubKernel = regexp.MustCompile(`(?m)^\s*linux\s+(?:/boot)?/(vmlinuz-\S+)`)
|
||||
var grubTitle = regexp.MustCompile(`(?m)^\s*menuentry\s+'([^']*)'`)
|
||||
var bootImage = regexp.MustCompile(`BOOT_IMAGE=(?:/boot)?/?(\S+)`)
|
||||
|
||||
// Boot reads the machine, the loader the way the boot-entries script does: the loader's EFI variable means
|
||||
// systemd-boot, a grub.cfg without it GRUB.
|
||||
func (m *Machine) Boot() (*BootState, error) {
|
||||
b := &BootState{Loader: "none", Entries: []string{}, Kernels: []string{}}
|
||||
files, _ := filepath.Glob(m.path("boot/vmlinuz-*"))
|
||||
for _, f := range files {
|
||||
b.Kernels = append(b.Kernels, filepath.Base(f))
|
||||
}
|
||||
sort.Strings(b.Kernels)
|
||||
hasKernel := func(name string) bool {
|
||||
for _, k := range b.Kernels {
|
||||
if k == name {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
var s syscall.Statfs_t
|
||||
if err := syscall.Statfs(m.path("boot"), &s); err == nil {
|
||||
b.BootFreeKiB = int64(s.Bavail) * int64(s.Bsize) / 1024
|
||||
}
|
||||
if raw, err := os.ReadFile(m.path(loaderSelected)); err == nil {
|
||||
b.Loader = "systemd-boot"
|
||||
if len(raw) > 4 {
|
||||
var u []uint16
|
||||
for i := 4; i+1 < len(raw); i += 2 {
|
||||
u = append(u, uint16(raw[i])|uint16(raw[i+1])<<8)
|
||||
}
|
||||
b.Booted = strings.TrimRight(string(utf16.Decode(u)), "\x00")
|
||||
}
|
||||
files, _ := filepath.Glob(m.path("boot/loader/entries/*.conf"))
|
||||
for _, f := range files {
|
||||
b.Entries = append(b.Entries, filepath.Base(f))
|
||||
}
|
||||
sort.Strings(b.Entries)
|
||||
if raw, err := os.ReadFile(m.path("boot/loader/entries/mesh-linux-lts.conf")); err == nil {
|
||||
b.FallbackPresent = strings.Contains(string(raw), "/vmlinuz-linux-lts") && hasKernel("vmlinuz-linux-lts")
|
||||
}
|
||||
return b, nil
|
||||
}
|
||||
if raw, err := os.ReadFile(m.path("boot/grub/grub.cfg")); err == nil {
|
||||
b.Loader = "grub"
|
||||
for _, t := range grubTitle.FindAllStringSubmatch(string(raw), -1) {
|
||||
b.Entries = append(b.Entries, t[1])
|
||||
}
|
||||
for _, k := range grubKernel.FindAllStringSubmatch(string(raw), -1) {
|
||||
if k[1] == "vmlinuz-linux-lts" && hasKernel("vmlinuz-linux-lts") {
|
||||
b.FallbackPresent = true
|
||||
}
|
||||
}
|
||||
if cmd, err := m.read("proc/cmdline"); err == nil {
|
||||
if i := bootImage.FindStringSubmatch(cmd); i != nil {
|
||||
b.Booted = i[1]
|
||||
}
|
||||
}
|
||||
}
|
||||
return b, nil
|
||||
}
|
||||
@@ -0,0 +1,90 @@
|
||||
package main
|
||||
|
||||
// kernel_boot over stand-in roots: a systemd-boot machine with and without the fallback entry, a GRUB machine
|
||||
// with and without the fallback kernel in its menu, and a machine with neither (novox/hq ADR 0508 §9, issue 1841).
|
||||
|
||||
import (
|
||||
"os"
|
||||
"path/filepath"
|
||||
"testing"
|
||||
)
|
||||
|
||||
const selectedVar = "sys/firmware/efi/efivars/LoaderEntrySelected-4a67b082-0a4c-41cf-b6c7-440b29bb8c4f"
|
||||
|
||||
func put(t *testing.T, root, p, content string) {
|
||||
t.Helper()
|
||||
full := filepath.Join(root, p)
|
||||
if err := os.MkdirAll(filepath.Dir(full), 0o755); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
write(t, full, content, 0o644)
|
||||
}
|
||||
|
||||
func bootOf(t *testing.T, root string) *BootState {
|
||||
t.Helper()
|
||||
got, err := (&Machine{Root: root}).Boot()
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
return got
|
||||
}
|
||||
|
||||
func TestKernelBootReadsASystemdBootMachineWithItsFallbackEntry(t *testing.T) {
|
||||
root := t.TempDir()
|
||||
put(t, root, selectedVar, "\x07\x00\x00\x00a\x00r\x00c\x00h\x00.\x00c\x00o\x00n\x00f\x00\x00\x00")
|
||||
put(t, root, "boot/loader/entries/arch.conf", defaultEntry)
|
||||
put(t, root, "boot/loader/entries/mesh-linux-lts.conf", "title Arch Linux (linux-lts, fallback)\nlinux /vmlinuz-linux-lts\n")
|
||||
put(t, root, "boot/vmlinuz-linux", "k")
|
||||
put(t, root, "boot/vmlinuz-linux-lts", "k")
|
||||
got := bootOf(t, root)
|
||||
if got.Loader != "systemd-boot" || got.Booted != "arch.conf" || !got.FallbackPresent {
|
||||
t.Fatalf("%+v", got)
|
||||
}
|
||||
if len(got.Entries) != 2 || got.Entries[0] != "arch.conf" || got.Entries[1] != "mesh-linux-lts.conf" {
|
||||
t.Fatalf("entries: %v", got.Entries)
|
||||
}
|
||||
if len(got.Kernels) != 2 || got.Kernels[0] != "vmlinuz-linux" || got.Kernels[1] != "vmlinuz-linux-lts" {
|
||||
t.Fatalf("kernels: %v", got.Kernels)
|
||||
}
|
||||
if got.BootFreeKiB <= 0 {
|
||||
t.Fatalf("free: %d", got.BootFreeKiB)
|
||||
}
|
||||
}
|
||||
|
||||
func TestKernelBootSaysASystemdBootMachineHasNoFallbackYet(t *testing.T) {
|
||||
root := t.TempDir()
|
||||
put(t, root, selectedVar, "\x07\x00\x00\x00a\x00r\x00c\x00h\x00.\x00c\x00o\x00n\x00f\x00\x00\x00")
|
||||
put(t, root, "boot/loader/entries/arch.conf", defaultEntry)
|
||||
put(t, root, "boot/vmlinuz-linux", "k")
|
||||
// The entry file alone is not the fallback: the kernel it names must be in /boot.
|
||||
put(t, root, "boot/loader/entries/mesh-linux-lts.conf", "linux /vmlinuz-linux-lts\n")
|
||||
if got := bootOf(t, root); got.FallbackPresent {
|
||||
t.Fatalf("a fallback entry without its kernel: %+v", got)
|
||||
}
|
||||
}
|
||||
|
||||
func TestKernelBootReadsAGrubMachine(t *testing.T) {
|
||||
root := t.TempDir()
|
||||
put(t, root, "boot/grub/grub.cfg", grubMenu)
|
||||
put(t, root, "boot/vmlinuz-linux", "k")
|
||||
put(t, root, "proc/cmdline", bootedCmdline)
|
||||
got := bootOf(t, root)
|
||||
if got.Loader != "grub" || got.FallbackPresent || got.Booted != "vmlinuz-linux" {
|
||||
t.Fatalf("%+v", got)
|
||||
}
|
||||
if len(got.Entries) != 2 || got.Entries[0] != "Arch Linux" {
|
||||
t.Fatalf("entries: %v", got.Entries)
|
||||
}
|
||||
put(t, root, "boot/vmlinuz-linux-lts", "k")
|
||||
put(t, root, "boot/grub/grub.cfg", grubMenu+"menuentry 'Arch Linux LTS' {\n\tlinux\t/boot/vmlinuz-linux-lts root=x rw\n}\n")
|
||||
if got := bootOf(t, root); !got.FallbackPresent {
|
||||
t.Fatalf("%+v", got)
|
||||
}
|
||||
}
|
||||
|
||||
func TestKernelBootSaysWhenThereIsNoLoader(t *testing.T) {
|
||||
got := bootOf(t, t.TempDir())
|
||||
if got.Loader != "none" || got.FallbackPresent || len(got.Entries) != 0 {
|
||||
t.Fatalf("%+v", got)
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,284 @@
|
||||
package main
|
||||
|
||||
// The boot-entries script as the mesh writes it, run with sh over a stand-in root (novox/hq ADR 0508 §9 and
|
||||
// §10, issue 1841). The fixtures are the shapes read from machines: a systemd-boot entries directory, a GRUB
|
||||
// menu, a hook that copies every kernel image to the default kernel's name, and the command line a machine
|
||||
// booted with.
|
||||
|
||||
import (
|
||||
"os"
|
||||
"os/exec"
|
||||
"path/filepath"
|
||||
"strings"
|
||||
"testing"
|
||||
)
|
||||
|
||||
const bootedCmdline = "initrd=\\initramfs-linux.img BOOT_IMAGE=/vmlinuz-linux root=UUID=2f6c1e0a-5c0e-4d52-9a57-1d8a2a7a3c11 rw quiet loglevel=3 initrd=\\intel-ucode.img\n"
|
||||
|
||||
const defaultEntry = "title Arch Linux\nlinux /vmlinuz-linux\ninitrd /intel-ucode.img\ninitrd /initramfs-linux.img\noptions root=UUID=2f6c1e0a-5c0e-4d52-9a57-1d8a2a7a3c11 rw quiet\n"
|
||||
|
||||
const loaderConf = "default arch.conf\ntimeout 3\nconsole-mode max\n"
|
||||
|
||||
// grubMenu is the part of a generated grub.cfg that names kernels, for a machine holding one kernel.
|
||||
const grubMenu = `menuentry 'Arch Linux' --class arch {
|
||||
set root='hd0,gpt2'
|
||||
linux /boot/vmlinuz-linux root=UUID=2f6c1e0a-5c0e-4d52-9a57-1d8a2a7a3c11 rw quiet
|
||||
initrd /boot/initramfs-linux.img
|
||||
}
|
||||
menuentry 'Arch Linux, with fallback initramfs' --class arch {
|
||||
linux /boot/vmlinuz-linux root=UUID=2f6c1e0a-5c0e-4d52-9a57-1d8a2a7a3c11 rw quiet
|
||||
initrd /boot/initramfs-linux-fallback.img
|
||||
}
|
||||
`
|
||||
|
||||
// kernelCopyingHook copies every installed kernel image to the default kernel's name on the boot partition.
|
||||
const kernelCopyingHook = `[Trigger]
|
||||
Operation = Install
|
||||
Operation = Upgrade
|
||||
Type = Path
|
||||
Target = usr/lib/modules/*/vmlinuz
|
||||
|
||||
[Action]
|
||||
Description = Copying the kernel to the default name
|
||||
When = PostTransaction
|
||||
Exec = /usr/bin/sh -c '/usr/bin/cp /usr/lib/modules/$(uname -r)/vmlinuz /boot/vmlinuz-linux'
|
||||
`
|
||||
|
||||
// otherHook is a hook of the distribution's shape that has nothing to do with kernel images.
|
||||
const otherHook = `[Trigger]
|
||||
Operation = Upgrade
|
||||
Type = Package
|
||||
Target = pacman-mirrorlist
|
||||
|
||||
[Action]
|
||||
Description = Updating the mirror list
|
||||
When = PostTransaction
|
||||
Exec = /usr/bin/cp /etc/pacman.d/mirrorlist.pacnew /etc/pacman.d/mirrorlist
|
||||
`
|
||||
|
||||
type bootBed struct {
|
||||
script, root, state, bin, mkconfig string
|
||||
need string
|
||||
}
|
||||
|
||||
func newBootBed(t *testing.T) bootBed {
|
||||
t.Helper()
|
||||
dir := t.TempDir()
|
||||
b := bootBed{script: filepath.Join(dir, "boot-entries"), root: filepath.Join(dir, "root"), state: filepath.Join(dir, "state"),
|
||||
bin: filepath.Join(dir, "bin"), mkconfig: filepath.Join(dir, "mkconfig")}
|
||||
content := manifest(t).resource(t, "boot-entries").text(t)
|
||||
content = strings.ReplaceAll(content, "${dir:state}", b.state)
|
||||
if strings.Contains(content, "${") {
|
||||
t.Fatalf("a substitution the test does not make is left in the script")
|
||||
}
|
||||
for _, d := range []string{b.root + "/boot", b.root + "/proc", b.root + "/etc/pacman.d/hooks", b.state, b.bin} {
|
||||
if err := os.MkdirAll(d, 0o755); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
}
|
||||
write(t, b.script, content, 0o755)
|
||||
write(t, b.root+"/proc/cmdline", bootedCmdline, 0o644)
|
||||
// the stand-in grub-mkconfig records the kernel it was told to put first, then its arguments.
|
||||
write(t, filepath.Join(b.bin, "grub-mkconfig"), "#!/bin/sh\necho \"top=$GRUB_TOP_LEVEL $@\" >>"+b.mkconfig+"\n", 0o755)
|
||||
return b
|
||||
}
|
||||
|
||||
func (b bootBed) put(t *testing.T, p, content string) {
|
||||
t.Helper()
|
||||
full := filepath.Join(b.root, p)
|
||||
if err := os.MkdirAll(filepath.Dir(full), 0o755); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
write(t, full, content, 0o644)
|
||||
}
|
||||
|
||||
func (b bootBed) systemdBoot(t *testing.T) {
|
||||
b.put(t, "sys/firmware/efi/efivars/LoaderEntrySelected-4a67b082-0a4c-41cf-b6c7-440b29bb8c4f", "\x07\x00\x00\x00a\x00r\x00c\x00h\x00.\x00c\x00o\x00n\x00f\x00\x00\x00")
|
||||
b.put(t, "boot/loader/loader.conf", loaderConf)
|
||||
b.put(t, "boot/loader/entries/arch.conf", defaultEntry)
|
||||
b.put(t, "boot/vmlinuz-linux", "k")
|
||||
}
|
||||
|
||||
func (b bootBed) run(t *testing.T) (string, error) {
|
||||
t.Helper()
|
||||
args := []string{b.script, "run", b.root, b.state}
|
||||
if b.need != "" {
|
||||
args = append(args, b.need)
|
||||
}
|
||||
cmd := exec.Command("sh", args...)
|
||||
cmd.Env = append(os.Environ(), "PATH="+b.bin+string(os.PathListSeparator)+os.Getenv("PATH"))
|
||||
out, err := cmd.CombinedOutput()
|
||||
return string(out), err
|
||||
}
|
||||
|
||||
func (b bootBed) read(t *testing.T, p string) string {
|
||||
t.Helper()
|
||||
raw, _ := os.ReadFile(filepath.Join(b.root, p))
|
||||
return string(raw)
|
||||
}
|
||||
|
||||
func TestASystemdBootMachineGetsTheFallbackEntryFromItsBootedCommandLine(t *testing.T) {
|
||||
b := newBootBed(t)
|
||||
b.systemdBoot(t)
|
||||
b.put(t, "boot/intel-ucode.img", "u")
|
||||
out, err := b.run(t)
|
||||
if err != nil || !strings.Contains(out, "wrote") {
|
||||
t.Fatalf("%v %s", err, out)
|
||||
}
|
||||
want := "title Arch Linux (linux-lts, fallback)\nlinux /vmlinuz-linux-lts\ninitrd /intel-ucode.img\ninitrd /initramfs-linux-lts.img\n" +
|
||||
"options root=UUID=2f6c1e0a-5c0e-4d52-9a57-1d8a2a7a3c11 rw quiet loglevel=3\n"
|
||||
if got := b.read(t, "boot/loader/entries/mesh-linux-lts.conf"); got != want {
|
||||
t.Fatalf("entry:\n%s\nwant:\n%s", got, want)
|
||||
}
|
||||
// The default entry and the loader's own file are as they were.
|
||||
if b.read(t, "boot/loader/entries/arch.conf") != defaultEntry || b.read(t, "boot/loader/loader.conf") != loaderConf {
|
||||
t.Fatalf("the default entry or loader.conf was touched")
|
||||
}
|
||||
if out, err := b.run(t); err != nil || !strings.Contains(out, "as it was") {
|
||||
t.Fatalf("again: %v %s", err, out)
|
||||
}
|
||||
}
|
||||
|
||||
func TestWithoutMicrocodeThereIsNoMicrocodeLine(t *testing.T) {
|
||||
b := newBootBed(t)
|
||||
b.systemdBoot(t)
|
||||
if out, err := b.run(t); err != nil {
|
||||
t.Fatalf("%v %s", err, out)
|
||||
}
|
||||
if got := b.read(t, "boot/loader/entries/mesh-linux-lts.conf"); strings.Contains(got, "ucode") || !strings.Contains(got, "initrd /initramfs-linux-lts.img") {
|
||||
t.Fatalf("entry: %s", got)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAGrubMachineIsGeneratedAgainOnlyWhenItsKernelsChanged(t *testing.T) {
|
||||
b := newBootBed(t)
|
||||
b.put(t, "boot/grub/grub.cfg", grubMenu)
|
||||
b.put(t, "boot/vmlinuz-linux", "k")
|
||||
if out, err := b.run(t); err != nil || !strings.Contains(out, "left as it is") {
|
||||
t.Fatalf("same: %v %s", err, out)
|
||||
}
|
||||
if _, err := os.Stat(b.mkconfig); err == nil {
|
||||
t.Fatalf("the menu was generated again with the same kernels")
|
||||
}
|
||||
b.put(t, "boot/vmlinuz-linux-lts", "k")
|
||||
if out, err := b.run(t); err != nil || !strings.Contains(out, "generated again") {
|
||||
t.Fatalf("changed: %v %s", err, out)
|
||||
}
|
||||
raw, _ := os.ReadFile(b.mkconfig)
|
||||
if string(raw) != "top=/boot/vmlinuz-linux -o "+filepath.Join(b.root, "boot/grub/grub.cfg")+"\n" {
|
||||
t.Fatalf("grub-mkconfig was asked %q", raw)
|
||||
}
|
||||
}
|
||||
|
||||
// ADR 0508 §9: GRUB's menu lists kernels in reverse version order, so linux-lts would come first and become the
|
||||
// default entry (GRUB_DEFAULT=0); the kernel the machine booted is put first, and the fallback kernel never.
|
||||
func TestGrubKeepsTheBootedKernelFirstAndNeverTheFallback(t *testing.T) {
|
||||
for name, c := range map[string]struct{ cmdline, want string }{
|
||||
"booted from the default kernel": {"BOOT_IMAGE=/vmlinuz-linux root=UUID=x rw\n", "top=/boot/vmlinuz-linux"},
|
||||
"booted from a grub device path": {"BOOT_IMAGE=(hd0,msdos1)/vmlinuz-linux-zen root=UUID=x rw\n", "top=/boot/vmlinuz-linux-zen"},
|
||||
"booted from the fallback kernel": {"BOOT_IMAGE=/vmlinuz-linux-lts root=UUID=x rw\n", "top=/boot/vmlinuz-linux"},
|
||||
"no image on the command line": {"root=UUID=x rw\n", "top=/boot/vmlinuz-linux"},
|
||||
"an image no longer in /boot": {"BOOT_IMAGE=/vmlinuz-linux-gone root=UUID=x rw\n", "top=/boot/vmlinuz-linux"},
|
||||
} {
|
||||
b := newBootBed(t)
|
||||
b.put(t, "boot/grub/grub.cfg", grubMenu)
|
||||
b.put(t, "proc/cmdline", c.cmdline)
|
||||
for _, k := range []string{"vmlinuz-linux", "vmlinuz-linux-lts", "vmlinuz-linux-zen"} {
|
||||
b.put(t, "boot/"+k, "k")
|
||||
}
|
||||
if out, err := b.run(t); err != nil {
|
||||
t.Fatalf("%s: %v %s", name, err, out)
|
||||
}
|
||||
raw, _ := os.ReadFile(b.mkconfig)
|
||||
if !strings.HasPrefix(string(raw), c.want+" -o ") {
|
||||
t.Errorf("%s: grub-mkconfig was asked %q, want %s first", name, raw, c.want)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestAMachineWithNoLoaderFailsLoudly(t *testing.T) {
|
||||
b := newBootBed(t)
|
||||
out, err := b.run(t)
|
||||
if err == nil || !strings.Contains(out, "no boot loader found") {
|
||||
t.Fatalf("%v %s", err, out)
|
||||
}
|
||||
}
|
||||
|
||||
func TestABootPartitionWithoutRoomFailsBeforeAnythingIsWritten(t *testing.T) {
|
||||
b := newBootBed(t)
|
||||
b.systemdBoot(t)
|
||||
b.need = "999999999999"
|
||||
out, err := b.run(t)
|
||||
if err == nil || !strings.Contains(out, "free room there before the kernel is installed") {
|
||||
t.Fatalf("%v %s", err, out)
|
||||
}
|
||||
if b.read(t, "boot/loader/entries/mesh-linux-lts.conf") != "" {
|
||||
t.Fatalf("an entry was written")
|
||||
}
|
||||
// Once the fallback kernel is in place the room is not asked for again.
|
||||
b.put(t, "boot/vmlinuz-linux-lts", "k")
|
||||
if out, err := b.run(t); err != nil {
|
||||
t.Fatalf("%v %s", err, out)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAHookThatCopiesKernelsToOneFixedNameIsMovedIntoTheStateAndNamed(t *testing.T) {
|
||||
b := newBootBed(t)
|
||||
b.systemdBoot(t)
|
||||
b.put(t, "etc/pacman.d/hooks/60-copy-kernel.hook", kernelCopyingHook)
|
||||
b.put(t, "etc/pacman.d/hooks/70-mirrorlist.hook", otherHook)
|
||||
b.put(t, "etc/pacman.d/hooks/90-mesh-kernel-boot-entries.hook", "[Trigger]\nTarget = usr/lib/modules/*/vmlinuz\n[Action]\nExec = /bin/cp x /boot/y\n")
|
||||
out, err := b.run(t)
|
||||
if err != nil || !strings.Contains(out, "taken over: the pacman hook 60-copy-kernel.hook") {
|
||||
t.Fatalf("%v %s", err, out)
|
||||
}
|
||||
if b.read(t, "etc/pacman.d/hooks/60-copy-kernel.hook") != "" {
|
||||
t.Fatalf("the copying hook is still in the hook directory")
|
||||
}
|
||||
if raw, _ := os.ReadFile(filepath.Join(b.state, "taken-over", "60-copy-kernel.hook")); string(raw) != kernelCopyingHook {
|
||||
t.Fatalf("moved hook: %q", raw)
|
||||
}
|
||||
if b.read(t, "etc/pacman.d/hooks/70-mirrorlist.hook") != otherHook || b.read(t, "etc/pacman.d/hooks/90-mesh-kernel-boot-entries.hook") == "" {
|
||||
t.Fatalf("a hook that copies no kernel was moved")
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheBootEntriesUnitAndHookAreTheModules(t *testing.T) {
|
||||
m := manifest(t)
|
||||
unit := m.resource(t, "boot-unit").text(t)
|
||||
for _, line := range []string{"Type=oneshot", "RemainAfterExit=yes", "ExecStart=${dir:state}/boot-entries run", "ProtectSystem=strict",
|
||||
"ReadWritePaths=/boot ${dir:state} /etc/pacman.d/hooks", "NoNewPrivileges=yes", "WantedBy=multi-user.target"} {
|
||||
if !strings.Contains(unit, line+"\n") {
|
||||
t.Errorf("the unit lacks %q", line)
|
||||
}
|
||||
}
|
||||
hook := m.resource(t, "boot-hook")
|
||||
text := hook.text(t)
|
||||
if hook["path"] != "/etc/pacman.d/hooks/90-mesh-kernel-boot-entries.hook" || !strings.Contains(text, "Target = usr/lib/modules/*/vmlinuz\n") ||
|
||||
!strings.Contains(text, "When = PostTransaction\n") || !strings.Contains(text, "Exec = ${dir:state}/boot-entries run\n") {
|
||||
t.Fatalf("the hook: %v\n%s", hook["path"], text)
|
||||
}
|
||||
apply := m.resource(t, "boot-apply")
|
||||
if apply["unit"] != "mesh-kernel-boot-entries.service" || apply["state"] != "running" || apply["boot"] != "enabled" {
|
||||
t.Fatalf("apply: %v", apply)
|
||||
}
|
||||
}
|
||||
|
||||
// The fallback kernel and its headers are declared after the boot-entries service, so the entry is placed
|
||||
// before the kernel is installed and every DKMS module is built for it (ADR 0508 §9).
|
||||
func TestTheFallbackKernelAndItsHeadersAreDeclaredAfterTheBootEntriesUnit(t *testing.T) {
|
||||
m := manifest(t)
|
||||
at := map[string]int{}
|
||||
for i, r := range m.Resources {
|
||||
at[r["id"].(string)] = i
|
||||
}
|
||||
for id, pkg := range map[string]string{"linux-lts": "linux-lts", "linux-lts-headers": "linux-lts-headers"} {
|
||||
r := m.resource(t, id)
|
||||
if r["type"] != "package" || r["package"] != pkg || r["absent"] == true {
|
||||
t.Errorf("%s: %v", id, r)
|
||||
}
|
||||
if at[id] < at["boot-apply"] || at[id] < at["boot-hook"] {
|
||||
t.Errorf("%s is declared before the boot-entries unit and hook", id)
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -46,5 +46,15 @@ func tools(m *Machine) []stdio.Tool {
|
||||
return m.HugePages(n)
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "kernel_boot",
|
||||
Description: "How this machine boots and whether the fallback kernel is on its menu (novox/hq ADR 0508 §9): the loader that booted it " +
|
||||
"(systemd-boot, read from the EFI variable it sets, grub, or none), the menu's entries, the kernels in /boot, the entry " +
|
||||
"the machine booted (systemd-boot) or the kernel image it booted (grub), whether the fallback kernel linux-lts is on " +
|
||||
"the menu with its kernel in /boot, and the room free on /boot in KiB (the fallback kernel needs about 72 MB). " +
|
||||
"Replaces ls /boot, bootctl list, reading grub.cfg and df /boot. (r)",
|
||||
Input: schema(map[string]any{}),
|
||||
Run: func(args map[string]any) (any, error) { return m.Boot() },
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
@@ -24,7 +24,7 @@ func TestTheDefaultIsMadviseAndTheUnitFollowsTheSetting(t *testing.T) {
|
||||
for _, r := range m.Resources {
|
||||
ids = append(ids, r["id"].(string))
|
||||
}
|
||||
if strings.Join(ids, " ") != "state huge-pages unit apply" {
|
||||
if strings.Join(ids, " ") != "state huge-pages unit apply boot-entries boot-unit boot-hook boot-apply linux-lts linux-lts-headers" {
|
||||
t.Fatalf("order: %v", ids)
|
||||
}
|
||||
script := m.resource(t, "huge-pages")
|
||||
|
||||
@@ -0,0 +1,13 @@
|
||||
# Module kernel, novox/hq ADR 0508 §9, issue 1841. After a kernel image is installed, upgraded or removed, the
|
||||
# fallback kernel's boot entry (or the GRUB menu) is placed again. Written by the mesh; change the module.
|
||||
[Trigger]
|
||||
Operation = Install
|
||||
Operation = Upgrade
|
||||
Operation = Remove
|
||||
Type = Path
|
||||
Target = usr/lib/modules/*/vmlinuz
|
||||
|
||||
[Action]
|
||||
Description = Placing the fallback kernel on the boot menu
|
||||
When = PostTransaction
|
||||
Exec = ${dir:state}/boot-entries run
|
||||
@@ -0,0 +1,133 @@
|
||||
#!/bin/sh
|
||||
# The fallback kernel's place on this machine's boot menu (module kernel, novox/hq ADR 0508 §9 and §10,
|
||||
# issue 1841). Written by the mesh and run as root by mesh-kernel-boot-entries.service at boot and when the
|
||||
# module is applied, and by the pacman hook 90-mesh-kernel-boot-entries.hook after a kernel is installed,
|
||||
# upgraded or removed. Replaced on every push; change the module, never this file.
|
||||
#
|
||||
# boot-entries run [<root> [<this module's state> [<room needed in /boot, KiB>]]]
|
||||
#
|
||||
# The three are arguments only so the module's tests can run this against a stand-in tree.
|
||||
#
|
||||
# It reads from the machine which loader booted it, never from a setting:
|
||||
# systemd-boot the EFI variable LoaderEntrySelected is present: the entry mesh-linux-lts.conf is written
|
||||
# from the command line the machine booted with.
|
||||
# GRUB /boot/grub/grub.cfg exists and that variable does not: the menu is generated again, only
|
||||
# when the set of kernels in /boot differs from the menu's.
|
||||
# neither it fails loudly. The default entry is never touched.
|
||||
# And a pacman hook of someone else's that copies kernel images to one fixed name in /boot is moved into this
|
||||
# module's state first (ADR 0508 §10): it would put the fallback kernel in the default kernel's place.
|
||||
set -eu
|
||||
|
||||
root=/
|
||||
state='${dir:state}'
|
||||
[ $# -ge 2 ] && root=$(printf '%s' "$2" | sed 's|/$||')
|
||||
[ $# -ge 3 ] && state=$3
|
||||
boot=$root/boot
|
||||
hooks=$root/etc/pacman.d/hooks
|
||||
efivar=$root/sys/firmware/efi/efivars/LoaderEntrySelected-4a67b082-0a4c-41cf-b6c7-440b29bb8c4f
|
||||
ours=90-mesh-kernel-boot-entries.hook
|
||||
# The fallback kernel and its initramfs are about 72 MB in /boot (issue 1836): less room fails the unit
|
||||
# before a package writes half of them.
|
||||
need_kib=98304
|
||||
[ $# -ge 4 ] && need_kib=$4
|
||||
|
||||
[ $# -ge 1 ] && [ "$1" = run ] || {
|
||||
echo "usage: boot-entries run [<root> [<state directory> [<KiB needed>]]]" >&2
|
||||
exit 2
|
||||
}
|
||||
|
||||
# A hook that is not ours, triggered by kernel images, whose action copies to one fixed name in /boot.
|
||||
copies_kernels() {
|
||||
grep -Eqi '^[[:space:]]*Target[[:space:]]*=.*(vmlinuz|usr/lib/modules)' "$1" &&
|
||||
grep -Eq '^[[:space:]]*Exec[[:space:]]*=.*(cp|install|ln|rsync)[[:space:]].*/boot/[A-Za-z0-9._+-]+([^A-Za-z0-9._+/%$*-]|$)' "$1"
|
||||
}
|
||||
|
||||
if [ -d "$hooks" ]; then
|
||||
for hook in "$hooks"/*.hook; do
|
||||
[ -f "$hook" ] || continue
|
||||
[ "$(basename "$hook")" = "$ours" ] && continue
|
||||
copies_kernels "$hook" || continue
|
||||
mkdir -p "$state/taken-over"
|
||||
mv -f "$hook" "$state/taken-over/$(basename "$hook")"
|
||||
echo "taken over: the pacman hook $(basename "$hook") copied kernel images to one fixed name in /boot; moved to $state/taken-over/ so each kernel keeps its own package's name"
|
||||
done
|
||||
fi
|
||||
|
||||
if [ -e "$efivar" ]; then
|
||||
loader=systemd-boot
|
||||
elif [ -f "$boot/grub/grub.cfg" ]; then
|
||||
loader=grub
|
||||
else
|
||||
echo "no boot loader found: the machine was not booted by systemd-boot (no LoaderEntrySelected variable) and there is no $boot/grub/grub.cfg; the fallback kernel has no menu to be placed on" >&2
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# Room first, only while the fallback kernel is not in /boot yet.
|
||||
if [ ! -e "$boot/vmlinuz-linux-lts" ]; then
|
||||
free=$(df -Pk "$boot" | awk 'NR==2 {print $4}')
|
||||
[ -n "$free" ] || free=0
|
||||
if [ "$free" -lt "$need_kib" ]; then
|
||||
echo "$boot has $free KiB free and the fallback kernel needs $need_kib KiB: free room there before the kernel is installed" >&2
|
||||
exit 1
|
||||
fi
|
||||
fi
|
||||
|
||||
case "$loader" in
|
||||
systemd-boot)
|
||||
entries=$boot/loader/entries
|
||||
cmdline=$(sed -e 's/BOOT_IMAGE=[^ ]*//g' -e 's/initrd=[^ ]*//g' "$root/proc/cmdline" | tr -s ' \n' ' ' | sed -e 's/^ //' -e 's/ $//')
|
||||
[ -n "$cmdline" ] || {
|
||||
echo "$root/proc/cmdline holds no command line to boot the fallback kernel with" >&2
|
||||
exit 1
|
||||
}
|
||||
want=$(
|
||||
echo "title Arch Linux (linux-lts, fallback)"
|
||||
echo "linux /vmlinuz-linux-lts"
|
||||
for ucode in intel-ucode.img amd-ucode.img; do
|
||||
[ -e "$boot/$ucode" ] && echo "initrd /$ucode"
|
||||
done
|
||||
echo "initrd /initramfs-linux-lts.img"
|
||||
echo "options $cmdline"
|
||||
)
|
||||
mkdir -p "$entries"
|
||||
if [ "$(cat "$entries/mesh-linux-lts.conf" 2>/dev/null || true)" = "$want" ]; then
|
||||
echo "systemd-boot: mesh-linux-lts.conf, as it was"
|
||||
else
|
||||
printf '%s\n' "$want" >"$entries/mesh-linux-lts.conf"
|
||||
echo "systemd-boot: wrote $entries/mesh-linux-lts.conf from the command line the machine booted with; the default entry is as it was"
|
||||
fi
|
||||
;;
|
||||
grub)
|
||||
cfg=$boot/grub/grub.cfg
|
||||
menu=$(sed -n 's|^[[:space:]]*linux[[:space:]]*\(/boot\)\{0,1\}/\(vmlinuz-[^[:space:]]*\).*|\2|p' "$cfg" | sort -u)
|
||||
have=$(cd "$boot" && ls -d vmlinuz-* 2>/dev/null | sort -u || true)
|
||||
if [ "$menu" = "$have" ]; then
|
||||
echo "grub: the menu lists the kernels in /boot, left as it is"
|
||||
else
|
||||
# GRUB lists kernels in reverse version order, so the fallback kernel would come first and become the
|
||||
# default entry: the kernel the machine booted is put first (GRUB_TOP_LEVEL; /etc/default/grub still
|
||||
# overrides it when it sets one), the fallback kernel never.
|
||||
top=$(tr ' ' '\n' <"$root/proc/cmdline" | sed -n 's|^BOOT_IMAGE=\(.*\)$|\1|p' | head -n 1)
|
||||
top=$(printf '%s' "$top" | sed 's|.*/||')
|
||||
case "$top" in
|
||||
vmlinuz-linux-lts | "") top= ;;
|
||||
esac
|
||||
[ -n "$top" ] && [ -e "$boot/$top" ] || top=
|
||||
if [ -z "$top" ]; then
|
||||
if [ -e "$boot/vmlinuz-linux" ]; then
|
||||
top=vmlinuz-linux
|
||||
else
|
||||
top=$(printf '%s\n' $have | grep -v '^vmlinuz-linux-lts$' | head -n 1 || true)
|
||||
fi
|
||||
fi
|
||||
if [ -n "$top" ]; then
|
||||
GRUB_TOP_LEVEL=/boot/$top grub-mkconfig -o "$cfg"
|
||||
else
|
||||
grub-mkconfig -o "$cfg"
|
||||
fi
|
||||
first=$top
|
||||
[ -n "$first" ] || first="the menu's own order"
|
||||
echo "grub: the kernels in /boot changed ($(echo $have | tr '\n' ' ')); the menu is generated again with $first first; the default entry is as it was"
|
||||
fi
|
||||
;;
|
||||
esac
|
||||
@@ -0,0 +1,22 @@
|
||||
[Unit]
|
||||
Description=The fallback kernel's place on this machine's boot menu
|
||||
# Module kernel, novox/hq ADR 0508 §9, issue 1841. Starting this unit reads which loader booted the machine and
|
||||
# places the fallback kernel's boot entry (systemd-boot) or generates the menu again (GRUB); it fails, and says
|
||||
# why, when there is no loader or the boot partition has no room. The kernel packages are declared after this
|
||||
# unit, so the entry is in place before they are installed.
|
||||
After=local-fs.target
|
||||
Wants=local-fs.target
|
||||
|
||||
[Service]
|
||||
Type=oneshot
|
||||
RemainAfterExit=yes
|
||||
ExecStart=${dir:state}/boot-entries run
|
||||
# It writes the boot partition, the pacman hook directory and its own state, and nothing else.
|
||||
ProtectSystem=strict
|
||||
ReadWritePaths=/boot ${dir:state} /etc/pacman.d/hooks
|
||||
ProtectHome=yes
|
||||
PrivateNetwork=yes
|
||||
NoNewPrivileges=yes
|
||||
|
||||
[Install]
|
||||
WantedBy=multi-user.target
|
||||
@@ -13,7 +13,8 @@
|
||||
}
|
||||
},
|
||||
"tools": [
|
||||
"kernel_huge_pages"
|
||||
"kernel_huge_pages",
|
||||
"kernel_boot"
|
||||
],
|
||||
"resources": [
|
||||
{
|
||||
@@ -53,6 +54,52 @@
|
||||
"health": {
|
||||
"kind": "unit"
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "boot-entries",
|
||||
"type": "file",
|
||||
"path": "${dir:state}/boot-entries",
|
||||
"mode": "0755",
|
||||
"trusted": true,
|
||||
"template": "files/boot-entries.sh"
|
||||
},
|
||||
{
|
||||
"id": "boot-unit",
|
||||
"type": "file",
|
||||
"path": "/etc/systemd/system/mesh-kernel-boot-entries.service",
|
||||
"mode": "0644",
|
||||
"template": "files/mesh-kernel-boot-entries.service"
|
||||
},
|
||||
{
|
||||
"id": "boot-hook",
|
||||
"type": "file",
|
||||
"path": "/etc/pacman.d/hooks/90-mesh-kernel-boot-entries.hook",
|
||||
"mode": "0644",
|
||||
"template": "files/90-mesh-kernel-boot-entries.hook"
|
||||
},
|
||||
{
|
||||
"id": "boot-apply",
|
||||
"type": "service",
|
||||
"unit": "mesh-kernel-boot-entries.service",
|
||||
"state": "running",
|
||||
"boot": "enabled",
|
||||
"restart-on": [
|
||||
"boot-unit",
|
||||
"boot-entries"
|
||||
],
|
||||
"health": {
|
||||
"kind": "unit"
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "linux-lts",
|
||||
"type": "package",
|
||||
"package": "linux-lts"
|
||||
},
|
||||
{
|
||||
"id": "linux-lts-headers",
|
||||
"type": "package",
|
||||
"package": "linux-lts-headers"
|
||||
}
|
||||
],
|
||||
"build": {
|
||||
|
||||
@@ -1,6 +1,7 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"reflect"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
@@ -9,15 +10,14 @@ import (
|
||||
// One build per change (novox/hq ADR 0350): a pull request closed unmerged has its check builds let go of, and a
|
||||
// walk's record says per module whether its build reused a check's build or built, and why.
|
||||
|
||||
// A pull request closed unmerged is stopped. Since issue 1318 (ADR 0503 §3) the controller is no longer asked to let go
|
||||
// of what its checks built (§5's forget-checks goes with the merge gate's checks); a controller that is down changes
|
||||
// nothing.
|
||||
func TestAPullRequestClosedUnmergedIsStoppedAndAsksTheControllerNothing(t *testing.T) {
|
||||
// A pull request closed unmerged asks the controller to let go of what its checks built for its merge (§5); one
|
||||
// the controller cannot answer is said, never fatal.
|
||||
func TestAPullRequestClosedUnmergedLetsGoOfItsCheckBuilds(t *testing.T) {
|
||||
w := newWorldOver(t, newMemStore(), newFakeController(), newFakeForge(), atCC)
|
||||
cc := pr(cat, 101, headCC, "feat/claude-code", "modules/claude-code/module.json")
|
||||
w.h.PullUpdated(cc)
|
||||
w.h.PullClosed(cc)
|
||||
if len(w.ctl.forgot) != 0 {
|
||||
if want := []string{"novox/mesh-catalog#101"}; !reflect.DeepEqual(w.ctl.forgot, want) {
|
||||
t.Fatalf("closed unmerged, the controller was asked to let go of %v", w.ctl.forgot)
|
||||
}
|
||||
if s := w.state(IDOf(cat, headCC)); s != Stopped {
|
||||
|
||||
@@ -18,7 +18,6 @@ import (
|
||||
|
||||
// Context names of the statuses this owner can say more of than the forge keeps.
|
||||
const (
|
||||
// ctxGate is the controller's merge gate, kept where a base still requires it; mesh/check is ctxCheck (onestatus.go).
|
||||
ctxGate = "mesh/merge-gate"
|
||||
ctxRepo = "mesh/repo-check"
|
||||
ctxDelivery = "mesh/delivery"
|
||||
@@ -231,10 +230,6 @@ func (h *Holder) addWhatIsKnown(s *CheckStatus, v *Verdict, d *checksDelivery) {
|
||||
} else if v.Repo != "" {
|
||||
s.Verdict = strings.ToUpper(v.Repo) + " — " + v.RepoSaid
|
||||
}
|
||||
case ctxCheck:
|
||||
if v != nil && v.Status == ctxCheck {
|
||||
s.Verdict = strings.ToUpper(firstOf(v.Gate, "error")) + " — " + v.Summary
|
||||
}
|
||||
case ctxDelivery:
|
||||
if d != nil {
|
||||
s.Verdict = string(d.State)
|
||||
|
||||
@@ -103,13 +103,6 @@ func (h *Holder) askProposedAgain(id, why string) {
|
||||
h.mu.Unlock()
|
||||
return
|
||||
}
|
||||
if d.judgedByCheck() {
|
||||
// The checker's (issue 1302): its check is not asked here, and whether a held config value is weighed is the
|
||||
// checker's composition's.
|
||||
h.Logf("[mesh-delivery] %s: a held config value changed; %v", id, checkerRefusal(id))
|
||||
h.mu.Unlock()
|
||||
return
|
||||
}
|
||||
owner, repo, commit, page, number := d.Owner(), d.Repo(), d.Commit, d.HTMLURL, d.Number
|
||||
h.mu.Unlock()
|
||||
if err := h.checkingAgain(owner, repo, number, commit, page, why); err != nil {
|
||||
|
||||
@@ -19,9 +19,6 @@ type ForgeStatusEvent struct {
|
||||
SHA string `json:"sha"`
|
||||
Context string `json:"context"`
|
||||
State string `json:"state"`
|
||||
// Description is the status's words (the gitea module's status event carries them since issue 1302): mesh/check's
|
||||
// line names the head and the trunk it judged.
|
||||
Description string `json:"description,omitempty"`
|
||||
}
|
||||
|
||||
// wantForge says the forge's open pull requests are to be read at the next look: a check may have answered.
|
||||
@@ -47,8 +44,7 @@ func (h *Holder) AnswersChanged() {
|
||||
}
|
||||
}
|
||||
|
||||
// CommitStatus takes a commit status set on the forge: one of the merge checks (mesh/check; hq's mesh/repo-check and, on a
|
||||
// base that still requires it, mesh/merge-gate) says a
|
||||
// CommitStatus takes a commit status set on the forge: one of the merge checks (mesh/merge-gate, mesh/repo-check) says a
|
||||
// check started, answered, or was put back, so the forge's open pull requests are read to see which heads still wait on
|
||||
// one (SeeOpenPulls). A status of any other context shows nothing the merge needs.
|
||||
//
|
||||
@@ -64,10 +60,6 @@ func (h *Holder) CommitStatus(e ForgeStatusEvent) {
|
||||
h.wantForge()
|
||||
h.mu.Lock()
|
||||
defer h.mu.Unlock()
|
||||
if e.Context == ctxCheck && (e.Description != "" || e.State == "pending") {
|
||||
// mesh/check is the verdict on its base (issue 1302): taken at once, as the controller's `checked` was.
|
||||
h.takeCheckStatus(e.Owner+"/"+e.Repo, e.SHA, e.State, e.Description)
|
||||
}
|
||||
if e.State != "success" {
|
||||
h.withdrawForStatus(e)
|
||||
return
|
||||
|
||||
@@ -78,9 +78,6 @@ type Holder struct {
|
||||
refusals []string
|
||||
// unchecked are the open pull requests' heads seen without a merge check (unchecked.go).
|
||||
unchecked map[string]unchecked
|
||||
// checker are the checker's verdicts by the hash of the heads they judged, as its verdicts bucket holds them
|
||||
// (onestatus.go, issue 1302).
|
||||
checker map[string]checkerVerdict
|
||||
// walks are the walks' and batches' last records (answers.go); orphans the waiting walks no live delivery
|
||||
// holds, by when that was first seen; turning the walks whose turn came this look, for some deliveries.
|
||||
walks map[string]seenWalk
|
||||
@@ -173,7 +170,6 @@ func (h *Holder) init() {
|
||||
h.lostRead = map[string]time.Time{}
|
||||
h.walkAsked = map[string]time.Time{}
|
||||
h.unchecked = map[string]unchecked{}
|
||||
h.checker = map[string]checkerVerdict{}
|
||||
h.walks = map[string]seenWalk{}
|
||||
h.orphans = map[string]time.Time{}
|
||||
h.asking = map[string]bool{}
|
||||
@@ -864,8 +860,9 @@ func verdictOfState(s string) string {
|
||||
|
||||
// PullClosed takes a pull request closed unmerged: its open deliveries are stopped.
|
||||
//
|
||||
// What its checks built is no longer asked to be let go of here (issue 1318): the controller's forget-checks goes with
|
||||
// the merge gate's checks (ADR 0503 §3), and the build seat's holders let go of it within the week.
|
||||
// **And what its checks built for its merge goes** (novox/hq ADR 0350 §5): nothing will take it now. Asked of the
|
||||
// controller once the deliveries are settled, outside the lock; a refusal is said, and the build seat's holders let
|
||||
// go of it within the week anyway.
|
||||
func (h *Holder) PullClosed(p PullEvent) {
|
||||
h.mu.Lock()
|
||||
h.forgetUnchecked(p.Owner, p.Repo, p.Number)
|
||||
@@ -902,6 +899,15 @@ func (h *Holder) PullClosed(p PullEvent) {
|
||||
h.pullUpdated(*over)
|
||||
}
|
||||
h.mu.Unlock()
|
||||
if h.Controller == nil || p.Owner == "" || p.Repo == "" || p.Number <= 0 {
|
||||
return
|
||||
}
|
||||
if err := h.Controller.ForgetChecks(p.Owner, p.Repo, p.Number); err != nil {
|
||||
h.mu.Lock()
|
||||
h.refuse(fmt.Errorf("%s#%d closed, and its check builds stay until the week's sweep: %w", p.repository(),
|
||||
p.Number, err))
|
||||
h.mu.Unlock()
|
||||
}
|
||||
}
|
||||
|
||||
// CheckedEvent is the controller's `checked`.
|
||||
@@ -1953,7 +1959,7 @@ func standingVerdict(p OpenPull, now time.Time) *Verdict {
|
||||
if p.Checks == nil {
|
||||
return nil
|
||||
}
|
||||
contexts := []string{"mesh/merge-gate", "mesh/repo-check"} // the controller's; mesh/check is taken in askChecks
|
||||
contexts := slices.Clone(mergeChecks)
|
||||
for _, c := range p.Required {
|
||||
if !slices.Contains(contexts, c) {
|
||||
contexts = append(contexts, c)
|
||||
@@ -1989,8 +1995,7 @@ func (h *Holder) askChecks(now time.Time) func() {
|
||||
}
|
||||
w := want{id: d.ID, head: d.Commit, number: d.Number, member: MemberOf(d)}
|
||||
w.owner, w.repo, _ = strings.Cut(d.Repository, "/")
|
||||
// A base that requires mesh/check is the checker's: nothing is asked of the controller (issue 1302).
|
||||
if !askedCheck(d) && !d.judgedByCheck() {
|
||||
if !askedCheck(d) {
|
||||
if tried, was := h.askTried[d.ID]; was && now.Sub(tried) < time.Minute {
|
||||
h.lookAt(tried.Add(time.Minute), "a proposed delivery's check that could not be asked")
|
||||
} else {
|
||||
@@ -2009,30 +2014,14 @@ func (h *Holder) askChecks(now time.Time) func() {
|
||||
open, err := h.Forge.OpenPulls()
|
||||
for _, w := range wants {
|
||||
var standing *Verdict
|
||||
oneStatus := false
|
||||
if err == nil {
|
||||
for _, p := range open {
|
||||
if p.Number == w.number && strings.EqualFold(p.Owner, w.owner) && strings.EqualFold(p.Repo, w.repo) &&
|
||||
sameCommit(p.HeadSHA, w.head) {
|
||||
if slices.Contains(p.Required, ctxCheck) {
|
||||
// Its base requires mesh/check (issue 1302): its verdict is that status, when it stands, and
|
||||
// nothing is asked of the controller.
|
||||
oneStatus = true
|
||||
h.mu.Lock()
|
||||
h.noteRequired(h.deliveries[w.id], p.Required)
|
||||
if c := p.Checks[ctxCheck]; c.State != "" {
|
||||
h.takeCheckStatus(w.owner+"/"+w.repo, w.head, c.State, c.Description)
|
||||
}
|
||||
h.mu.Unlock()
|
||||
continue
|
||||
}
|
||||
standing = standingVerdict(p, h.Now())
|
||||
}
|
||||
}
|
||||
}
|
||||
if oneStatus {
|
||||
continue
|
||||
}
|
||||
if standing != nil {
|
||||
h.mu.Lock()
|
||||
if d := h.deliveries[w.id]; d != nil && d.State == Proposed && d.Check == nil {
|
||||
@@ -2180,20 +2169,14 @@ func (h *Holder) planGroup(g *Group, members []*Delivery, now time.Time) []func(
|
||||
allReady = false
|
||||
}
|
||||
}
|
||||
// A group judged by mesh/check (issue 1302, ADR 0447 §2): its verdict is the checker's one verdict at its members'
|
||||
// heads, read, not asked; nothing below asks a composed check or owes a member's mesh/merge-gate.
|
||||
oneStatus := oneStatusGroup(members)
|
||||
if oneStatus {
|
||||
h.composeFromChecker(g, members, now)
|
||||
}
|
||||
again := !oneStatus && g.Check != nil && (!sameMembers(g.Check.Commits, commits) || g.Check.OutOfDate != "" && rejudgedSince(members, g.Check.OutOfDateAt))
|
||||
again := g.Check != nil && (!sameMembers(g.Check.Commits, commits) || g.Check.OutOfDate != "" && rejudgedSince(members, g.Check.OutOfDateAt))
|
||||
// A closed group's member whose merge the tool refused, after the composed check answered, is judged again at once
|
||||
// (novox/hq ADR 0418 §4, issue 1082): the refusal is a sibling's merge having moved what the verdict stands on, and
|
||||
// nothing else asks the composed check again. Not while a check is on its way (its verdict is empty then).
|
||||
// **A refusal that names the member's OWN pending status** (mesh/repo-check, with the gate not pending: the composed
|
||||
// check set it) is not that: the composed check sets no repository check, so asking it again changes nothing. The
|
||||
// member's own check is asked, once for the head (issue 1049); its verdict sets the status.
|
||||
if !oneStatus && !again && g.Closed && g.Check != nil && g.Check.Verdict != "" {
|
||||
if !again && g.Closed && g.Check != nil && g.Check.Verdict != "" {
|
||||
for _, d := range members {
|
||||
if !refusedAfter(d, g.Check.At) {
|
||||
continue
|
||||
@@ -2210,7 +2193,7 @@ func (h *Holder) planGroup(g *Group, members []*Delivery, now time.Time) []func(
|
||||
break
|
||||
}
|
||||
}
|
||||
if !oneStatus && allReady && (g.Check == nil && !g.Closed || again) {
|
||||
if allReady && (g.Check == nil && !g.Closed || again) {
|
||||
ms := membersOfAsk(members)
|
||||
id := g.ID
|
||||
asks = append(asks, func() {
|
||||
@@ -2279,7 +2262,7 @@ func (h *Holder) planGroup(g *Group, members []*Delivery, now time.Time) []func(
|
||||
}
|
||||
// Each member's mesh/merge-gate is the composed check's (ADR 0403 §1, §6): pending, saying why, while no composed
|
||||
// verdict stands at these heads. The verdict itself sets it, success or failure, in the forge module.
|
||||
if why := composedPending(g, members); why != "" && !oneStatus {
|
||||
if why := composedPending(g, members); why != "" {
|
||||
if g.SaidGate == nil {
|
||||
g.SaidGate = map[string]string{}
|
||||
}
|
||||
@@ -2837,8 +2820,8 @@ func ownStatusPending(answer string) bool {
|
||||
// its own. Under the lock; the ask is made outside it. Nil when it was asked already for this head.
|
||||
func (h *Holder) askOwnAfterRefusal(d *Delivery) func() {
|
||||
r := d.Ready
|
||||
if !r.OwnAskedAt.IsZero() || d.judgedByCheck() {
|
||||
return nil // asked already; or the checker's, who checks it again on its own (issue 1302)
|
||||
if !r.OwnAskedAt.IsZero() {
|
||||
return nil
|
||||
}
|
||||
r.OwnAskedAt = h.Now()
|
||||
h.keep(d)
|
||||
|
||||
@@ -30,8 +30,8 @@ func (h *Holder) lostCheckBound(repository string) time.Duration {
|
||||
// awaitedSince is when a proposed delivery's check was asked, when it waits for one this owner asked: no verdict yet,
|
||||
// or a verdict out of date with its check against the trunk's head asked. False for a delivery waiting for none.
|
||||
func awaitedSince(d *Delivery) (time.Time, bool) {
|
||||
if d.State != Proposed || d.MergedAs != "" || d.Number == 0 || d.judgedByCheck() {
|
||||
return time.Time{}, false // a head on mesh/check asks the controller nothing (issue 1302)
|
||||
if d.State != Proposed || d.MergedAs != "" || d.Number == 0 {
|
||||
return time.Time{}, false
|
||||
}
|
||||
switch {
|
||||
case d.Check == nil && askedCheck(d):
|
||||
|
||||
@@ -115,25 +115,6 @@ func run(h *Holder, l *listening, occupancy *busOccupancy) {
|
||||
time.Sleep(wait)
|
||||
}
|
||||
}()
|
||||
// The checker's verdicts (issue 1302, ADR 0447 §2): the Approve on mesh/check is asked only on its verdict at exactly the
|
||||
// heads an ask covers, and a group's verdict is that one; each change asks for a look.
|
||||
go func() {
|
||||
for wait := 2 * time.Second; ; wait = min(wait*2, time.Minute) {
|
||||
err := stdio.State(checkerVerdicts).Watch("", func(c stdio.StateChange) error {
|
||||
if c.Op == "delete" {
|
||||
h.CheckerVerdictChanged(c.Key, nil)
|
||||
} else {
|
||||
h.CheckerVerdictChanged(c.Key, c.Value)
|
||||
}
|
||||
return nil
|
||||
})
|
||||
if err == nil {
|
||||
return
|
||||
}
|
||||
logf("[mesh-delivery] not hearing the checker's verdicts yet (%v); asking again in %s", err, wait)
|
||||
time.Sleep(wait)
|
||||
}
|
||||
}()
|
||||
logf("[mesh-delivery] listening: %s", strings.Join(followed, ", "))
|
||||
}
|
||||
|
||||
@@ -332,8 +313,7 @@ func tools(h *Holder, l *listening) []stdio.Tool {
|
||||
{Name: seat + "groups",
|
||||
Description: "Every delivery group: its members, its composed check and its state, derived from its members. " +
|
||||
"A group has no order: it is merged in one pass and delivered by one plan, whose tiers come from the module " +
|
||||
"graph (novox/hq ADR 0418). On mesh/check its check is the checker's one verdict at every member's head (ADR 0447); " +
|
||||
"composed_gate says its members' gate is the controller's composed check's (ADR 0403), for a base still on it.",
|
||||
"graph (novox/hq ADR 0418). composed_gate says its members' gate is the composed check's (ADR 0403).",
|
||||
Input: map[string]any{"type": "object", "properties": map[string]any{
|
||||
"all": map[string]any{"type": "string", "enum": []string{"true", "false"}}}},
|
||||
Run: func(a map[string]any) (any, error) { return h.Groups(boolArg(a, "all")), nil }},
|
||||
@@ -350,7 +330,7 @@ func tools(h *Holder, l *listening) []stdio.Tool {
|
||||
}},
|
||||
{Name: seat + "checks",
|
||||
Description: "What the mesh's checks said of a pull request's head or of one commit: each of the commit's " +
|
||||
"statuses (mesh/check, mesh/delivery, …; hq's mesh/repo-check) with its state, description and when it " +
|
||||
"statuses (mesh/merge-gate, mesh/repo-check, mesh/delivery, …) with its state, description and when it " +
|
||||
"was set; the merge check's full verdict as the controller said it — each layer's summary, the machine " +
|
||||
"that ran it, when, its build id and its report; and whether the branch's protection would let it " +
|
||||
"merge, every required status being success.",
|
||||
|
||||
@@ -83,17 +83,16 @@ func TestItClaimsTheDeliverySeatAndFollowsWhatItHandles(t *testing.T) {
|
||||
if !reflect.DeepEqual(buckets, []string{"deliveries", "groups", "pulls", "asks", "merges", "owed", "source"}) {
|
||||
t.Fatalf("state %v", buckets)
|
||||
}
|
||||
// What it reads: the carry from today's module's three buckets, the teller's answers, the checker's verdicts (issue
|
||||
// 1302), the seat's holder and the machines' haves (ADR 0450 §6 and its insight of 2026-10-20).
|
||||
// What it reads: the carry from today's module's three buckets, the teller's answers, the seat's holder and the
|
||||
// machines' haves (ADR 0450 §6 and its insight of 2026-10-20).
|
||||
if !reflect.DeepEqual(m.Reads, []string{"mesh-delivery.deliveries", "mesh-delivery.groups", "mesh-delivery.pulls",
|
||||
"mesh-delivery-teller.answers", "mesh-delivery-checker.verdicts", "mesh-controller.assignments", "node-engine.have"}) {
|
||||
"mesh-delivery-teller.answers", "mesh-controller.assignments", "node-engine.have"}) {
|
||||
t.Fatalf("reads %v", m.Reads)
|
||||
}
|
||||
// It calls exactly the controller's seven verbs (ADR 0450 §7; queue since issue 1274, forget-checks no more since issue 1318) and reads the forge with
|
||||
// three tools; the forge's
|
||||
// It calls exactly the controller's seven verbs (ADR 0450 §7) and reads the forge with three tools; the forge's
|
||||
// writes are the bridge's (ADR 0437 §8). It uses no operator-channel: the teller asks (ADR 0437 §6).
|
||||
want := []string{}
|
||||
for _, v := range []string{"deliver", "delivery-stop", "delivery-backups-again", "delivery-walks", "delivery-check", "queue", "plans"} {
|
||||
for _, v := range []string{"deliver", "delivery-stop", "delivery-backups-again", "delivery-walks", "delivery-check", "forget-checks", "queue", "plans"} {
|
||||
want = append(want, "seat:"+ControllerSeat+"."+v)
|
||||
}
|
||||
want = append(want, "gitea.gitea_commit_statuses", "gitea.gitea_contains", "gitea.gitea_open_pulls")
|
||||
|
||||
@@ -54,12 +54,6 @@ type Delivery struct {
|
||||
Plan *DeliveryPlan `json:"plan,omitempty"`
|
||||
// Check is its own check's verdict.
|
||||
Check *Verdict `json:"check,omitempty"`
|
||||
// Required are the statuses its base requires, as the forge last said them (gitea_open_pulls): mesh/check makes it
|
||||
// the checker's (issue 1302). Empty until the forge said them.
|
||||
Required []string `json:"required,omitempty"`
|
||||
// CheckState is its head's mesh/check as the forge last said it (pending, success, failure, error): a pending one
|
||||
// after a pass is the checker judging it again, and no Approve is asked meanwhile (issue 1302).
|
||||
CheckState string `json:"check_state,omitempty"`
|
||||
// CheckAsk is this owner's own ask of its check, when the forge's announcement did not bring one: a head
|
||||
// adopted while it already waited, or one whose announcement the controller missed.
|
||||
CheckAsk *CheckAsk `json:"check_ask,omitempty"`
|
||||
@@ -185,9 +179,6 @@ type Verdict struct {
|
||||
Trunk string `json:"trunk,omitempty"`
|
||||
Beside map[string]string `json:"beside,omitempty"`
|
||||
Took string `json:"took,omitempty"`
|
||||
// Status is the forge status it was read from when it is not the controller's verdict: mesh/check, the checker's
|
||||
// (novox/hq ADR 0447, issue 1302). Empty for the controller's `checked`.
|
||||
Status string `json:"status,omitempty"`
|
||||
}
|
||||
|
||||
// Judged is what a verdict judged, in words: "head 1a2b3c4d5e6f merged into trunk 7a8b9c0d1e2f" (ADR 0323 §3);
|
||||
@@ -1043,8 +1034,7 @@ type Group struct {
|
||||
Updated time.Time `json:"updated"`
|
||||
// Check is its composed check: asked of the controller for these member commits, and its verdict.
|
||||
Check *GroupCheck `json:"check,omitempty"`
|
||||
// SaidGate is, by member id, why the member's mesh/merge-gate was last said pending (a group the controller judges;
|
||||
// a group on mesh/check is owed none, issue 1302): written again only when it
|
||||
// SaidGate is, by member id, why the member's mesh/merge-gate was last said pending: written again only when it
|
||||
// changes (novox/hq ADR 0403 §1, §6, issue 978). Cleared once a composed verdict stands at the heads, which the
|
||||
// forge module then sets as every member's gate.
|
||||
SaidGate map[string]string `json:"said_gate,omitempty"`
|
||||
|
||||
@@ -1,289 +0,0 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"regexp"
|
||||
"slices"
|
||||
"sort"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
sdk "git.novox.be/novox/mesh-sdk/go/delivery"
|
||||
)
|
||||
|
||||
// **One required status** (novox/hq ADR 0447, issue 1302; ADR 0503 §3, §5).
|
||||
//
|
||||
// Main requires `mesh/check` everywhere but hq: the checker's verdict on a pull request's head merged onto its trunk's
|
||||
// head, written by its forge bridge. On such a base this owner reads that one status, and nothing of the controller:
|
||||
//
|
||||
// - the verdict of a head is its mesh/check, heard as the forge's status event (its words name the head and the trunk
|
||||
// it judged: "pass <head12> on <trunk12> | …") or read from the forge's open pull requests; it is kept as the
|
||||
// delivery's Check, marked as mesh/check's, and moves the delivery as the controller's verdict did;
|
||||
// - the Approve is asked only when the checker's verdict for exactly the heads the ask covers passed (its verdicts
|
||||
// bucket, keyed by the heads' hash, ADR 0447 §2): a member's pass from a check that judged it alone, before a sibling
|
||||
// joined, is not the group's;
|
||||
// - a group's verdict is that one verdict (no composed check is asked, and no member is owed a mesh/merge-gate); a
|
||||
// member's mesh/check that is no longer a success withdraws the group's ask (ADR 0441 §1, withdrawForStatus);
|
||||
// - no check is asked of the controller: the checker checks every head on every move, and the trunk's move is its to
|
||||
// say on mesh/check (ADR 0447 §4). The controller is asked only for a base that requires mesh/repo-check without
|
||||
// mesh/check (hq), and only what it was asked before (the lead's decision for issue 1302 in R4e).
|
||||
//
|
||||
// A base whose required statuses are not known yet is read as before: the forge's open pull requests say them at the next
|
||||
// look, and a mesh/check heard on the head says it too.
|
||||
|
||||
// ctxCheck is the one status main requires (ADR 0447 §1).
|
||||
const ctxCheck = "mesh/check"
|
||||
|
||||
// checkerVerdicts is the checker's verdicts bucket as this module reads it (its manifest's reads).
|
||||
const checkerVerdicts = "mesh-delivery-checker.verdicts"
|
||||
|
||||
// The states of a checker's verdict (mesh-sdk go/delivery VerdictRunning, VerdictPass, VerdictFail; this module's mesh-sdk
|
||||
// pin is older than them).
|
||||
const (
|
||||
checkerRunning = "running"
|
||||
checkerPass = "pass"
|
||||
)
|
||||
|
||||
// checkerVerdict is a verdict of the checker as its verdicts bucket keeps it (mesh-sdk go/delivery Verdict, read here by
|
||||
// the fields the gate needs): the group, the hash of the heads it judged, each repository's trunk head and its state.
|
||||
type checkerVerdict struct {
|
||||
Group string `json:"group"`
|
||||
Heads string `json:"heads"`
|
||||
Trunk map[string]string `json:"trunk,omitempty"`
|
||||
State string `json:"state"`
|
||||
Failed struct {
|
||||
Test string `json:"test,omitempty"`
|
||||
Machine string `json:"machine,omitempty"`
|
||||
Words string `json:"words,omitempty"`
|
||||
} `json:"failed,omitzero"`
|
||||
}
|
||||
|
||||
// CheckerVerdictChanged takes a key of the checker's verdicts bucket (`<group>.<heads hash>`) put or deleted (an empty
|
||||
// value): the view of the checker's verdicts the Approve and a group's verdict are read from. A value that is no verdict
|
||||
// is said and dropped.
|
||||
func (h *Holder) CheckerVerdictChanged(key string, value []byte) {
|
||||
h.mu.Lock()
|
||||
defer h.mu.Unlock()
|
||||
if h.checker == nil {
|
||||
h.checker = map[string]checkerVerdict{}
|
||||
}
|
||||
hash := key
|
||||
if i := strings.LastIndex(key, "."); i >= 0 {
|
||||
hash = key[i+1:]
|
||||
}
|
||||
if len(value) == 0 || string(value) == "null" {
|
||||
delete(h.checker, hash)
|
||||
h.Poke()
|
||||
return
|
||||
}
|
||||
var v checkerVerdict
|
||||
if err := json.Unmarshal(value, &v); err != nil || v.State == "" {
|
||||
h.Logf("[mesh-delivery] the checker's verdict %s cannot be read (%v); it is not taken", key, err)
|
||||
return
|
||||
}
|
||||
if v.Heads == "" {
|
||||
v.Heads = hash
|
||||
}
|
||||
h.checker[v.Heads] = v
|
||||
h.Poke()
|
||||
}
|
||||
|
||||
// checkWords are the head and the trunk mesh/check's line names: "<head12> on <trunk12>" (ADR 0447 §3).
|
||||
var checkWords = regexp.MustCompile(`\b([0-9a-f]{12}) on ([0-9a-f]{12})\b`)
|
||||
|
||||
// checkLine is the head and the trunk commit a mesh/check line names, twelve characters each; empty when it names none.
|
||||
func checkLine(desc string) (head, trunk string) {
|
||||
m := checkWords.FindStringSubmatch(strings.ToLower(desc))
|
||||
if m == nil {
|
||||
return "", ""
|
||||
}
|
||||
return m[1], m[2]
|
||||
}
|
||||
|
||||
// verdictOfCheck is a delivery's verdict from its head's mesh/check: its state as a verdict, its line as the summary, and
|
||||
// the head and trunk it names.
|
||||
func verdictOfCheck(state, desc string, now time.Time) *Verdict {
|
||||
head, trunk := checkLine(desc)
|
||||
return &Verdict{At: now, Gate: verdictOfState(state), Summary: firstOf(desc, "mesh/check "+state), Head: head, Trunk: trunk,
|
||||
Status: ctxCheck}
|
||||
}
|
||||
|
||||
// judgedByCheck is whether the delivery's base requires mesh/check, as the forge said it, or its verdict is mesh/check's.
|
||||
func (d *Delivery) judgedByCheck() bool {
|
||||
return slices.Contains(d.Required, ctxCheck) || (d.Check != nil && d.Check.Status == ctxCheck)
|
||||
}
|
||||
|
||||
// noteRequired keeps what the forge said the delivery's base requires. Under the lock.
|
||||
func (h *Holder) noteRequired(d *Delivery, required []string) {
|
||||
if d == nil || required == nil || slices.Equal(d.Required, required) {
|
||||
return
|
||||
}
|
||||
d.Required = slices.Clone(required)
|
||||
h.keep(d)
|
||||
}
|
||||
|
||||
// takeCheckStatus takes a head's mesh/check as its delivery's verdict, as the controller's `checked` was taken: a
|
||||
// proposed delivery moves on it, a rejected one judged again against another trunk or now passing is proposed again
|
||||
// first. A pending mesh/check is no verdict and changes nothing here (its ask is withdrawn by withdrawForStatus). The same
|
||||
// verdict heard twice changes nothing. Under the lock.
|
||||
func (h *Holder) takeCheckStatus(repository, sha, state, desc string) {
|
||||
d := h.deliveries[IDOf(strings.ToLower(repository), sha)]
|
||||
if d == nil || state == "" {
|
||||
return // its announcement makes it; the forge's open pull requests bring the verdict then
|
||||
}
|
||||
if d.CheckState != state {
|
||||
d.CheckState = state
|
||||
h.keep(d)
|
||||
}
|
||||
if state == "pending" {
|
||||
return
|
||||
}
|
||||
if !slices.Contains(d.Required, ctxCheck) {
|
||||
d.Required = append(d.Required, ctxCheck)
|
||||
}
|
||||
now := h.Now()
|
||||
v := verdictOfCheck(state, desc, now)
|
||||
if c := d.Check; c != nil && c.Status == ctxCheck && c.Gate == v.Gate && c.Trunk == v.Trunk && c.Summary == v.Summary {
|
||||
return
|
||||
}
|
||||
if d.State == Rejected && d.Check != nil && (v.Passes() || v.Trunk != "" && !sameCommit(d.Check.Trunk, v.Trunk)) {
|
||||
t, err := Apply(d, EvRecheck, true, Facts{Now: now, By: byOwner, Why: fmt.Sprintf("its mesh/check judged it again: %s",
|
||||
oneLine(clip(desc, 120)))})
|
||||
if err != nil {
|
||||
h.refuse(err)
|
||||
return
|
||||
}
|
||||
h.moved(d, []Transition{t})
|
||||
}
|
||||
if d.State != Proposed {
|
||||
h.keep(d)
|
||||
return // past its verdict: the merge reads the forge's status itself
|
||||
}
|
||||
d.Check, d.AskedAgain, d.CheckAsk = v, nil, nil
|
||||
h.mergeAgainNow(d)
|
||||
if why := d.outOfDate(); why != "" {
|
||||
h.Logf("[mesh-delivery] %s: its mesh/check is %s; the checker checks it again", d.ID, why)
|
||||
}
|
||||
h.Logf("[mesh-delivery] %s: its mesh/check is %s — %s", d.ID, state, oneLine(clip(desc, 120)))
|
||||
h.settleAndKeep(d)
|
||||
}
|
||||
|
||||
// headsOf are the heads the checker judges together for an ask: the members the ask covers.
|
||||
func headsOf(members []*Delivery) []sdk.Head {
|
||||
out := make([]sdk.Head, 0, len(members))
|
||||
for _, m := range members {
|
||||
out = append(out, sdk.Head{Repository: strings.ToLower(m.Repository), Number: m.Number, Head: strings.ToLower(m.Commit)})
|
||||
}
|
||||
return out
|
||||
}
|
||||
|
||||
// checkerAt is the checker's verdict for exactly these members' heads, and their hash.
|
||||
func (h *Holder) checkerAt(members []*Delivery) (checkerVerdict, string, bool) {
|
||||
hash, err := sdk.HeadsHash(headsOf(members))
|
||||
if err != nil {
|
||||
return checkerVerdict{}, "", false
|
||||
}
|
||||
v, ok := h.checker[hash]
|
||||
return v, hash, ok
|
||||
}
|
||||
|
||||
// checkerWaits is why the Approve of a delivery on mesh/check is not asked yet: the checker's verdict for the heads the ask
|
||||
// covers is not held, or did not pass. Empty when it passed, for a delivery not judged by mesh/check, and once an ask
|
||||
// covers the head (the merge then reads the forge's statuses itself). Under the lock.
|
||||
func (h *Holder) checkerWaits(d *Delivery) string {
|
||||
if !d.judgedByCheck() || h.covering(d) != nil {
|
||||
return ""
|
||||
}
|
||||
members := h.askMembers(d)
|
||||
for _, m := range members {
|
||||
if m.CheckState != "" && m.CheckState != "success" {
|
||||
return fmt.Sprintf("the mesh/check of %s is %s at its head %s", m.ID, m.CheckState, shortOf(m.Commit, 12))
|
||||
}
|
||||
}
|
||||
v, hash, ok := h.checkerAt(members)
|
||||
what := "its head"
|
||||
if len(members) > 1 {
|
||||
what = fmt.Sprintf("its group's %d heads", len(members))
|
||||
}
|
||||
switch {
|
||||
case !ok:
|
||||
return fmt.Sprintf("the checker's verdict at %s (heads %s) is not held yet", what, hash)
|
||||
case v.State == checkerRunning:
|
||||
return fmt.Sprintf("the checker's verdict at %s (heads %s) is running", what, hash)
|
||||
case v.State != checkerPass:
|
||||
return fmt.Sprintf("the checker's verdict at %s (heads %s) is %s", what, hash, v.State)
|
||||
}
|
||||
return ""
|
||||
}
|
||||
|
||||
// oneStatusGroup is whether a group is judged by mesh/check: one of its members is.
|
||||
func oneStatusGroup(members []*Delivery) bool {
|
||||
return slices.ContainsFunc(members, (*Delivery).judgedByCheck)
|
||||
}
|
||||
|
||||
// composeFromChecker makes a group's verdict on mesh/check from the checker's one verdict at its members' heads (ADR 0447
|
||||
// §2): no composed check is asked. It is kept in the group's check as a composed verdict was, so the group's state, its
|
||||
// Approve and its merges read it unchanged: running or not held is no verdict yet, a pass or a failure is the group's, and
|
||||
// a member whose trunk moved past what it judged puts it out of date. Under the lock.
|
||||
func (h *Holder) composeFromChecker(g *Group, members []*Delivery, now time.Time) {
|
||||
live := make([]*Delivery, 0, len(members))
|
||||
for _, d := range members {
|
||||
if d.MergedAs == "" && !d.State.Final() {
|
||||
live = append(live, d)
|
||||
}
|
||||
}
|
||||
if len(live) == 0 || (g.Closed && g.Check != nil && g.Check.Verdict != "") {
|
||||
return // every member on the trunk, or a closed group: the verdict its Approve was asked on stands
|
||||
}
|
||||
v, hash, ok := h.checkerAt(live)
|
||||
c := &GroupCheck{Asked: "mesh/check " + hash, At: now, Commits: commitsOf(members)}
|
||||
if g.Check != nil && g.Check.Asked == c.Asked && sameMembers(g.Check.Commits, c.Commits) {
|
||||
c.At = g.Check.At
|
||||
}
|
||||
if ok && v.State != checkerRunning {
|
||||
c.Verdict = v.State
|
||||
c.Summary = "the checker's verdict at the members' heads (" + hash + ")"
|
||||
if w := firstOf(v.Failed.Test, v.Failed.Machine, v.Failed.Words); w != "" {
|
||||
c.Summary += ": " + w
|
||||
}
|
||||
c.Trunks = map[string]string{}
|
||||
ids := make([]string, 0, len(live))
|
||||
for _, d := range live {
|
||||
ids = append(ids, d.ID)
|
||||
}
|
||||
sort.Strings(ids)
|
||||
for _, id := range ids {
|
||||
d := h.deliveries[id]
|
||||
judged := v.Trunk[strings.ToLower(d.Repository)]
|
||||
if judged == "" {
|
||||
continue
|
||||
}
|
||||
c.Trunks[d.ID] = judged
|
||||
if c.OutOfDate == "" && d.TrunkNow != "" && !sameCommit(judged, d.TrunkNow) {
|
||||
c.OutOfDate = fmt.Sprintf("the checker's verdict is out of date: %s was judged on trunk %s, the trunk is now %s",
|
||||
d.ID, shortOf(judged, 12), shortOf(d.TrunkNow, 12))
|
||||
c.OutOfDateAt = now
|
||||
}
|
||||
}
|
||||
}
|
||||
if g.Check != nil && g.Check.Asked == c.Asked && g.Check.Verdict == c.Verdict && g.Check.OutOfDate == c.OutOfDate &&
|
||||
sameMembers(g.Check.Commits, c.Commits) {
|
||||
return
|
||||
}
|
||||
g.Check, g.Updated = c, now
|
||||
h.keepGroup(g)
|
||||
if c.Verdict != "" {
|
||||
for _, d := range members {
|
||||
h.mergeAgainNow(d)
|
||||
}
|
||||
h.Logf("[mesh-delivery] group %s: %s — %s", g.ID, c.Verdict, c.Summary)
|
||||
}
|
||||
}
|
||||
|
||||
// checkerRefusal is why a delivery on mesh/check is asked nothing of the controller: the checker checks it.
|
||||
func checkerRefusal(id string) error {
|
||||
return fmt.Errorf("%s is judged by mesh/check, the checker's: the checker checks its head again on every move of the head, "+
|
||||
"its trunk or its group, and nothing is asked of it here (novox/hq ADR 0447). Read where its check stands with the "+
|
||||
"checker's checker_pull tool", id)
|
||||
}
|
||||
@@ -1,82 +0,0 @@
|
||||
//go:build integration
|
||||
|
||||
package main
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
sdk "git.novox.be/novox/mesh-sdk/go/delivery"
|
||||
)
|
||||
|
||||
// **Issue 1302's acceptance over a real bus** (ADR 0447, 1302-S10): a pull request into a main that requires mesh/check goes
|
||||
// ready on its mesh/check alone — the forge's status event and the checker's verdict in its verdicts bucket — has its Approve
|
||||
// asked, and on the Approve its merge is handed to the bridge; nothing is asked of the controller (no delivery-check, no
|
||||
// forget-checks) and no controller verdict is heard at all.
|
||||
func TestADeliveryOnMeshCheckGoesReadyAskedAndMergedOverARealBus(t *testing.T) {
|
||||
b := newTestBus(t)
|
||||
ctx := context.Background()
|
||||
rt := b.start(t)
|
||||
evctx, cancel := context.WithCancel(ctx)
|
||||
defer cancel()
|
||||
rt.deliverEvents(evctx)
|
||||
rt.until("the module to subscribe to the events it follows", 20*time.Second, func() bool {
|
||||
rt.mu.Lock()
|
||||
defer rt.mu.Unlock()
|
||||
return len(rt.subs) >= len(followed)
|
||||
})
|
||||
rt.untilHolding()
|
||||
|
||||
const head = "2222222222222222222222222222222222222222"
|
||||
const trunk = "3333333333333333333333333333333333333333"
|
||||
rt.mu.Lock()
|
||||
rt.openPulls = []OpenPull{{Owner: "novox", Repo: "app", Number: 8, Base: "main", HeadSHA: head, Required: []string{ctxCheck},
|
||||
Statuses: []string{}, Checks: map[string]HeadCheck{}}}
|
||||
rt.mu.Unlock()
|
||||
announced := pr("novox/app", 8, head, "fix/y", "src/b.go")
|
||||
announced.Body = "Reviewed-by: lead@shanks at " + head
|
||||
b.publish(t, "mesh.mod.gitea.event.pull.updated", announced)
|
||||
|
||||
// The checker's verdict at the head, then the forge's word that its bridge set mesh/check.
|
||||
hash, err := sdk.HeadsHash([]sdk.Head{{Repository: "novox/app", Number: 8, Head: head}})
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
verdict, _ := json.Marshal(map[string]any{"v": 1, "group": "fix/y", "heads": hash, "trunk": map[string]string{"novox/app": trunk},
|
||||
"state": "pass"})
|
||||
if _, err := b.kvs[checkerVerdicts].Put(ctx, "fix/y."+hash, verdict); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
line := "pass " + head[:12] + " on " + trunk[:12] + " | unit 3 run, 0 kept"
|
||||
rt.mu.Lock()
|
||||
rt.openPulls[0].Statuses = []string{ctxCheck}
|
||||
rt.openPulls[0].Checks = map[string]HeadCheck{ctxCheck: {State: "success", Description: line}}
|
||||
rt.mu.Unlock()
|
||||
set := b.publish(t, "mesh.mod.gitea.event.forge.status", map[string]any{"owner": "novox", "repo": "app", "sha": head,
|
||||
"context": ctxCheck, "state": "success", "description": line})
|
||||
|
||||
var key string
|
||||
rt.until("the ask to be written on mesh/check alone", 10*time.Second, func() bool { key = onlyAsk(b); return key != "" })
|
||||
t.Logf("the ask was written %s after mesh/check was set", time.Since(set))
|
||||
b.answer(t, key, OptionApprove)
|
||||
var handed sdk.Merge
|
||||
rt.until("the merge to be handed to the bridge", 10*time.Second, func() bool { handed = mergeAt(b, "fix/y"); return len(handed.Pulls) > 0 })
|
||||
if len(handed.Pulls) != 1 || handed.Pulls[0].Head.Head != head || handed.Pulls[0].State != sdk.MergeMerging {
|
||||
t.Fatalf("the merge handed is %+v", handed)
|
||||
}
|
||||
for _, verb := range []string{"delivery-check", "forget-checks"} {
|
||||
if n := len(rt.asksOf("seat:" + ControllerSeat + "." + verb)); n != 0 {
|
||||
t.Fatalf("the controller was asked %s %d times on a mesh/check base", verb, n)
|
||||
}
|
||||
}
|
||||
// No mesh/merge-gate is owed on the head.
|
||||
keys, _ := b.kvs["owed"].Keys(ctx)
|
||||
for _, k := range keys {
|
||||
if strings.Contains(k, "merge-gate") {
|
||||
t.Fatalf("a mesh/merge-gate was owed: %s", k)
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -1,245 +0,0 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
sdk "git.novox.be/novox/mesh-sdk/go/delivery"
|
||||
)
|
||||
|
||||
// The tests of issue 1302 (ADR 0447 §2, §4; ADR 0441 §1; ADR 0421 §1, §3; lead decisions L6 and the gate's part of
|
||||
// 1318): on a base that requires mesh/check the gate reads that one status. A delivery is ready, its Approve asked and its
|
||||
// merge handed on when mesh/check passed at its head and the checker's verdict covers every member's head as they are; no
|
||||
// check is asked of the controller, no composed check is asked, and no mesh/merge-gate is owed.
|
||||
|
||||
const (
|
||||
oneRepo, oneRepo2 = "novox/mesh-sdk", "novox/mesh-delivery"
|
||||
oneHead, oneHead2 = "e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5", "f6f6f6f6f6f6f6f6f6f6f6f6f6f6f6f6f6f6f6f6"
|
||||
oneTrunk = "a1a1a1a1a1a1a1a1a1a1a1a1a1a1a1a1a1a1a1a1"
|
||||
oneTrunk2 = "b2b2b2b2b2b2b2b2b2b2b2b2b2b2b2b2b2b2b2b2"
|
||||
)
|
||||
|
||||
// requiresCheck makes the forge say each pull request's base requires mesh/check alone, as gitea_open_pulls answers.
|
||||
func requiresCheck(w *world, pulls ...PullEvent) {
|
||||
w.forge.mu.Lock()
|
||||
defer w.forge.mu.Unlock()
|
||||
w.forge.open = nil
|
||||
for _, p := range pulls {
|
||||
w.forge.open = append(w.forge.open, OpenPull{Owner: p.Owner, Repo: p.Repo, Number: p.Number, Base: "main",
|
||||
HeadSHA: p.HeadSHA, Required: []string{ctxCheck}, Statuses: []string{}, Checks: map[string]HeadCheck{}})
|
||||
}
|
||||
}
|
||||
|
||||
// checkStatus is the forge's word that mesh/check was set on a head, with its line as the checker writes it.
|
||||
func checkStatus(repository, head, state, trunk string) ForgeStatusEvent {
|
||||
owner, repo, _ := strings.Cut(repository, "/")
|
||||
words := map[string]string{"success": "pass", "pending": "checking", "failure": "failed: a test"}[state]
|
||||
return ForgeStatusEvent{Owner: owner, Repo: repo, SHA: head, Context: ctxCheck, State: state,
|
||||
Description: words + " " + head[:12] + " on " + trunk[:12] + " | unit 4 run, 0 kept"}
|
||||
}
|
||||
|
||||
// putCheckerVerdict puts the checker's verdict for these heads in the gate's view of the checker's verdicts bucket.
|
||||
func putCheckerVerdict(w *world, state string, heads ...sdk.Head) {
|
||||
w.t.Helper()
|
||||
hash, err := sdk.HeadsHash(heads)
|
||||
if err != nil {
|
||||
w.t.Fatal(err)
|
||||
}
|
||||
trunk := map[string]string{}
|
||||
for _, h := range heads {
|
||||
trunk[h.Repository] = oneTrunk
|
||||
}
|
||||
value, _ := json.Marshal(map[string]any{"v": 1, "group": "g", "heads": hash, "trunk": trunk, "state": state})
|
||||
w.h.CheckerVerdictChanged("g."+hash, value)
|
||||
}
|
||||
|
||||
func headOf(repository string, number int, head string) sdk.Head {
|
||||
return sdk.Head{Repository: repository, Number: number, Head: head}
|
||||
}
|
||||
|
||||
func mergeGateOwed(w *world) []string {
|
||||
w.forge.mu.Lock()
|
||||
defer w.forge.mu.Unlock()
|
||||
var out []string
|
||||
for k := range w.forge.statuses {
|
||||
if strings.HasSuffix(k, " mesh/merge-gate") {
|
||||
out = append(out, k)
|
||||
}
|
||||
}
|
||||
return out
|
||||
}
|
||||
|
||||
// 1302 acceptance: the gate's table moves a lone delivery to ready on mesh/check alone, asks its Approve and merges it,
|
||||
// with no check asked of the controller and no mesh/merge-gate written.
|
||||
func TestALoneDeliveryGoesReadyAskedAndMergedOnMeshCheckAlone(t *testing.T) {
|
||||
w := newWorld(t)
|
||||
p := pr(oneRepo, 1, oneHead, "feat/one")
|
||||
requiresCheck(w, p)
|
||||
w.h.PullUpdated(p)
|
||||
w.settleAll()
|
||||
if got := w.ctl.checks; len(got) != 0 {
|
||||
t.Fatalf("a check was asked of the controller for a mesh/check base: %v", got)
|
||||
}
|
||||
id := IDOf(oneRepo, oneHead)
|
||||
if s := w.state(id); s != Proposed {
|
||||
t.Fatalf("before mesh/check it is %s, want proposed", s)
|
||||
}
|
||||
putCheckerVerdict(w, "pass", headOf(oneRepo, 1, oneHead))
|
||||
w.h.CommitStatus(checkStatus(oneRepo, oneHead, "success", oneTrunk))
|
||||
w.settleAll()
|
||||
if s := w.state(id); s == Proposed || s == Rejected {
|
||||
t.Fatalf("after mesh/check passed it is %s (%s), want ready or on", s, w.logs())
|
||||
}
|
||||
if d := w.delivery(id); d.Check == nil || d.Check.Status != ctxCheck || d.Check.Trunk != oneTrunk[:12] || d.Check.Head != oneHead[:12] {
|
||||
t.Fatalf("the verdict is not mesh/check's: %+v", d.Check)
|
||||
}
|
||||
if w.op.count() != 1 {
|
||||
t.Fatalf("its Approve was asked %d times, want once", w.op.count())
|
||||
}
|
||||
if len(w.forge.merges) != 1 || !strings.Contains(w.forge.merges[0], "mesh-sdk#1") {
|
||||
t.Fatalf("merges %v, want mesh-sdk#1", w.forge.merges)
|
||||
}
|
||||
if got := mergeGateOwed(w); len(got) != 0 {
|
||||
t.Fatalf("mesh/merge-gate was written: %v", got)
|
||||
}
|
||||
if len(w.ctl.checks) != 0 {
|
||||
t.Fatalf("a check was asked of the controller: %v", w.ctl.checks)
|
||||
}
|
||||
}
|
||||
|
||||
// 1302 (lead: the verdict read must be the group's): mesh/check passing on a head is not enough while the checker's
|
||||
// verdict for exactly the delivery's heads is not held: the Approve waits for it, and says so.
|
||||
func TestTheApproveWaitsForTheCheckersVerdictAtTheDeliverysHeads(t *testing.T) {
|
||||
w := newWorld(t)
|
||||
p := pr(oneRepo, 1, oneHead, "feat/one")
|
||||
requiresCheck(w, p)
|
||||
w.h.PullUpdated(p)
|
||||
w.h.CommitStatus(checkStatus(oneRepo, oneHead, "success", oneTrunk))
|
||||
w.settleAll()
|
||||
if w.op.count() != 0 {
|
||||
t.Fatalf("the Approve was asked with no checker verdict at the head")
|
||||
}
|
||||
if d := w.delivery(IDOf(oneRepo, oneHead)); d.Ready == nil || !strings.Contains(d.Ready.Waits, "checker") {
|
||||
t.Fatalf("it does not say it waits for the checker's verdict: %+v", d.Ready)
|
||||
}
|
||||
putCheckerVerdict(w, "running", headOf(oneRepo, 1, oneHead))
|
||||
w.settleAll()
|
||||
if w.op.count() != 0 {
|
||||
t.Fatalf("the Approve was asked on a running verdict")
|
||||
}
|
||||
putCheckerVerdict(w, "pass", headOf(oneRepo, 1, oneHead))
|
||||
w.settleAll()
|
||||
if w.op.count() != 1 {
|
||||
t.Fatalf("the Approve was asked %d times after the checker's pass, want once", w.op.count())
|
||||
}
|
||||
}
|
||||
|
||||
// 1302 acceptance (ADR 0447 §2, ADR 0441 §1): a group on mesh/check is asked once when every member's mesh/check passed
|
||||
// and the checker's verdict names every member's head; a member's mesh/check that changes withdraws the ask. No composed
|
||||
// check is asked of the controller and no member is owed a mesh/merge-gate.
|
||||
func TestAGroupOnMeshCheckIsAskedOnTheCheckersOneVerdictAndWithdrawnWhenAMemberChanges(t *testing.T) {
|
||||
w := newWorld(t)
|
||||
w.op.auto = false
|
||||
a, b := pr(oneRepo, 1, oneHead, "feat/two"), pr(oneRepo2, 2, oneHead2, "feat/two")
|
||||
requiresCheck(w, a, b)
|
||||
w.h.PullUpdated(a)
|
||||
w.h.PullUpdated(b)
|
||||
w.h.wantForge() // as the event's handler does for a head not seen before (main.go)
|
||||
w.settleAll()
|
||||
// One member's verdict alone, of a check that judged it by itself: not the group's.
|
||||
putCheckerVerdict(w, "pass", headOf(oneRepo, 1, oneHead))
|
||||
w.h.CommitStatus(checkStatus(oneRepo, oneHead, "success", oneTrunk))
|
||||
w.h.CommitStatus(checkStatus(oneRepo2, oneHead2, "success", oneTrunk))
|
||||
w.settleAll()
|
||||
if w.op.count() != 0 {
|
||||
t.Fatalf("the group was asked on a verdict that does not name every member's head")
|
||||
}
|
||||
putCheckerVerdict(w, "pass", headOf(oneRepo, 1, oneHead), headOf(oneRepo2, 2, oneHead2))
|
||||
w.settleAll()
|
||||
if w.op.count() != 1 || len(w.op.open()) != 1 {
|
||||
t.Fatalf("the group was asked %d times (open %v), want once", w.op.count(), w.op.open())
|
||||
}
|
||||
if len(w.ctl.checks) != 0 {
|
||||
t.Fatalf("a check was asked of the controller: %v", w.ctl.checks)
|
||||
}
|
||||
if got := mergeGateOwed(w); len(got) != 0 {
|
||||
t.Fatalf("mesh/merge-gate was owed: %v", got)
|
||||
}
|
||||
w.h.CommitStatus(checkStatus(oneRepo2, oneHead2, "pending", oneTrunk2))
|
||||
w.settleAll()
|
||||
if len(w.op.open()) != 0 {
|
||||
t.Fatalf("the group's ask stands though a member's mesh/check changed: %v", w.op.open())
|
||||
}
|
||||
}
|
||||
|
||||
// 1302: a failing mesh/check rejects the delivery; nothing is asked.
|
||||
func TestAFailingMeshCheckRejects(t *testing.T) {
|
||||
w := newWorld(t)
|
||||
p := pr(oneRepo, 1, oneHead, "feat/one")
|
||||
requiresCheck(w, p)
|
||||
w.h.PullUpdated(p)
|
||||
w.h.CommitStatus(checkStatus(oneRepo, oneHead, "failure", oneTrunk))
|
||||
w.settleAll()
|
||||
if s := w.state(IDOf(oneRepo, oneHead)); s != Rejected {
|
||||
t.Fatalf("it is %s, want rejected", s)
|
||||
}
|
||||
if w.op.count() != 0 || len(w.ctl.checks) != 0 {
|
||||
t.Fatalf("asked: operator %d, controller %v", w.op.count(), w.ctl.checks)
|
||||
}
|
||||
}
|
||||
|
||||
// 1302 (lead's (2)): the recheck verb refuses a delivery on a mesh/check base, saying the checker checks again on every
|
||||
// move and where to read it.
|
||||
func TestRecheckRefusesAMeshCheckDelivery(t *testing.T) {
|
||||
w := newWorld(t)
|
||||
p := pr(oneRepo, 1, oneHead, "feat/one")
|
||||
requiresCheck(w, p)
|
||||
w.h.PullUpdated(p)
|
||||
w.h.CommitStatus(checkStatus(oneRepo, oneHead, "failure", oneTrunk))
|
||||
w.settleAll()
|
||||
_, err := w.h.Recheck(IDOf(oneRepo, oneHead), "try again", "tester")
|
||||
if err == nil || !strings.Contains(err.Error(), "checker_pull") || !strings.Contains(err.Error(), "checker") {
|
||||
t.Fatalf("recheck answered %v, want a refusal naming the checker and checker_pull", err)
|
||||
}
|
||||
if len(w.ctl.checks) != 0 {
|
||||
t.Fatalf("a check was asked of the controller: %v", w.ctl.checks)
|
||||
}
|
||||
}
|
||||
|
||||
// 1318 (the gate's part): a pull request closed unmerged asks the controller to forget nothing: forget-checks is not called.
|
||||
func TestAClosedPullRequestAsksNoForgetChecks(t *testing.T) {
|
||||
w := newWorld(t)
|
||||
p := pr(oneRepo, 1, oneHead, "feat/one")
|
||||
w.h.PullUpdated(p)
|
||||
w.h.PullClosed(p)
|
||||
if len(w.ctl.forgot) != 0 {
|
||||
t.Fatalf("forget-checks was called: %v", w.ctl.forgot)
|
||||
}
|
||||
}
|
||||
|
||||
// 1302: the mesh/check line gives the head and the trunk it judged.
|
||||
func TestTheCheckLineGivesHeadAndTrunk(t *testing.T) {
|
||||
head, trunk := checkLine("pass " + oneHead[:12] + " on " + oneTrunk[:12] + " | unit 4 run, 0 kept | group g of 2 pull requests")
|
||||
if head != oneHead[:12] || trunk != oneTrunk[:12] {
|
||||
t.Fatalf("head %q trunk %q", head, trunk)
|
||||
}
|
||||
if head, trunk = checkLine("failed: a test"); head != "" || trunk != "" {
|
||||
t.Fatalf("a line that names no commits gave %q %q", head, trunk)
|
||||
}
|
||||
}
|
||||
|
||||
// 1302 with issue 1274's lost checks: a head on mesh/check never awaits a controller's check, so the build queue is not
|
||||
// read for it and nothing is asked again.
|
||||
func TestAHeadOnMeshCheckAwaitsNoControllersCheck(t *testing.T) {
|
||||
at := time.Date(2026, 10, 22, 12, 0, 0, 0, time.UTC)
|
||||
d := &Delivery{State: Proposed, Number: 1, CheckAsk: &CheckAsk{ID: "check-1", At: at}}
|
||||
if _, ok := awaitedSince(d); !ok {
|
||||
t.Fatal("the controller's asked check is not awaited")
|
||||
}
|
||||
d.Required = []string{ctxCheck}
|
||||
if _, ok := awaitedSince(d); ok {
|
||||
t.Fatal("a head on mesh/check awaits a controller's check")
|
||||
}
|
||||
}
|
||||
@@ -6,6 +6,7 @@ import (
|
||||
"fmt"
|
||||
"slices"
|
||||
"sort"
|
||||
"strconv"
|
||||
"strings"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
@@ -64,6 +65,8 @@ type Controller interface {
|
||||
StopWithout(walk, why, by string, without []string) (StopAnswer, error)
|
||||
// Walks are the walks the controller keeps — one, given its id — and whether this seat has a holder on record.
|
||||
Walks(walk string) (bool, []Walk, error)
|
||||
// ForgetChecks lets go of the builds a pull request's merge checks made for its merge (novox/hq ADR 0350 §5).
|
||||
ForgetChecks(owner, repo string, number int) error
|
||||
// Queue is every ask the build queue holds, waiting or in flight (the controller's `queue`, novox/hq issue 455): what
|
||||
// tells a check still on its way from one the queue lost (issue 1274). A dead ask is not held.
|
||||
Queue() ([]QueuedAsk, error)
|
||||
@@ -287,6 +290,11 @@ func (c seatController) BackupsAgain(walk, why string) error {
|
||||
return err
|
||||
}
|
||||
|
||||
func (c seatController) ForgetChecks(owner, repo string, number int) error {
|
||||
_, err := c.verb("forget-checks", map[string]any{"owner": owner, "repo": repo, "number": strconv.Itoa(number)})
|
||||
return err
|
||||
}
|
||||
|
||||
func (c seatController) Stop(walk, why, by string) error {
|
||||
_, err := c.verb("delivery-stop", map[string]any{"plan": walk, "why": why, "by": by})
|
||||
return err
|
||||
|
||||
@@ -299,8 +299,8 @@ func (h *Holder) sayAfterLine(d *Delivery) func() {
|
||||
// second Approve. Nil when nothing needs asking. Under the lock; the ask runs outside it.
|
||||
func (h *Holder) askFresh(d *Delivery, now time.Time) func() {
|
||||
if d.State != Proposed || d.outOfDate() == "" || d.AskedAgain != nil || !d.Check.Passes() || d.Base != MergeBranch ||
|
||||
d.Number == 0 || d.TrunkNow == "" || d.judgedByCheck() {
|
||||
return nil // a head on mesh/check is the checker's to judge again on its trunk's move (ADR 0447 §4)
|
||||
d.Number == 0 || d.TrunkNow == "" {
|
||||
return nil
|
||||
}
|
||||
// A group's member too (issue 1047, design 47: a move of the trunk asks every check again): the group's one Approve is
|
||||
// asked after its composed check, so a member ready before that has no Approve to carry; its check is asked at once
|
||||
@@ -383,9 +383,6 @@ func (r *ReadyRun) noteRefusal(d *Delivery, answer string) {
|
||||
// stays through a trunk move, ADR 0421 §3) and its check against the trunk's head is not on its way; one without an
|
||||
// Approve waits for that check first.
|
||||
func (h *Holder) readyWaits(d *Delivery) string {
|
||||
if d.judgedByCheck() && d.State == Proposed && d.outOfDate() != "" && h.covering(d) == nil {
|
||||
return "its mesh/check is " + d.outOfDate() + ": the checker checks it again against the trunk's head"
|
||||
}
|
||||
switch {
|
||||
case d.State == Ready:
|
||||
case d.State == Proposed && d.outOfDate() != "" && d.AskedAgain != nil:
|
||||
@@ -405,10 +402,6 @@ func (h *Holder) readyWaits(d *Delivery) string {
|
||||
return w
|
||||
}
|
||||
}
|
||||
// On mesh/check, the checker's verdict at exactly the heads the ask covers (issue 1302, ADR 0447 §2).
|
||||
if w := h.checkerWaits(d); w != "" {
|
||||
return w
|
||||
}
|
||||
if d.Group == "" {
|
||||
return ""
|
||||
}
|
||||
|
||||
@@ -54,10 +54,9 @@ const outOfDatePrefix = "out of date:"
|
||||
// minute and the check is asked at once, so ten minutes is a fault, never a queue.
|
||||
const UncheckedAfter = 10 * time.Minute
|
||||
|
||||
// mergeChecks are the statuses that are the mesh's check: mesh/check, the one main requires (novox/hq ADR 0447, issue
|
||||
// 1302), and the two it replaced, which hq's base (mesh/repo-check, ADR 0503 §5) and a reverted one still require. A base
|
||||
// requiring any is one the mesh is meant to check; only those a base requires are waited on.
|
||||
var mergeChecks = []string{ctxCheck, "mesh/merge-gate", "mesh/repo-check"}
|
||||
// mergeChecks are the statuses that are the mesh's merge check (novox/hq ADR 0237): a base requiring either is
|
||||
// one the mesh is meant to check.
|
||||
var mergeChecks = []string{"mesh/merge-gate", "mesh/repo-check"}
|
||||
|
||||
// OpenPull is an open pull request as the forge says it (the gitea module's gitea_open_pulls): its head, its base,
|
||||
// the contexts the base's protection requires, the contexts its head has a status of, in any state, and each of
|
||||
@@ -180,15 +179,6 @@ func (h *Holder) SeeOpenPulls() error {
|
||||
now := h.Now()
|
||||
seen := map[string]unchecked{}
|
||||
for _, p := range open {
|
||||
if slices.Contains(p.Required, ctxCheck) {
|
||||
// What the forge says of a head on a mesh/check base is its verdict (issue 1302): kept, and taken.
|
||||
if d := h.deliveries[IDOf(p.Owner+"/"+p.Repo, p.HeadSHA)]; d != nil {
|
||||
h.noteRequired(d, p.Required)
|
||||
if c := p.Checks[ctxCheck]; c.State != "" {
|
||||
h.takeCheckStatus(p.Owner+"/"+p.Repo, p.HeadSHA, c.State, c.Description)
|
||||
}
|
||||
}
|
||||
}
|
||||
waiting, pending, _ := p.unanswered()
|
||||
if len(waiting) == 0 {
|
||||
continue
|
||||
|
||||
@@ -152,9 +152,8 @@ type GroupLine struct {
|
||||
ID string `json:"id"`
|
||||
State string `json:"state"`
|
||||
Why string `json:"why"`
|
||||
// ComposedGate says the group's gate is the controller's composed check's: each member's own pull request check runs
|
||||
// the repository's check alone, and the composed verdict sets its mesh/merge-gate (novox/hq ADR 0403 §1, issue 978).
|
||||
// On mesh/check the group's check is the checker's one verdict at every member's head (ADR 0447 §2, issue 1302).
|
||||
// ComposedGate says the group's gate is the composed check's: each member's own pull request check runs the
|
||||
// repository's check alone, and the composed verdict sets its mesh/merge-gate (novox/hq ADR 0403 §1, issue 978).
|
||||
ComposedGate bool `json:"composed_gate"`
|
||||
Check string `json:"composed,omitempty"`
|
||||
Members []string `json:"members"`
|
||||
@@ -216,8 +215,8 @@ type StalledLine struct {
|
||||
State State `json:"state"`
|
||||
// Walk is the walk a line of the state Unheld is about (novox/hq issue 362).
|
||||
Walk string `json:"walk,omitempty"`
|
||||
// Waiting are the merge checks a line of the state Unanswered waits on, each as it reads — "mesh/check
|
||||
// never set", "mesh/check pending since <UTC time>" — so the controller's words name them without reading
|
||||
// Waiting are the merge checks a line of the state Unanswered waits on, each as it reads — "mesh/repo-check
|
||||
// never set", "mesh/merge-gate pending since <UTC time>" — so the controller's words name them without reading
|
||||
// Says (novox/hq issue 438).
|
||||
Waiting []string `json:"waiting,omitempty"`
|
||||
// Checks are the same merge checks as data — context, state, since when — so the controller says their times in
|
||||
@@ -304,10 +303,6 @@ func (h *Holder) Recheck(id, why, by string) (string, error) {
|
||||
h.mu.Unlock()
|
||||
return "", fmt.Errorf("no delivery %q", id)
|
||||
}
|
||||
if d.judgedByCheck() {
|
||||
h.mu.Unlock()
|
||||
return "", checkerRefusal(id) // the lead's decision for issue 1302: the checker serves no ask
|
||||
}
|
||||
if d.State == Proposed {
|
||||
// A proposed delivery's check is asked again when the build queue holds none of its head (issue 1274).
|
||||
h.mu.Unlock()
|
||||
@@ -675,10 +670,6 @@ func (h *Holder) Close(id, why string) (string, error) {
|
||||
// does not take spends nothing. Called under the lock; it unlocks.
|
||||
func (h *Holder) reask(d *Delivery, why string) (string, error) {
|
||||
id := d.ID
|
||||
if d.judgedByCheck() {
|
||||
h.mu.Unlock()
|
||||
return "", checkerRefusal(id)
|
||||
}
|
||||
trial := *d
|
||||
trial.Transitions = nil
|
||||
if _, err := Apply(&trial, EvReask, true, Facts{Now: h.Now(), By: byH2, Why: why}); err != nil {
|
||||
|
||||
@@ -46,6 +46,7 @@
|
||||
"seat:mesh-controller.delivery-backups-again",
|
||||
"seat:mesh-controller.delivery-walks",
|
||||
"seat:mesh-controller.delivery-check",
|
||||
"seat:mesh-controller.forget-checks",
|
||||
"seat:mesh-controller.queue",
|
||||
"seat:mesh-controller.plans",
|
||||
"gitea.gitea_commit_statuses",
|
||||
@@ -78,7 +79,6 @@
|
||||
"mesh-delivery.groups",
|
||||
"mesh-delivery.pulls",
|
||||
"mesh-delivery-teller.answers",
|
||||
"mesh-delivery-checker.verdicts",
|
||||
"mesh-controller.assignments",
|
||||
"node-engine.have"
|
||||
],
|
||||
|
||||
@@ -11,7 +11,6 @@ Each directory or archive this module declares says how it treats a file it did
|
||||
- `${dir:logs}`: everything in it is the program's; the mesh moves nothing out (ADR 0422 §2).
|
||||
- `${dir:data}`: everything in it is the program's; the mesh moves nothing out (ADR 0422 §2).
|
||||
- `${dir:server-lib}`: everything in it is the program's; the mesh moves nothing out (ADR 0422 §2).
|
||||
- `${dir:lib}`: everything in it is the program's; the mesh moves nothing out (ADR 0422 §2).
|
||||
- `${dir:db}`: everything in it is the program's; the mesh moves nothing out (ADR 0422 §2).
|
||||
- `${dir:rabbitmq}`: everything in it is the program's; the mesh moves nothing out (ADR 0422 §2).
|
||||
- `${dir:redis}`: everything in it is the program's; the mesh moves nothing out (ADR 0422 §2).
|
||||
@@ -23,8 +22,3 @@ Everything the container mounts sits in `${dir:box}` (`/var/lib/only-office/box`
|
||||
The image opens what is below each mount to every account on every start (`chmod -R 755`), so the barrier is a
|
||||
parent the container never sees. The module root stays root-owned 0711: the converter, which runs on the host
|
||||
as `only-office-converter`, passes through it to its two secrets and nothing else.
|
||||
|
||||
`${dir:lib}` (`/var/lib/only-office/lib`) is the copy the container used before, no longer mounted, kept by the
|
||||
mesh because it is not empty and closed to 0700. It keeps its id so mesh-host sets that mode: a new id at a path
|
||||
the image owns would be used as found (issue 339). Its contents, and those of the old `logs`, `data`, `db`,
|
||||
`rabbitmq`, `redis` and `fonts` beside it, are removed on the operator's word, and then `lib` leaves the manifest.
|
||||
|
||||
@@ -184,19 +184,16 @@ func TestConverterPassesOnlyTheRoot(t *testing.T) {
|
||||
}
|
||||
}
|
||||
|
||||
// 1843/AC3 (unit: manifest check)
|
||||
// Given the catalogue, when only-office's manifest is read, then the directory lib is declared with no
|
||||
// path (its old place, /var/lib/only-office/lib) and mode 0700, so the next apply closes the copy the
|
||||
// image opened, which mesh-host keeps because it is not empty; no container volume names ${dir:lib};
|
||||
// and every data entry points at a directory that is mounted now.
|
||||
func TestOldLibIsClosedAndUnmounted(t *testing.T) {
|
||||
// 1854/AC1 (unit: manifest check)
|
||||
// Given the catalogue, when only-office's manifest is read, then it declares no directory lib (the old
|
||||
// copy the container used before issue 1843, emptied on the operator's word), no container volume names
|
||||
// ${dir:lib}, and every data entry still points at a directory the server container mounts.
|
||||
func TestOldLibIsGone(t *testing.T) {
|
||||
m := readLayout(t)
|
||||
lib := m.dir(t, "lib")
|
||||
if lib.Path != "" || lib.Place != "" {
|
||||
t.Errorf("lib = path %q, place %q, want neither: it stays where the old copy is", lib.Path, lib.Place)
|
||||
}
|
||||
if lib.Mode != "0700" {
|
||||
t.Errorf("lib mode = %q, want 0700", lib.Mode)
|
||||
for _, r := range m.Resources {
|
||||
if r.ID == "lib" && r.Type == "directory" {
|
||||
t.Errorf("the manifest still declares the directory lib (path %q, mode %q)", r.Path, r.Mode)
|
||||
}
|
||||
}
|
||||
mounted := map[string]bool{}
|
||||
for _, v := range m.container(t, "server").Volumes {
|
||||
@@ -213,6 +210,22 @@ func TestOldLibIsClosedAndUnmounted(t *testing.T) {
|
||||
}
|
||||
}
|
||||
|
||||
// 1854/AC2 (unit: manifest check)
|
||||
// Given the catalogue, when only-office's README is read, then it names no ${dir:lib} and no longer
|
||||
// describes the old copy as kept.
|
||||
func TestReadmeNamesNoOldLib(t *testing.T) {
|
||||
raw, err := os.ReadFile("README.md")
|
||||
if err != nil {
|
||||
t.Fatalf("reading README.md: %v", err)
|
||||
}
|
||||
if strings.Contains(string(raw), "${dir:lib}") {
|
||||
t.Error("README.md still names ${dir:lib}")
|
||||
}
|
||||
if strings.Contains(string(raw), "/var/lib/only-office/lib") {
|
||||
t.Error("README.md still describes the old copy at /var/lib/only-office/lib")
|
||||
}
|
||||
}
|
||||
|
||||
type layoutResource struct {
|
||||
ID string `json:"id"`
|
||||
Type string `json:"type"`
|
||||
|
||||
@@ -188,12 +188,6 @@
|
||||
"mode": "0711",
|
||||
"unknown": "program"
|
||||
},
|
||||
{
|
||||
"id": "lib",
|
||||
"type": "directory",
|
||||
"mode": "0700",
|
||||
"unknown": "program"
|
||||
},
|
||||
{
|
||||
"id": "db",
|
||||
"type": "directory",
|
||||
|
||||
+53
-26
@@ -10,10 +10,10 @@ that seat (ADR 0207).
|
||||
(ADR 0207).
|
||||
- **`/etc/pacman.conf`, whole.** A block cannot be added to `[options]` by appending: anything added
|
||||
at the end of the file lands in the last repository's section. So the module owns the file:
|
||||
- The repositories are the union of what the four machines had enabled on 2026-10-04: `core`,
|
||||
`extra` and `multilib`. All four had all three.
|
||||
- The repositories are the union of what the mesh's machines had enabled on 2026-10-04: `core`,
|
||||
`extra` and `multilib`. All had all three.
|
||||
- The options are the distribution's defaults, plus `Color`, `ParallelDownloads = 5`,
|
||||
`VerbosePkgLists`, and `DownloadUser = alpm` (pacman 7; the `alpm` user exists on all four).
|
||||
`VerbosePkgLists`, and `DownloadUser = alpm` (pacman 7; the `alpm` user exists on every machine).
|
||||
- The manifest test runs `pacman-conf` on the rendered file and checks the repository list and the
|
||||
options as pacman reads them. It skips that check where `pacman-conf` is absent.
|
||||
- The host keeps the machine's previous file once, the first time it writes over it (ADR 0102).
|
||||
@@ -23,23 +23,44 @@ that seat (ADR 0207).
|
||||
list stays reflector's to write, not the mesh's.
|
||||
- **Cache cleaning.** The `pacman-contrib` package and `paccache.timer` running and enabled. Each
|
||||
week it keeps the last three versions of each package.
|
||||
- **The weekly upgrade (ADR 0477 clause 5, issue 1574).** `mesh-pacman-weekly-upgrade.timer` and its
|
||||
oneshot service, placed whole in `/etc/systemd/system`. The timer fires every Saturday at 04:00 in the
|
||||
node's local time, the same moment on every node (no random delay), and is `Persistent=true`: a node that
|
||||
was off or asleep then upgrades at its next start. The service runs `/usr/bin/pacman -Syu --noconfirm`, the
|
||||
same arguments `pacman_upgrade` starts (`UpgradeArgs` in the code; a manifest test holds the two together).
|
||||
No other moment upgrades but a hand `pacman_upgrade`, and the mesh restarts no machine for an upgrade: a new
|
||||
kernel waits for the next restart.
|
||||
- **The outcome is the module's health.** The timer's resource asks `pacman_upgrade_health`: unhealthy,
|
||||
naming the exit and the last `error:` line of the journal, when the last weekly upgrade failed; healthy
|
||||
while it runs, after it passed, and before it ever ran. The controller raises the module's condition for
|
||||
the node from it. A failed service stays failed until
|
||||
the next run, so the condition stands until an upgrade passes; a hand `pacman_upgrade` does not clear
|
||||
it. A restart of the machine clears the failed state and with it the condition, and the next Saturday
|
||||
runs again. A run that goes on longer than two hours is unhealthy too, naming when it started and the
|
||||
last line of its journal: it is hung and holds pacman's database lock. The service has no start
|
||||
timeout on purpose, because killing pacman in the middle of a transaction is worse; the hung run is
|
||||
said, not killed.
|
||||
- **The package view on the bus (ADR 0507).** A process of its own, the keeper (`pacman-tools keeper`, account
|
||||
`pacman-keeper`, its own bus account), keeps the node's packages in the state `installed` and what an upgrade
|
||||
would change in `updates`, written only on change: at its start, whenever pacman's database lock is released
|
||||
(an inotify watch), when reflector rewrites the mirror list, and once a day. Each change of the packages is
|
||||
`package.changed`. A list over the bus's cap on a value is not written: the key keeps its last value and
|
||||
`<node>.not-current` says the size and since when, which the health says too. `pacman_packages`,
|
||||
`pacman_updates`, `pacman_installed` and `pacman_foreign` answer every node from these states, and what no
|
||||
module declares from the controller's `declared-packages`, without asking a machine.
|
||||
- **Upgrades as runs, one node at a time (ADR 0508).** Every full upgrade and every rollback is a run in the
|
||||
state `runs` at `<node>.<run>`, written at each step and emitted as it: the package record taken and read back
|
||||
(`records`, kept `record-days`), the DKMS gate before the transaction (a kernel an installed DKMS module's own
|
||||
`Linux-Maximum` or `BUILD_EXCLUSIVE_KERNEL` excludes is held with its headers, `upgrade.kernel-held`), the
|
||||
transaction as a unit of its own, its end, and the after-upgrade check (`.pacnew`/`.pacsave` files, the news
|
||||
since the previous upgrade, units newly failed, a restart needed, processes still mapping a replaced file by
|
||||
unit and executable, DKMS modules not built for an installed kernel). pacman's log is read once into `runs`
|
||||
as history.
|
||||
- **The round.** The nodes upgrade in the order of the mesh-level key `upgrade-order` (a trusted list,
|
||||
`predefined.json`; the module's config value of the same name falls back to it). On Saturday at 04:00 local
|
||||
time the first node of the order opens the week's round in `rounds`; each node's keeper watches `rounds` and
|
||||
`runs` and starts its own run when the node before it passed its after-upgrade check. A failed run or check
|
||||
stops the round there; only the operator passes a node (`pacman_round` with `pass` and a reason asks on the
|
||||
operator's channel, and the round goes on only on the operator's Approve). With no order set no round opens.
|
||||
- **Acts are asks.** `pacman_upgrade` and `pacman_rollback` with `apply` write an ask in `asks` at
|
||||
`<node>.<id>`, which that node's keeper alone takes.
|
||||
- **The root helper (ADR 0510).** The keeper never runs as root. The acts that need root go through a second
|
||||
process of the same build, `pacman-root` (`pacman-tools root`), running as root with no bus account, on a
|
||||
unix socket in `${dir:state}/root/` that only the keeper's account may open, the peer's account checked on
|
||||
every connection. It serves four verbs and refuses everything else: start a run's full upgrade as the unit
|
||||
`mesh-pacman-upgrade-<run>` (leaving out held packages), start a rollback's unit with the plan in the request
|
||||
(packages from the cache or the Arch Linux Archive, the dated repository's file found by name), read which
|
||||
processes map a replaced file, and read a transaction unit's journal. No sudo rule is placed.
|
||||
- **The health is read from the bus.** The keeper's health check (`pacman_upgrade_health`) is unhealthy while
|
||||
no `upgrade-order` is set; while the node's latest run was refused, failed or did not pass its check (until
|
||||
a later run passes); while its transaction goes on past two hours (hung, said and not killed); while a
|
||||
kernel is held; and while the round stands still at this node or at the node it waits for, naming it and
|
||||
since when. A node that slept through a run is judged by what the bus holds, not by a unit's age (issue
|
||||
1579). The weekly timer of ADR 0477 §5 and its service are dropped from the manifest, so the node-engine
|
||||
removes them.
|
||||
|
||||
## What it improves
|
||||
|
||||
@@ -62,21 +83,27 @@ that seat (ADR 0207).
|
||||
|---|---|---|
|
||||
| `pacman_search` | r | `pacman -Ss`: repository, name, version, groups, installed and at which version, description |
|
||||
| `pacman_info` | r | `-Qi`, or `-Si` when not installed, with lists as lists |
|
||||
| `pacman_installed` | r | every package with version, explicit or dependency, foreign; filters and totals |
|
||||
| `pacman_installed` | r | every package with version, explicit or dependency, foreign; filters and totals, from `installed` on the bus |
|
||||
| `pacman_packages` | r | every node's packages from the bus: one package everywhere, differing versions, packages on some nodes only, each node's explicit packages no module declares, foreign apart, nodes not current |
|
||||
| `pacman_owns` | r | which package owns a path, or `owned: false` |
|
||||
| `pacman_files` | r | what a package placed, bounded |
|
||||
| `pacman_updates` | r | `checkupdates`: what a full upgrade would change, never setting up a partial upgrade |
|
||||
| `pacman_upgrade` | a | starts `pacman -Syu --noconfirm` (sudo -n) as a transient unit that outlives the call; answers the unit and the news since the last upgrade; given the unit, how it went |
|
||||
| `pacman_upgrade_health` | r | whether the weekly upgrade last passed: the timer's health check (see the weekly upgrade above) |
|
||||
| `pacman_updates` | r | what a full upgrade would change, for one node or all, from `updates` on the bus |
|
||||
| `pacman_upgrade` | a | asks a node's keeper for a full upgrade as a run (an ask in `asks`); answers the ask and the news since the last upgrade; given a transaction unit, how it goes |
|
||||
| `pacman_upgrade_health` | r | the keeper's health check, read from the bus (see the health above) |
|
||||
| `pacman_upgrades` | r | every run of every node from `runs`: narrowed to a node and days, or one run whole |
|
||||
| `pacman_round` | r/a | where every node stands in the newest rounds; `pass` with a reason asks the operator to let a round go on past a node |
|
||||
| `pacman_rollback` | r/a | what returning a node to a run's record would change, each package with its source; `apply` asks the node's keeper to run it |
|
||||
| `pacman_orphans` | r | `pacman -Qdt` |
|
||||
| `pacman_remove_orphans` | a | `pacman -Rs` on named orphans, or all of them, as a unit of its own; a name that is not an orphan is refused |
|
||||
| `pacman_cache` | r/a | size, interrupted downloads, what paccache would free keeping N; `clean: true` removes them |
|
||||
| `pacman_history` | r | `/var/log/pacman.log`: installs, upgrades, downgrades, reinstalls and removals since a day, and the last full upgrade |
|
||||
| `pacman_mirrors` | r/a | the list, its generator and age, reflector's options, timer and last run; `refresh: true` starts reflector |
|
||||
| `pacman_foreign` | r | `pacman -Qm` |
|
||||
| `pacman_foreign` | r | the foreign packages, from `installed` on the bus |
|
||||
| `pacman_news` | r | the distribution's news since the last full upgrade (or a day), over https; no network is `reachable: false` |
|
||||
| `pacman_config` | r | `pacman-conf`: options, repositories, and whether `/etc/pacman.conf` is the module's |
|
||||
|
||||
A transaction never runs as the tool's own child. An upgrade takes longer than the 20 s a call may
|
||||
take, and a pacman killed mid-transaction leaves a half-upgraded machine and a lock. So a transaction
|
||||
runs as a transient unit (`systemd-run`, named `mesh-pacman-…`), and its log is read from the journal.
|
||||
runs as a transient unit (`systemd-run`, named `mesh-pacman-…`), and its log is read from the journal. A
|
||||
run's transaction is started by the root helper; removing orphans and cleaning the cache still go through
|
||||
`sudo -n` from the tools bundle.
|
||||
|
||||
@@ -55,30 +55,11 @@ func (m *Machine) journal(unit string, n int) []string {
|
||||
return lines(out)
|
||||
}
|
||||
|
||||
// UpgradeArgs is the full system upgrade, in one place: pacman_upgrade starts pacman with these arguments as
|
||||
// a transient unit, and the weekly upgrade's service (files/mesh-pacman-weekly-upgrade.service) runs
|
||||
// pacman with the same ones; a test holds the two together (issue 1574, ADR 0477 clause 5).
|
||||
// UpgradeArgs is the full system upgrade, in one place: the root helper starts a run's transaction with these
|
||||
// arguments (ADR 0510), and the packages a run holds back are added to them as --ignore.
|
||||
var UpgradeArgs = []string{"-Syu", "--noconfirm"}
|
||||
|
||||
// Upgrade starts a full system upgrade as a transient unit and answers at once, with the
|
||||
// distribution's news since the last upgrade; or, given a unit it started, answers how it went.
|
||||
func (m *Machine) Upgrade(unit string, n int) (map[string]any, error) {
|
||||
if unit != "" {
|
||||
return m.UpgradeStatus(unit, n)
|
||||
}
|
||||
news := m.News("")
|
||||
started, _, err := m.transaction("upgrade", false, UpgradeArgs...)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return map[string]any{
|
||||
"started": started,
|
||||
"follow": "call pacman_upgrade with this unit to read how it goes",
|
||||
"news": news,
|
||||
}, nil
|
||||
}
|
||||
|
||||
var unitName = regexp.MustCompile(`^` + unitPrefix + `[a-z-]+-[0-9]+$`)
|
||||
var unitName = regexp.MustCompile(`^` + unitPrefix + `[a-z-]+-[0-9a-z]+$`)
|
||||
|
||||
// UpgradeStatus is a transaction unit's state and the tail of what it logged.
|
||||
func (m *Machine) UpgradeStatus(unit string, n int) (map[string]any, error) {
|
||||
|
||||
@@ -0,0 +1,151 @@
|
||||
package main
|
||||
|
||||
// The after-upgrade check (novox/hq ADR 0508 §4), written to the run as its last step: the .pacnew and .pacsave
|
||||
// files the transaction left, the distribution's news since the node's previous upgrade, the units failed now
|
||||
// that were not failed in the record, and whether a restart is needed (the running kernel is no longer the one
|
||||
// its package installed). It passes when the transaction succeeded and no unit newly failed (and, with the
|
||||
// second part, every DKMS module is built); the rest is said and does not stop the round. The mesh restarts no
|
||||
// machine (ADR 0477 §5).
|
||||
|
||||
import (
|
||||
"sort"
|
||||
"strings"
|
||||
"time"
|
||||
)
|
||||
|
||||
// AfterCheck is the after-upgrade check of a run whose transaction ended, against the record taken before it
|
||||
// (nil when there is none: every unit failed now is then said as newly failed).
|
||||
func (u *Upgrades) AfterCheck(run Run, rec *Record) Check {
|
||||
c := Check{Pacnew: []string{}, Pacsave: []string{}, News: []NewsTitle{}, Mapped: []Mapped{}, NewlyFailed: []string{}, Unbuilt: []DKMSBuild{}}
|
||||
succeeded := transactionSucceeded(run)
|
||||
var started time.Time
|
||||
for _, s := range run.Steps {
|
||||
if s.Step == StepStarted {
|
||||
started = s.At
|
||||
}
|
||||
}
|
||||
if !started.IsZero() {
|
||||
if b := u.transactionBlock(run.Kind, started); b != nil {
|
||||
c.Pacnew = append(c.Pacnew, b.Pacnew...)
|
||||
c.Pacsave = append(c.Pacsave, b.Pacsave...)
|
||||
}
|
||||
}
|
||||
c.News, c.NewsError = u.newsSince(u.previousUpgrade(run))
|
||||
healthy := true
|
||||
if failed, err := u.Machine.FailedUnits(); err != nil {
|
||||
healthy = false
|
||||
c.NewlyFailed = append(c.NewlyFailed, "(the failed units could not be read: "+err.Error()+")")
|
||||
} else {
|
||||
before := map[string]bool{}
|
||||
if rec != nil {
|
||||
for _, f := range rec.FailedUnits {
|
||||
before[f] = true
|
||||
}
|
||||
}
|
||||
for _, f := range failed {
|
||||
if !before[f] {
|
||||
c.NewlyFailed = append(c.NewlyFailed, f)
|
||||
}
|
||||
}
|
||||
}
|
||||
c.RunningKernel, c.InstalledKernel, c.RestartNeeded = u.restartNeeded()
|
||||
c.Passed = succeeded && healthy && len(c.NewlyFailed) == 0
|
||||
u.afterCheckII(&c)
|
||||
return c
|
||||
}
|
||||
|
||||
// previousUpgrade is when the node's upgrade before this run started: its previous run with a transaction.
|
||||
func (u *Upgrades) previousUpgrade(run Run) time.Time {
|
||||
runs, err := u.ownRuns()
|
||||
if err != nil {
|
||||
return time.Time{}
|
||||
}
|
||||
var last time.Time
|
||||
for _, r := range runs {
|
||||
if r.Run.Run >= run.Run || r.Kind != RunUpgrade {
|
||||
continue
|
||||
}
|
||||
for _, s := range r.Steps {
|
||||
if s.Step == StepStarted && s.At.After(last) {
|
||||
last = s.At
|
||||
}
|
||||
}
|
||||
}
|
||||
return last
|
||||
}
|
||||
|
||||
// newsSince is the distribution's news published after a time, titles only; or why it could not be read.
|
||||
func (u *Upgrades) newsSince(since time.Time) ([]NewsTitle, string) {
|
||||
body, err := fetch(NewsFeed)
|
||||
if err != nil {
|
||||
return []NewsTitle{}, err.Error()
|
||||
}
|
||||
items, err := ParseNews(body, since)
|
||||
if err != nil {
|
||||
return []NewsTitle{}, "the feed could not be read: " + err.Error()
|
||||
}
|
||||
titles := []NewsTitle{}
|
||||
for _, it := range items {
|
||||
titles = append(titles, NewsTitle{Title: it.Title, Link: it.Link, Published: it.Published})
|
||||
}
|
||||
return titles, ""
|
||||
}
|
||||
|
||||
// kernelTag is a release's flavour, its letters alone: `7.2.8-arch1-2` is arch, `6.18.54-2-lts` is lts.
|
||||
func kernelTag(release string) string {
|
||||
return strings.Map(func(r rune) rune {
|
||||
if r >= 'a' && r <= 'z' {
|
||||
return r
|
||||
}
|
||||
return -1
|
||||
}, release)
|
||||
}
|
||||
|
||||
// installedKernels are the releases the installed kernel packages install, sorted.
|
||||
func (m *Machine) installedKernels() ([]string, error) {
|
||||
out, err := m.Out("pacman", "-Qq")
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
var releases []string
|
||||
for _, name := range lines(out) {
|
||||
name = strings.TrimSpace(name)
|
||||
if name != "linux" && !strings.HasPrefix(name, "linux-") {
|
||||
continue
|
||||
}
|
||||
release, kernel, err := m.kernelRelease(name)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
if kernel {
|
||||
releases = append(releases, release)
|
||||
}
|
||||
}
|
||||
sort.Strings(releases)
|
||||
return releases, nil
|
||||
}
|
||||
|
||||
// restartNeeded compares the running kernel with what its package installs now: when no installed kernel is the
|
||||
// running one, the one of its flavour is what a restart would boot.
|
||||
func (u *Upgrades) restartNeeded() (running, installed string, needed bool) {
|
||||
raw, err := u.Machine.ReadFile(OSRelease)
|
||||
if err != nil {
|
||||
return "(the running kernel could not be read: " + err.Error() + ")", "", false
|
||||
}
|
||||
running = strings.TrimSpace(string(raw))
|
||||
kernels, err := u.Machine.installedKernels()
|
||||
if err != nil {
|
||||
return running, "(the installed kernels could not be read: " + err.Error() + ")", false
|
||||
}
|
||||
for _, k := range kernels {
|
||||
if k == running {
|
||||
return running, k, false
|
||||
}
|
||||
}
|
||||
for _, k := range kernels {
|
||||
if kernelTag(k) == kernelTag(running) {
|
||||
return running, k, true
|
||||
}
|
||||
}
|
||||
return running, "", true
|
||||
}
|
||||
@@ -0,0 +1,121 @@
|
||||
package main
|
||||
|
||||
// The after-upgrade check, its first part (novox/hq ADR 0508 §4; acceptance 1841/AC5): .pacnew and .pacsave
|
||||
// files from the transaction's log lines, the news since the previous upgrade, units newly failed against the
|
||||
// record, and a restart needed.
|
||||
|
||||
import (
|
||||
"errors"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
)
|
||||
|
||||
const checkLog = "[2026-10-04T12:00:01+0000] [PACMAN] Running 'pacman -Syu --noconfirm'\n" +
|
||||
"[2026-10-04T12:00:02+0000] [PACMAN] starting full system upgrade\n" +
|
||||
"[2026-10-04T12:00:09+0000] [ALPM] transaction started\n" +
|
||||
"[2026-10-04T12:00:10+0000] [ALPM] upgraded pacman-mirrorlist (20260901-1 -> 20261001-1)\n" +
|
||||
"[2026-10-04T12:00:10+0000] [ALPM] warning: /etc/pacman.d/mirrorlist installed as /etc/pacman.d/mirrorlist.pacnew\n" +
|
||||
"[2026-10-04T12:00:10+0000] [ALPM] warning: /etc/lemurs/config.toml saved as /etc/lemurs/config.toml.pacsave\n" +
|
||||
"[2026-10-04T12:00:10+0000] [ALPM] removed lemurs-git (0.4.0.r17.g7151336-2)\n" +
|
||||
"[2026-10-04T12:00:11+0000] [ALPM] transaction completed\n" +
|
||||
"[2026-10-04T12:05:00+0000] [PACMAN] Running 'pacman -S --noconfirm made-up'\n" +
|
||||
"[2026-10-04T12:05:01+0000] [ALPM] transaction started\n" +
|
||||
"[2026-10-04T12:05:01+0000] [ALPM] warning: /etc/made-up.conf installed as /etc/made-up.conf.pacnew\n" +
|
||||
"[2026-10-04T12:05:02+0000] [ALPM] transaction completed\n"
|
||||
|
||||
const newsFeed = `<?xml version="1.0"?><rss><channel>
|
||||
<item><title>A manual intervention</title><link>https://example.org/news/2</link><pubDate>Fri, 02 Oct 2026 10:00:00 +0000</pubDate><description>x</description></item>
|
||||
<item><title>An older post</title><link>https://example.org/news/1</link><pubDate>Mon, 21 Sep 2026 10:00:00 +0000</pubDate><description>y</description></item>
|
||||
</channel></rss>`
|
||||
|
||||
// checkMachine is a machine after an upgrade: these units failed now, this kernel running, and the linux
|
||||
// package now installing linuxNow.
|
||||
func checkMachine(failed, linuxNow string) *Machine {
|
||||
m := machine(byLine(map[string]Ran{
|
||||
"systemctl list-units --state=failed --plain --no-legend --no-pager": {Stdout: failed},
|
||||
"pacman -Qq": {Stdout: "bash\nlinux\nlinux-firmware\nlinux-lts\n"},
|
||||
"dkms status": {Status: 127, Err: "ENOENT"},
|
||||
"pacman -Qlq -- linux": {Stdout: "/usr/lib/modules/" + linuxNow + "/vmlinuz\n"},
|
||||
"pacman -Qlq -- linux-firmware": {Stdout: "/usr/lib/firmware/\n"},
|
||||
"pacman -Qlq -- linux-lts": {Stdout: "/usr/lib/modules/6.18.54-2-lts/vmlinuz\n"},
|
||||
}, nil), 1000)
|
||||
m.ReadFile = func(path string) ([]byte, error) {
|
||||
switch path {
|
||||
case OSRelease:
|
||||
return []byte("7.2.8-arch1-2\n"), nil
|
||||
case PacmanLog:
|
||||
return []byte(checkLog), nil
|
||||
}
|
||||
return nil, errNoFile
|
||||
}
|
||||
return m
|
||||
}
|
||||
|
||||
func checkedRun(steps ...string) Run {
|
||||
r := Run{Node: "node-a", Run: "20261004T120000Z", Kind: RunUpgrade, From: FromAsk}
|
||||
at := time.Date(2026, 10, 4, 12, 0, 0, 0, time.UTC)
|
||||
for _, s := range steps {
|
||||
r.Steps = append(r.Steps, Step{Step: s, At: at})
|
||||
}
|
||||
return r
|
||||
}
|
||||
|
||||
func checkUpgrades(t *testing.T, m *Machine) *Upgrades {
|
||||
runs := newBucket()
|
||||
previous := Run{Node: "node-a", Run: "20260926T040000Z", Kind: RunUpgrade, From: FromRound,
|
||||
Steps: []Step{{Step: StepStarted, At: time.Date(2026, 9, 26, 4, 0, 0, 0, time.UTC)}}}
|
||||
if _, err := runs.Create(NodeKey("node-a", previous.Run), previous); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
saved := fetch
|
||||
fetch = func(string) ([]byte, error) { return []byte(newsFeed), nil }
|
||||
t.Cleanup(func() { fetch = saved })
|
||||
return &Upgrades{Node: "node-a", Machine: m, Runs: runs}
|
||||
}
|
||||
|
||||
func TestTheCheckNamesTheFilesTheTransactionLeftAndTheNewsSinceThePreviousUpgrade(t *testing.T) {
|
||||
u := checkUpgrades(t, checkMachine("", "7.2.8-arch1-2"))
|
||||
c := u.AfterCheck(checkedRun(StepStarted, StepFinished), &Record{})
|
||||
if strings.Join(c.Pacnew, " ") != "/etc/pacman.d/mirrorlist.pacnew" || strings.Join(c.Pacsave, " ") != "/etc/lemurs/config.toml.pacsave" {
|
||||
t.Errorf("only this transaction's files: %v %v", c.Pacnew, c.Pacsave)
|
||||
}
|
||||
if len(c.News) != 1 || c.News[0].Title != "A manual intervention" || c.NewsError != "" {
|
||||
t.Errorf("the news since the previous upgrade (26 September): %+v %q", c.News, c.NewsError)
|
||||
}
|
||||
if !c.Passed || c.RestartNeeded || c.RunningKernel != "7.2.8-arch1-2" || c.InstalledKernel != "7.2.8-arch1-2" {
|
||||
t.Errorf("%+v", c)
|
||||
}
|
||||
}
|
||||
|
||||
func TestNewsThatCannotBeReadIsSaidNotShownAsNone(t *testing.T) {
|
||||
u := checkUpgrades(t, checkMachine("", "7.2.8-arch1-2"))
|
||||
fetch = func(string) ([]byte, error) { return nil, errors.New("no route to host") }
|
||||
c := u.AfterCheck(checkedRun(StepStarted, StepFinished), &Record{})
|
||||
if !strings.Contains(c.NewsError, "no route to host") || !c.Passed {
|
||||
t.Errorf("%+v", c)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAUnitNewlyFailedFailsTheCheckAndOneFailedBeforeDoesNot(t *testing.T) {
|
||||
u := checkUpgrades(t, checkMachine("backup.service loaded failed failed The backup\nweb.service loaded failed failed The web\n", "7.2.8-arch1-2"))
|
||||
c := u.AfterCheck(checkedRun(StepStarted, StepFinished), &Record{FailedUnits: []string{"backup.service"}})
|
||||
if strings.Join(c.NewlyFailed, " ") != "web.service" || c.Passed {
|
||||
t.Errorf("%+v", c)
|
||||
}
|
||||
}
|
||||
|
||||
func TestARestartIsNeededWhenTheRunningKernelIsNoLongerTheInstalledOne(t *testing.T) {
|
||||
u := checkUpgrades(t, checkMachine("", "7.2.9-arch1-1"))
|
||||
c := u.AfterCheck(checkedRun(StepStarted, StepFinished), &Record{})
|
||||
if !c.RestartNeeded || c.InstalledKernel != "7.2.9-arch1-1" || !c.Passed {
|
||||
t.Errorf("a restart is said and does not fail the check: %+v", c)
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheCheckPassesOnlyWhenTheTransactionSucceeded(t *testing.T) {
|
||||
u := checkUpgrades(t, checkMachine("", "7.2.8-arch1-2"))
|
||||
if c := u.AfterCheck(checkedRun(StepStarted, StepFailed), &Record{}); c.Passed {
|
||||
t.Errorf("%+v", c)
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,153 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"sort"
|
||||
"strings"
|
||||
"sync"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
)
|
||||
|
||||
// memBucket is a state in memory, with the bus's revisions and its refusal of a lost race.
|
||||
type memBucket struct {
|
||||
mu sync.Mutex
|
||||
values map[string][]byte
|
||||
revs map[string]uint64
|
||||
rev uint64
|
||||
writes []string // "op key", in order
|
||||
fail map[string]error
|
||||
}
|
||||
|
||||
func newBucket() *memBucket {
|
||||
return &memBucket{values: map[string][]byte{}, revs: map[string]uint64{}, fail: map[string]error{}}
|
||||
}
|
||||
|
||||
func (b *memBucket) Get(key string) (*stdio.StateEntry, error) {
|
||||
b.mu.Lock()
|
||||
defer b.mu.Unlock()
|
||||
if err := b.fail["get "+key]; err != nil {
|
||||
return nil, err
|
||||
}
|
||||
v, ok := b.values[key]
|
||||
if !ok {
|
||||
return nil, nil
|
||||
}
|
||||
return &stdio.StateEntry{Key: key, Value: json.RawMessage(v), Revision: b.revs[key]}, nil
|
||||
}
|
||||
|
||||
func (b *memBucket) write(op, key string, value any) (uint64, error) {
|
||||
if err := b.fail[op+" "+key]; err != nil {
|
||||
return 0, err
|
||||
}
|
||||
data, err := json.Marshal(value)
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
b.rev++
|
||||
b.values[key], b.revs[key] = data, b.rev
|
||||
b.writes = append(b.writes, op+" "+key)
|
||||
return b.rev, nil
|
||||
}
|
||||
|
||||
func (b *memBucket) Create(key string, value any) (uint64, error) {
|
||||
b.mu.Lock()
|
||||
defer b.mu.Unlock()
|
||||
if _, ok := b.values[key]; ok {
|
||||
return 0, fmt.Errorf("%s: %w", key, stdio.ErrChanged)
|
||||
}
|
||||
return b.write("create", key, value)
|
||||
}
|
||||
|
||||
func (b *memBucket) Update(key string, value any, revision uint64) (uint64, error) {
|
||||
b.mu.Lock()
|
||||
defer b.mu.Unlock()
|
||||
if b.revs[key] != revision {
|
||||
return 0, fmt.Errorf("%s: %w", key, stdio.ErrChanged)
|
||||
}
|
||||
return b.write("update", key, value)
|
||||
}
|
||||
|
||||
func (b *memBucket) Put(key string, value any) (uint64, error) {
|
||||
b.mu.Lock()
|
||||
defer b.mu.Unlock()
|
||||
return b.write("put", key, value)
|
||||
}
|
||||
|
||||
func (b *memBucket) Delete(key string) error {
|
||||
b.mu.Lock()
|
||||
defer b.mu.Unlock()
|
||||
if err := b.fail["delete "+key]; err != nil {
|
||||
return err
|
||||
}
|
||||
delete(b.values, key)
|
||||
delete(b.revs, key)
|
||||
b.writes = append(b.writes, "delete "+key)
|
||||
return nil
|
||||
}
|
||||
|
||||
func (b *memBucket) Keys() ([]string, error) {
|
||||
b.mu.Lock()
|
||||
defer b.mu.Unlock()
|
||||
keys := []string{}
|
||||
for k := range b.values {
|
||||
keys = append(keys, k)
|
||||
}
|
||||
sort.Strings(keys)
|
||||
return keys, nil
|
||||
}
|
||||
|
||||
// value reads a key into its shape; a missing key is the zero value and false.
|
||||
func (b *memBucket) value(key string, v any) bool {
|
||||
b.mu.Lock()
|
||||
defer b.mu.Unlock()
|
||||
raw, ok := b.values[key]
|
||||
if ok {
|
||||
if err := json.Unmarshal(raw, v); err != nil {
|
||||
panic(err)
|
||||
}
|
||||
}
|
||||
return ok
|
||||
}
|
||||
|
||||
// writesTo are the writes to keys starting with prefix.
|
||||
func (b *memBucket) writesTo(prefix string) []string {
|
||||
b.mu.Lock()
|
||||
defer b.mu.Unlock()
|
||||
var out []string
|
||||
for _, w := range b.writes {
|
||||
if _, key, _ := strings.Cut(w, " "); strings.HasPrefix(key, prefix) {
|
||||
out = append(out, w)
|
||||
}
|
||||
}
|
||||
return out
|
||||
}
|
||||
|
||||
// sentEvent is one event a fake sink was handed.
|
||||
type sentEvent struct {
|
||||
Name string
|
||||
Body any
|
||||
}
|
||||
|
||||
type eventLog struct {
|
||||
mu sync.Mutex
|
||||
sent []sentEvent
|
||||
}
|
||||
|
||||
func (e *eventLog) Emit(name string, body any, _ ...stdio.EmitOptions) error {
|
||||
e.mu.Lock()
|
||||
defer e.mu.Unlock()
|
||||
e.sent = append(e.sent, sentEvent{name, body})
|
||||
return nil
|
||||
}
|
||||
|
||||
func (e *eventLog) names() []string {
|
||||
e.mu.Lock()
|
||||
defer e.mu.Unlock()
|
||||
var out []string
|
||||
for _, s := range e.sent {
|
||||
out = append(out, s.Name)
|
||||
}
|
||||
return out
|
||||
}
|
||||
@@ -0,0 +1,203 @@
|
||||
package main
|
||||
|
||||
// The keeper's bus shapes carry only their named fields (novox/hq ADR 0507 §1, AC7; ADR 0508): every key's value
|
||||
// and every event's body, filled whole, marshals to exactly the fields named here, and no shape holds a map or a
|
||||
// free-form value through which anything else could reach the bus.
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"reflect"
|
||||
"regexp"
|
||||
"sort"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
)
|
||||
|
||||
// busFields is every shape's JSON fields, by Go type: the interface the keeper and the verbs share.
|
||||
var busFields = map[string][]string{
|
||||
"InstalledValue": {"at", "packages"},
|
||||
"InstalledPackage": {"name", "version", "explicit", "repo"},
|
||||
"NotCurrent": {"bytes", "since"},
|
||||
"UpdatesValue": {"reachable", "packages"},
|
||||
"VersionChange": {"name", "from", "to"},
|
||||
"PackageVersion": {"name", "version"},
|
||||
"Run": {"node", "run", "kind", "from", "round", "ask", "of", "steps"},
|
||||
"Step": {"step", "at", "why", "held", "unit", "ignored", "plan", "exit", "error", "changes", "check"},
|
||||
"HeldKernel": {"package", "version", "module", "bound"},
|
||||
"Changes": {"added", "removed", "changed"},
|
||||
"Check": {"passed", "pacnew", "pacsave", "news", "news_error", "mapped", "mapped_error", "newly_failed", "restart_needed", "running_kernel",
|
||||
"installed_kernel", "unbuilt", "dkms_error"},
|
||||
"NewsTitle": {"title", "link", "published"},
|
||||
"Mapped": {"unit", "container", "executable"},
|
||||
"DKMSBuild": {"module", "version", "kernel"},
|
||||
"RollbackItem": {"name", "from", "to", "source", "location"},
|
||||
"Record": {"node", "run", "at", "installed", "kernel", "failed_units", "dkms"},
|
||||
"DKMSModule": {"module", "version", "kernels"},
|
||||
"Round": {"round", "opened", "opened_by", "order"},
|
||||
"RoundNode": {"round", "node", "run", "pass"},
|
||||
"Pass": {"reason", "ask", "asked", "answer", "answered"},
|
||||
"Ask": {"id", "node", "act", "at", "of", "remove_added", "round", "pass", "reason"},
|
||||
"PackageChanged": {"node", "at", "added", "removed", "changed"},
|
||||
"UpgradeRecorded": {"node", "run", "packages", "kernel"},
|
||||
"KernelHeld": {"node", "run", "held"},
|
||||
"UpgradeStarted": {"node", "run", "from", "round", "unit", "ignored"},
|
||||
"UpgradeFinished": {"node", "run", "changes"},
|
||||
"UpgradeFailed": {"node", "run", "step", "exit", "error"},
|
||||
"UpgradeChecked": {"node", "run", "kind", "passed"},
|
||||
"RollbackStarted": {"node", "run", "of", "unit"},
|
||||
"RollbackFinished": {"node", "run", "of", "exit", "error"},
|
||||
}
|
||||
|
||||
// The value of each state's keys.
|
||||
var stateValues = []any{InstalledValue{}, NotCurrent{}, UpdatesValue{}, Run{}, Record{}, Round{}, RoundNode{}, Ask{}}
|
||||
|
||||
var stamp = time.Date(2026, 10, 24, 4, 0, 12, 0, time.UTC)
|
||||
|
||||
// fill sets every field of v, recursively, to a value that is not empty: one element in each list, every pointer
|
||||
// set, so that every field marshals, omitempty or not.
|
||||
func fill(v reflect.Value) {
|
||||
switch v.Kind() {
|
||||
case reflect.String:
|
||||
v.SetString("x")
|
||||
case reflect.Bool:
|
||||
v.SetBool(true)
|
||||
case reflect.Int, reflect.Int64:
|
||||
v.SetInt(1)
|
||||
case reflect.Pointer:
|
||||
v.Set(reflect.New(v.Type().Elem()))
|
||||
fill(v.Elem())
|
||||
case reflect.Slice:
|
||||
s := reflect.MakeSlice(v.Type(), 1, 1)
|
||||
fill(s.Index(0))
|
||||
v.Set(s)
|
||||
case reflect.Struct:
|
||||
if v.Type() == reflect.TypeOf(time.Time{}) {
|
||||
v.Set(reflect.ValueOf(stamp))
|
||||
return
|
||||
}
|
||||
for i := 0; i < v.NumField(); i++ {
|
||||
fill(v.Field(i))
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// fieldsOf walks a filled value against its marshalled JSON: each struct's object holds exactly its named
|
||||
// fields. It answers the struct types it met.
|
||||
func fieldsOf(t *testing.T, v reflect.Value, raw json.RawMessage, met map[string]bool) {
|
||||
t.Helper()
|
||||
switch v.Kind() {
|
||||
case reflect.Map, reflect.Interface:
|
||||
t.Errorf("%s holds a free-form value: a bus shape names its fields", v.Type())
|
||||
case reflect.Pointer:
|
||||
fieldsOf(t, v.Elem(), raw, met)
|
||||
case reflect.Slice:
|
||||
var items []json.RawMessage
|
||||
if err := json.Unmarshal(raw, &items); err != nil || len(items) != v.Len() {
|
||||
t.Errorf("%s: %s is not its list", v.Type(), raw)
|
||||
return
|
||||
}
|
||||
for i := range items {
|
||||
fieldsOf(t, v.Index(i), items[i], met)
|
||||
}
|
||||
case reflect.Struct:
|
||||
if v.Type() == reflect.TypeOf(time.Time{}) {
|
||||
return
|
||||
}
|
||||
name := v.Type().Name()
|
||||
met[name] = true
|
||||
var obj map[string]json.RawMessage
|
||||
if err := json.Unmarshal(raw, &obj); err != nil {
|
||||
t.Errorf("%s: %s is not an object", name, raw)
|
||||
return
|
||||
}
|
||||
var got []string
|
||||
for k := range obj {
|
||||
got = append(got, k)
|
||||
}
|
||||
want, named := busFields[name]
|
||||
if !named {
|
||||
t.Errorf("%s is on the bus and its fields are not named in busFields", name)
|
||||
return
|
||||
}
|
||||
want = append([]string(nil), want...)
|
||||
sort.Strings(got)
|
||||
sort.Strings(want)
|
||||
if strings.Join(got, " ") != strings.Join(want, " ") {
|
||||
t.Errorf("%s marshals to %v, its named fields are %v", name, got, want)
|
||||
}
|
||||
for i := 0; i < v.NumField(); i++ {
|
||||
tag, _, _ := strings.Cut(v.Type().Field(i).Tag.Get("json"), ",")
|
||||
if r, ok := obj[tag]; ok {
|
||||
fieldsOf(t, v.Field(i), r, met)
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestEveryStateValueAndEventBodyCarriesOnlyItsNamedFields(t *testing.T) {
|
||||
roots := append([]any(nil), stateValues...)
|
||||
for _, e := range Events {
|
||||
roots = append(roots, e.Body)
|
||||
}
|
||||
met := map[string]bool{}
|
||||
for _, root := range roots {
|
||||
v := reflect.New(reflect.TypeOf(root)).Elem()
|
||||
fill(v)
|
||||
raw, err := json.Marshal(v.Interface())
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
fieldsOf(t, v, raw, met)
|
||||
}
|
||||
for name := range busFields {
|
||||
if !met[name] {
|
||||
t.Errorf("busFields names %s, which no state value or event body holds", name)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestEveryEventIsNamedOnceWithItsBody(t *testing.T) {
|
||||
seen := map[string]bool{}
|
||||
for _, e := range Events {
|
||||
if seen[e.Name] {
|
||||
t.Errorf("%s twice", e.Name)
|
||||
}
|
||||
seen[e.Name] = true
|
||||
if e.Body == nil || reflect.TypeOf(e.Body).Kind() != reflect.Struct {
|
||||
t.Errorf("%s has no body type", e.Name)
|
||||
}
|
||||
}
|
||||
if len(Events) != 9 {
|
||||
t.Errorf("package.changed and the eight upgrade and rollback events, not %d", len(Events))
|
||||
}
|
||||
}
|
||||
|
||||
// A key is node and id tokens joined by dots; each token is one NATS key token.
|
||||
func TestKeysAreTokensJoinedByDots(t *testing.T) {
|
||||
token := regexp.MustCompile(`^[A-Za-z0-9_-]+$`)
|
||||
run := RunID(time.Date(2026, 10, 24, 6, 0, 12, 0, time.FixedZone("CEST", 2*3600)))
|
||||
if run != "20261024T040012Z" {
|
||||
t.Errorf("RunID: %s", run)
|
||||
}
|
||||
round := RoundID(time.Date(2026, 10, 24, 4, 0, 0, 0, time.Local))
|
||||
if round != "2026-W43" {
|
||||
t.Errorf("RoundID: %s", round)
|
||||
}
|
||||
if RoundID(time.Date(2027, 1, 2, 4, 0, 0, 0, time.UTC)) != "2026-W53" {
|
||||
t.Errorf("the ISO week of 2 January 2027 is 2026's 53rd: %s", RoundID(time.Date(2027, 1, 2, 4, 0, 0, 0, time.UTC)))
|
||||
}
|
||||
for key, parts := range map[string]int{
|
||||
NodeKey("node-a", run): 2, NotCurrentKey("node-a"): 2, RoundNodeKey(round, "node-b"): 2,
|
||||
} {
|
||||
ts := strings.Split(key, ".")
|
||||
if len(ts) != parts {
|
||||
t.Errorf("%s: %d tokens", key, len(ts))
|
||||
}
|
||||
for _, x := range ts {
|
||||
if !token.MatchString(x) {
|
||||
t.Errorf("%s: token %q", key, x)
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,166 @@
|
||||
package main
|
||||
|
||||
// The DKMS gate (novox/hq ADR 0508 §3): before the transaction, every kernel package the upgrade would change is
|
||||
// compared with the kernel bound each installed DKMS module declares itself: `Linux-Maximum` in its source's
|
||||
// META, or `BUILD_EXCLUSIVE_KERNEL` (a pattern of kernel releases) in its dkms.conf. A kernel past a bound is
|
||||
// left out of the run with its headers, and the rest is upgraded. A kernel package is known by its files, not
|
||||
// its name: one whose files hold usr/lib/modules/<release>/vmlinuz.
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"regexp"
|
||||
"strconv"
|
||||
"strings"
|
||||
)
|
||||
|
||||
var (
|
||||
vmlinuzPath = regexp.MustCompile(`^/?usr/lib/modules/([^/]+)/vmlinuz$`)
|
||||
linuxMaximum = regexp.MustCompile(`(?m)^Linux-Maximum:\s*(\S+)\s*$`)
|
||||
buildExclusive = regexp.MustCompile(`(?m)^BUILD_EXCLUSIVE_KERNEL=(?:"([^"]*)"|'([^']*)'|(\S+))\s*$`)
|
||||
leadingNumber = regexp.MustCompile(`^[0-9]+(\.[0-9]+)*`)
|
||||
)
|
||||
|
||||
// kernelRelease is the release an installed kernel package's image is for, and whether the package is a kernel.
|
||||
func (m *Machine) kernelRelease(pkg string) (string, bool, error) {
|
||||
out, err := m.Out("pacman", "-Qlq", "--", pkg)
|
||||
if err != nil {
|
||||
return "", false, fmt.Errorf("the files of %s: %w", pkg, err)
|
||||
}
|
||||
for _, l := range lines(out) {
|
||||
if g := vmlinuzPath.FindStringSubmatch(strings.TrimSpace(l)); g != nil {
|
||||
return g[1], true, nil
|
||||
}
|
||||
}
|
||||
return "", false, nil
|
||||
}
|
||||
|
||||
// NewRelease is the release a kernel package's new version will install, derived from it as the installed
|
||||
// release is from the installed version: `7.2.8.arch1-2` installs `7.2.8-arch1-2`, `6.18.54-2` (the LTS
|
||||
// kernel) installs `6.18.54-2-lts`.
|
||||
func NewRelease(installedVersion, installedRelease, newVersion string) string {
|
||||
base := func(v string) string { return strings.Replace(v, ".arch", "-arch", 1) }
|
||||
suffix := ""
|
||||
if b := base(installedVersion); strings.HasPrefix(installedRelease, b) {
|
||||
suffix = strings.TrimPrefix(installedRelease, b)
|
||||
}
|
||||
return base(newVersion) + suffix
|
||||
}
|
||||
|
||||
// pastMaximum says whether a release is past a Linux-Maximum bound, compared on as many numbers as the bound
|
||||
// gives: 7.3.1 is past 7.2, 7.2.9 is not.
|
||||
func pastMaximum(release, maximum string) bool {
|
||||
bound := strings.Split(leadingNumber.FindString(maximum), ".")
|
||||
have := strings.Split(leadingNumber.FindString(release), ".")
|
||||
if bound[0] == "" || have[0] == "" {
|
||||
return false
|
||||
}
|
||||
for i, b := range bound {
|
||||
bn, _ := strconv.Atoi(b)
|
||||
hn := 0
|
||||
if i < len(have) {
|
||||
hn, _ = strconv.Atoi(have[i])
|
||||
}
|
||||
if hn != bn {
|
||||
return hn > bn
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
// dkmsBound is the bound one DKMS module declares, read from its source.
|
||||
type dkmsBound struct {
|
||||
Maximum string
|
||||
Exclusive string
|
||||
}
|
||||
|
||||
func (m *Machine) dkmsBound(mod DKMSModule) dkmsBound {
|
||||
dir := "/usr/src/" + mod.Module + "-" + mod.Version + "/"
|
||||
var b dkmsBound
|
||||
if meta, err := m.ReadFile(dir + "META"); err == nil {
|
||||
if g := linuxMaximum.FindSubmatch(meta); g != nil {
|
||||
b.Maximum = string(g[1])
|
||||
}
|
||||
}
|
||||
if conf, err := m.ReadFile(dir + "dkms.conf"); err == nil {
|
||||
if g := buildExclusive.FindSubmatch(conf); g != nil {
|
||||
b.Exclusive = string(g[1]) + string(g[2]) + string(g[3])
|
||||
}
|
||||
}
|
||||
return b
|
||||
}
|
||||
|
||||
// heldBy says which bound of a module holds a release, or nothing.
|
||||
func heldBy(b dkmsBound, release string) (string, error) {
|
||||
if b.Maximum != "" && pastMaximum(release, b.Maximum) {
|
||||
return "Linux-Maximum: " + b.Maximum, nil
|
||||
}
|
||||
if b.Exclusive != "" {
|
||||
re, err := regexp.Compile(b.Exclusive)
|
||||
if err != nil {
|
||||
return "", fmt.Errorf("BUILD_EXCLUSIVE_KERNEL %q is not a pattern this reads: %w", b.Exclusive, err)
|
||||
}
|
||||
if !re.MatchString(release) {
|
||||
return "BUILD_EXCLUSIVE_KERNEL=" + b.Exclusive, nil
|
||||
}
|
||||
}
|
||||
return "", nil
|
||||
}
|
||||
|
||||
// Gate is the DKMS gate for the node's next run: the kernels it holds, the packages left out of the transaction
|
||||
// for them (each held kernel and its headers), or why the run cannot start.
|
||||
func (u *Upgrades) Gate() (held []HeldKernel, ignore []string, why string) {
|
||||
e, err := u.Updates.Get(u.Node)
|
||||
if err != nil {
|
||||
return nil, nil, "what an upgrade would change could not be read from the bus: " + err.Error()
|
||||
}
|
||||
if e == nil {
|
||||
return nil, nil, "what an upgrade would change is not on the bus yet, so no kernel it would bring can be checked"
|
||||
}
|
||||
var updates UpdatesValue
|
||||
if err := json.Unmarshal(e.Value, &updates); err != nil {
|
||||
return nil, nil, "what an upgrade would change is not readable: " + err.Error()
|
||||
}
|
||||
if !updates.Reachable {
|
||||
return nil, nil, "the node could not reach its mirrors, so what an upgrade would change is not known"
|
||||
}
|
||||
var modules []DKMSModule
|
||||
var bounds []dkmsBound
|
||||
held, ignore = []HeldKernel{}, []string{}
|
||||
for _, p := range updates.Packages {
|
||||
if p.Name != "linux" && !strings.HasPrefix(p.Name, "linux-") {
|
||||
continue
|
||||
}
|
||||
release, kernel, err := u.Machine.kernelRelease(p.Name)
|
||||
if err != nil {
|
||||
return nil, nil, "whether " + p.Name + " is a kernel could not be read: " + err.Error()
|
||||
}
|
||||
if !kernel {
|
||||
continue
|
||||
}
|
||||
if modules == nil {
|
||||
if modules, err = u.Machine.DKMSModules(); err != nil {
|
||||
return nil, nil, "the DKMS modules could not be read: " + err.Error()
|
||||
}
|
||||
for _, mod := range modules {
|
||||
b := u.Machine.dkmsBound(mod)
|
||||
bounds = append(bounds, b)
|
||||
}
|
||||
}
|
||||
next := NewRelease(p.From, release, p.To)
|
||||
for i, mod := range modules {
|
||||
bound, err := heldBy(bounds[i], next)
|
||||
if err != nil {
|
||||
return nil, nil, mod.Module + " " + mod.Version + ": " + err.Error()
|
||||
}
|
||||
if bound == "" {
|
||||
continue
|
||||
}
|
||||
held = append(held, HeldKernel{Package: p.Name, Version: p.To, Module: mod.Module + "/" + mod.Version, Bound: bound})
|
||||
if !contains(ignore, p.Name) {
|
||||
ignore = append(ignore, p.Name, p.Name+"-headers")
|
||||
}
|
||||
}
|
||||
}
|
||||
return held, ignore, ""
|
||||
}
|
||||
@@ -0,0 +1,115 @@
|
||||
package main
|
||||
|
||||
// The DKMS gate (novox/hq ADR 0508 §3; acceptance 1841/AC4). The META is captured from a machine whose ZFS
|
||||
// module declares Linux-Maximum 7.2; the BUILD_EXCLUSIVE_KERNEL case is a made-up module's pattern.
|
||||
|
||||
import (
|
||||
"os"
|
||||
"strings"
|
||||
"testing"
|
||||
)
|
||||
|
||||
func TestTheNewKernelsReleaseIsDerivedAsTheInstalledOnesIs(t *testing.T) {
|
||||
for _, c := range [][4]string{
|
||||
{"7.2.8.arch1-2", "7.2.8-arch1-2", "7.3.1.arch1-1", "7.3.1-arch1-1"},
|
||||
{"6.18.54-2", "6.18.54-2-lts", "6.18.60-1", "6.18.60-1-lts"},
|
||||
} {
|
||||
if got := NewRelease(c[0], c[1], c[2]); got != c[3] {
|
||||
t.Errorf("%v: %s", c, got)
|
||||
}
|
||||
}
|
||||
for release, past := range map[string]bool{"7.3.1-arch1-1": true, "7.2.9-arch1-1": false, "7.2-arch1-1": false, "8.0.1": true, "6.18.60-1-lts": false} {
|
||||
if pastMaximum(release, "7.2") != past {
|
||||
t.Errorf("%s past 7.2: want %v", release, past)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// gateMachine is a machine with the linux and LTS kernels and these DKMS modules' sources.
|
||||
func gateMachine(t *testing.T, dkmsStatus string, sources map[string]string) *Machine {
|
||||
files := func(paths ...string) Ran { return Ran{Stdout: strings.Join(paths, "\n") + "\n"} }
|
||||
m := machine(byLine(map[string]Ran{
|
||||
"pacman -Qlq -- linux": files("/usr/lib/modules/7.2.8-arch1-2/", "/usr/lib/modules/7.2.8-arch1-2/vmlinuz"),
|
||||
"pacman -Qlq -- linux-headers": files("/usr/lib/modules/7.2.8-arch1-2/build/Makefile"),
|
||||
"pacman -Qlq -- linux-lts": files("/usr/lib/modules/6.18.54-2-lts/vmlinuz"),
|
||||
"pacman -Qlq -- linux-firmware": files("/usr/lib/firmware/amdgpu/"),
|
||||
"dkms status": {Stdout: dkmsStatus},
|
||||
}, nil), 1000)
|
||||
m.ReadFile = func(path string) ([]byte, error) {
|
||||
if s, ok := sources[path]; ok {
|
||||
return []byte(s), nil
|
||||
}
|
||||
return nil, errNoFile
|
||||
}
|
||||
return m
|
||||
}
|
||||
|
||||
func capturedZFSSource(t *testing.T) map[string]string {
|
||||
out := map[string]string{}
|
||||
for _, f := range []string{"META", "dkms.conf"} {
|
||||
raw, err := os.ReadFile("testdata/dkms-zfs-linux-maximum-7.2/" + f)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
out["/usr/src/zfs-2.4.4/"+f] = string(raw)
|
||||
}
|
||||
return out
|
||||
}
|
||||
|
||||
func gateUpgrades(t *testing.T, m *Machine, linuxTo string, reachable bool) *Upgrades {
|
||||
updates := newBucket()
|
||||
if _, err := updates.Put("node-a", UpdatesValue{Reachable: reachable, Packages: []VersionChange{
|
||||
{Name: "bash", From: "5.3.15-1", To: "5.3.20-1"},
|
||||
{Name: "linux", From: "7.2.8.arch1-2", To: linuxTo},
|
||||
{Name: "linux-firmware", From: "20261001-1", To: "20261015-1"},
|
||||
{Name: "linux-headers", From: "7.2.8.arch1-2", To: linuxTo},
|
||||
{Name: "linux-lts", From: "6.18.54-2", To: "6.18.60-1"},
|
||||
}}); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
return &Upgrades{Node: "node-a", Machine: m, Updates: updates}
|
||||
}
|
||||
|
||||
const zfsStatus = "zfs/2.4.4, 7.2.8-arch1-2, x86_64: installed\nzfs/2.4.4, 6.18.54-2-lts, x86_64: installed\n"
|
||||
|
||||
func TestAKernelPastAModulesLinuxMaximumIsHeldWithItsHeaders(t *testing.T) {
|
||||
u := gateUpgrades(t, gateMachine(t, zfsStatus, capturedZFSSource(t)), "7.3.1.arch1-1", true)
|
||||
held, ignore, why := u.Gate()
|
||||
if why != "" {
|
||||
t.Fatal(why)
|
||||
}
|
||||
if len(held) != 1 || held[0] != (HeldKernel{Package: "linux", Version: "7.3.1.arch1-1", Module: "zfs/2.4.4", Bound: "Linux-Maximum: 7.2"}) {
|
||||
t.Errorf("held: %+v", held)
|
||||
}
|
||||
if strings.Join(ignore, " ") != "linux linux-headers" {
|
||||
t.Errorf("left out: %v (the LTS kernel stays within the bound and is upgraded)", ignore)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAKernelWithinTheBoundIsNotHeld(t *testing.T) {
|
||||
u := gateUpgrades(t, gateMachine(t, zfsStatus, capturedZFSSource(t)), "7.2.9.arch1-1", true)
|
||||
if held, ignore, why := u.Gate(); why != "" || len(held) != 0 || len(ignore) != 0 {
|
||||
t.Errorf("%+v %v %q", held, ignore, why)
|
||||
}
|
||||
}
|
||||
|
||||
func TestABuildExclusivePatternHoldsTheKernelsItDoesNotMatch(t *testing.T) {
|
||||
src := map[string]string{"/usr/src/made-up-driver-1.0/dkms.conf": "PACKAGE_NAME=\"made-up-driver\"\nBUILD_EXCLUSIVE_KERNEL=\"^(6\\.|7\\.[0-2]\\.)\"\n"}
|
||||
u := gateUpgrades(t, gateMachine(t, "made-up-driver/1.0, 7.2.8-arch1-2, x86_64: installed\n", src), "7.3.1.arch1-1", true)
|
||||
held, ignore, why := u.Gate()
|
||||
if why != "" || len(held) != 1 || held[0].Package != "linux" || held[0].Bound != `BUILD_EXCLUSIVE_KERNEL=^(6\.|7\.[0-2]\.)` ||
|
||||
strings.Join(ignore, " ") != "linux linux-headers" {
|
||||
t.Errorf("%+v %v %q", held, ignore, why)
|
||||
}
|
||||
}
|
||||
|
||||
func TestNoRunWhenWhatTheUpgradeWouldChangeIsNotKnown(t *testing.T) {
|
||||
u := gateUpgrades(t, gateMachine(t, zfsStatus, capturedZFSSource(t)), "7.3.1.arch1-1", false)
|
||||
if _, _, why := u.Gate(); !strings.Contains(why, "could not reach its mirrors") {
|
||||
t.Errorf("unreachable: %q", why)
|
||||
}
|
||||
u.Updates = newBucket()
|
||||
if _, _, why := u.Gate(); !strings.Contains(why, "not on the bus") {
|
||||
t.Errorf("no updates key: %q", why)
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,32 @@
|
||||
package main
|
||||
|
||||
// A node whose packages are over the bus's cap is unhealthy, naming the size (novox/hq ADR 0507 §2, acceptance
|
||||
// 1840/AC3): `installed.<node>` keeps its last value, which is no longer the node's, and the module's health says
|
||||
// so until the keeper writes it again and deletes the mark.
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
)
|
||||
|
||||
// InstalledHealth is the health of this node's `installed` key: unhealthy while `<node>.not-current` is there,
|
||||
// with its size and since when; ok false when there is nothing to say (the key is current, or the state cannot
|
||||
// be read here, which the keeper's own log says).
|
||||
func (m *Machine) InstalledHealth() (Health, bool) {
|
||||
if m.States == nil || m.Node == "" {
|
||||
return Health{}, false
|
||||
}
|
||||
e, err := m.States(StateInstalled).Get(NotCurrentKey(m.Node))
|
||||
if err != nil {
|
||||
return Health{Why: "the node's packages on the bus could not be read: " + err.Error()}, true
|
||||
}
|
||||
if e == nil {
|
||||
return Health{}, false
|
||||
}
|
||||
var nc NotCurrent
|
||||
if err := json.Unmarshal(e.Value, &nc); err != nil {
|
||||
return Health{Why: fmt.Sprintf("installed.%s is marked not current, and its mark is not readable: %v", m.Node, err)}, true
|
||||
}
|
||||
return Health{Why: fmt.Sprintf("the node's packages come to %d bytes, over the bus's cap of %d on a value, since %s: "+
|
||||
"installed.%s keeps an older list", nc.Bytes, ValueCap, nc.Since.Format("2006-01-02 15:04 MST"), m.Node)}, true
|
||||
}
|
||||
@@ -0,0 +1,45 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"strings"
|
||||
"testing"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
)
|
||||
|
||||
// fakeReader reads a fakeStore as a verb does.
|
||||
type fakeReader struct{ *fakeStore }
|
||||
|
||||
func (r fakeReader) History(key string) ([]stdio.StateRevision, error) {
|
||||
e, _ := r.Get(key)
|
||||
if e == nil {
|
||||
return []stdio.StateRevision{}, nil
|
||||
}
|
||||
return []stdio.StateRevision{{Key: key, Value: e.Value, Revision: e.Revision, Op: "put", At: r.at[key]}}, nil
|
||||
}
|
||||
|
||||
// readerOf is a machine's States over fake states.
|
||||
func readerOf(states *fakeStates) func(string) StateReader {
|
||||
return func(name string) StateReader { return fakeReader{states.store(name)} }
|
||||
}
|
||||
|
||||
// 1840/AC3: the health is unhealthy naming the size while the node is marked not current, and says nothing of
|
||||
// it once the mark is gone.
|
||||
func TestTheHealthNamesTheSizeWhileTheNodeIsNotCurrent(t *testing.T) {
|
||||
states := &fakeStates{}
|
||||
m := machine(nil, 1000)
|
||||
m.Node, m.States = "node-a", readerOf(states)
|
||||
if _, said := m.InstalledHealth(); said {
|
||||
t.Fatal("a current node's health said something")
|
||||
}
|
||||
states.store(StateInstalled).Put("node-a.not-current", NotCurrent{Bytes: 300000, Since: t0})
|
||||
states.store(StateInstalled).Put("node-b.not-current", NotCurrent{Bytes: 1, Since: t0})
|
||||
h, said := m.InstalledHealth()
|
||||
if !said || h.Healthy || !strings.Contains(h.Why, "300000 bytes") || !strings.Contains(h.Why, "2026-10-24 04:00") {
|
||||
t.Fatalf("%+v", h)
|
||||
}
|
||||
states.store(StateInstalled).Delete("node-a.not-current")
|
||||
if _, said := m.InstalledHealth(); said {
|
||||
t.Fatal("the mark is gone and the health still says it")
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,639 @@
|
||||
package main
|
||||
|
||||
// The keeper (novox/hq ADR 0507 §1-§4; ADR 0508): the module's process on every node, under an account of its
|
||||
// own with a bus account of its own (ADR 0436 §1). It keeps the node's packages in the state `installed` and what
|
||||
// an upgrade would change in `updates`, and says each change of the packages as `package.changed`. It reads the
|
||||
// machine at its start and whenever the machine may have changed, never on a timer but the one daily read of
|
||||
// `updates` ADR 0507 §4 allows:
|
||||
//
|
||||
// - pacman's database lock released (an inotify watch on its directory: the lock file deleted) ends a
|
||||
// transaction, whoever ran it: `installed` and then `updates` are read again;
|
||||
// - the mirror list rewritten (reflector writes /etc/pacman.d/mirrorlist): `updates` is read again;
|
||||
// - a day after `updates` was last read: it is read again, since the repositories change what an upgrade
|
||||
// would do without any transaction on the node.
|
||||
//
|
||||
// A key is written only when what it holds changed (ADR 0289 §1). A value over the bus's cap is not written:
|
||||
// the key keeps its last value and `<node>.not-current` says the size and since when, written once and deleted
|
||||
// once the node's packages are current again (ADR 0507 §2).
|
||||
//
|
||||
// Everything the keeper touches is injected (its states, its events, the machine, the clock, the watch), so the
|
||||
// loop is tested without a machine or a bus. One goroutine runs it all: the watch's signals, the clock, and every
|
||||
// change a hook watches are handled one at a time, in the order they came.
|
||||
//
|
||||
// # Hooks (the upgrade parts of ADR 0508 join here)
|
||||
//
|
||||
// Hooks are where further work joins the loop, each run on the loop's goroutine:
|
||||
//
|
||||
// - Start: once, after the first reads of `installed` and `updates`, before anything else;
|
||||
// - AfterTransaction: after every released lock, once `installed` and `updates` were read again, with what the
|
||||
// transaction changed (nil when nothing);
|
||||
// - Changes: each a state and a key pattern (`*` one name, `**` the rest) watched for the life of the keeper;
|
||||
// its Handle sees the current values first (Current set), then every put and delete in order;
|
||||
// - Wakes: each a function the keeper calls at its start and again at the time it last answered; it answers when
|
||||
// it next wants to be called (a zero time: not again). It is how a hook keeps a time (a round's opening, a
|
||||
// stall) without polling.
|
||||
//
|
||||
// A hook's error is logged and the keeper goes on: a fault in one part never stops the view. keeperHooks builds
|
||||
// the hooks the keeper runs with, once connected; it is the one place a part is added.
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"fmt"
|
||||
"os"
|
||||
"os/signal"
|
||||
"path/filepath"
|
||||
"syscall"
|
||||
"time"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
"git.novox.be/novox/mesh-sdk/go/process"
|
||||
)
|
||||
|
||||
// ValueCap is the bus's cap on one value of a state (ADR 0201 §14): the mesh's bound, not the module's.
|
||||
const ValueCap = 256 << 10
|
||||
|
||||
// Where pacman keeps its database and its lock, and where the mirror list is.
|
||||
const (
|
||||
PacmanDBDir = "/var/lib/pacman"
|
||||
LockName = "db.lck"
|
||||
MirrorDir = "/etc/pacman.d"
|
||||
MirrorName = "mirrorlist"
|
||||
)
|
||||
|
||||
// UpdatesEvery is how long after the last read of `updates` it is read again with nothing else asking (ADR 0507
|
||||
// §4: once a day).
|
||||
const UpdatesEvery = 24 * time.Hour
|
||||
|
||||
// Store is one state as the keeper uses it: a process state on the bus (process.State), or a fake in a test.
|
||||
type Store interface {
|
||||
Get(key string) (*stdio.StateEntry, error)
|
||||
Create(key string, value any) (uint64, error)
|
||||
Update(key string, value any, revision uint64) (uint64, error)
|
||||
Put(key string, value any) (uint64, error)
|
||||
Delete(key string) error
|
||||
Keys() ([]string, error)
|
||||
Watch(key string, handle func(stdio.StateChange) error) error
|
||||
}
|
||||
|
||||
// Clock is the time as the keeper sees it.
|
||||
type Clock interface {
|
||||
Now() time.Time
|
||||
// After fires once, d from now.
|
||||
After(d time.Duration) <-chan time.Time
|
||||
}
|
||||
|
||||
type realClock struct{}
|
||||
|
||||
func (realClock) Now() time.Time { return time.Now() }
|
||||
func (realClock) After(d time.Duration) <-chan time.Time { return time.After(d) }
|
||||
|
||||
// Signal is what the machine's watch saw.
|
||||
type Signal int
|
||||
|
||||
const (
|
||||
// LockReleased: pacman's database lock was deleted, a transaction (or a database sync) ended.
|
||||
LockReleased Signal = iota + 1
|
||||
// MirrorsRewritten: the mirror list was written or replaced.
|
||||
MirrorsRewritten
|
||||
)
|
||||
|
||||
func (s Signal) String() string {
|
||||
switch s {
|
||||
case LockReleased:
|
||||
return "the lock released"
|
||||
case MirrorsRewritten:
|
||||
return "the mirror list rewritten"
|
||||
}
|
||||
return fmt.Sprintf("signal %d", int(s))
|
||||
}
|
||||
|
||||
// Watcher is the machine's watch: its signals, until it is closed or fails (the channel closes).
|
||||
type Watcher interface {
|
||||
Signals() <-chan Signal
|
||||
Close() error
|
||||
}
|
||||
|
||||
// Transaction is a released lock as AfterTransaction sees it: when, and what it changed in `installed`
|
||||
// (nil when it changed nothing, or the value could not be written).
|
||||
type Transaction struct {
|
||||
At time.Time
|
||||
Changed *PackageChanged
|
||||
}
|
||||
|
||||
// StateHook is a watch on one of the module's states for the life of the keeper.
|
||||
type StateHook struct {
|
||||
State string
|
||||
Key string
|
||||
Handle func(k *Keeper, c stdio.StateChange) error
|
||||
}
|
||||
|
||||
// Wake is called at the keeper's start and then at the time it last answered; it answers when next.
|
||||
type Wake func(k *Keeper, now time.Time) time.Time
|
||||
|
||||
// Hooks are the parts that join the keeper's loop (see the file's comment).
|
||||
type Hooks struct {
|
||||
Start []func(k *Keeper) error
|
||||
AfterTransaction []func(k *Keeper, t Transaction) error
|
||||
Changes []StateHook
|
||||
Wakes []Wake
|
||||
}
|
||||
|
||||
// keeperHooks is the hooks the keeper runs with, built once the keeper is connected: the one place a part of
|
||||
// the keeper is added. Today: the upgrades of ADR 0508 (upgrades.go, record.go, transaction.go, rounds.go), for
|
||||
// this node.
|
||||
//
|
||||
// - Start: the package manager's log read into `runs` once (ImportLog), records past record-days pruned
|
||||
// (PruneRecords), and every unfinished run carried on (Follow), a run a keeper that went away left included;
|
||||
// - AfterTransaction: every unfinished run carried on (Follow): a released lock may be a run's transaction ending;
|
||||
// - Changes: `asks` at `<node>.*`, each change handed to TakeAsk; `rounds` and `runs`, every change, to the
|
||||
// round's step (RoundStep): a node starts the moment the node before it passed;
|
||||
// - Wakes: PruneRecords once a day; the round's step at the time it answered (the Saturday's opening, a pass
|
||||
// whose answer is read again, a refused run tried again).
|
||||
func keeperHooks(k *Keeper) Hooks {
|
||||
root := k.Root
|
||||
if root == nil {
|
||||
root = newRootActs()
|
||||
}
|
||||
u := &Upgrades{Node: k.Node, Config: k.Config, Machine: k.Machine, Root: root, Events: k.Events,
|
||||
Installed: k.State(StateInstalled), Updates: k.State(StateUpdates), Runs: k.State(StateRuns),
|
||||
Records: k.State(StateRecords), Rounds: k.State(StateRounds), Asks: k.State(StateAsks), AskRecord: k.AskRecord,
|
||||
Check: k.Check}
|
||||
k.upgrades = u
|
||||
const roundWake = 1 // the round's step is the second wake
|
||||
roundStep := func(k *Keeper, now time.Time) time.Time {
|
||||
next, err := u.RoundStep()
|
||||
if err != nil {
|
||||
k.logf("[pacman] keeper: the round: %v", err)
|
||||
}
|
||||
return next
|
||||
}
|
||||
// onRoundChange takes the round's step on a change of `rounds` or `runs`, and keeps the time it answers. The
|
||||
// current values handed at the watch's start are not each a step: the wake at the start takes one.
|
||||
onRoundChange := func(k *Keeper, c stdio.StateChange) error {
|
||||
if c.Current {
|
||||
return nil
|
||||
}
|
||||
k.wakeAt(roundWake, roundStep(k, k.Clock.Now()))
|
||||
return nil
|
||||
}
|
||||
return Hooks{
|
||||
Start: []func(*Keeper) error{
|
||||
func(k *Keeper) error {
|
||||
n, err := u.ImportLog()
|
||||
if n > 0 {
|
||||
k.logf("[pacman] keeper: %d past upgrades read from the package manager's log into runs", n)
|
||||
}
|
||||
return err
|
||||
},
|
||||
func(*Keeper) error { _, err := u.PruneRecords(); return err },
|
||||
func(*Keeper) error { return u.Follow() },
|
||||
},
|
||||
AfterTransaction: []func(*Keeper, Transaction) error{
|
||||
func(*Keeper, Transaction) error { return u.Follow() },
|
||||
},
|
||||
Changes: []StateHook{
|
||||
{State: StateAsks, Key: k.Node + ".*", Handle: func(_ *Keeper, c stdio.StateChange) error {
|
||||
return u.TakeAsk(c)
|
||||
}},
|
||||
{State: StateRounds, Key: "**", Handle: onRoundChange},
|
||||
{State: StateRuns, Key: "**", Handle: onRoundChange},
|
||||
},
|
||||
Wakes: []Wake{
|
||||
func(k *Keeper, now time.Time) time.Time {
|
||||
if !now.Equal(k.started) {
|
||||
if _, err := u.PruneRecords(); err != nil {
|
||||
k.logf("[pacman] keeper: pruning records: %v", err)
|
||||
}
|
||||
}
|
||||
return now.Add(UpdatesEvery)
|
||||
},
|
||||
roundStep,
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
// newRootActs is the client of the module's root helper (ADR 0510), which carries out the acts that need root,
|
||||
// at the socket the manifest names to the keeper; with none named, every act is refused, named.
|
||||
var newRootActs = func() RootActs {
|
||||
if sock := os.Getenv(EnvRootSocket); sock != "" {
|
||||
return rootClient{Socket: sock}
|
||||
}
|
||||
return noRoot{}
|
||||
}
|
||||
|
||||
type noRoot struct{}
|
||||
|
||||
const noRootSaid = "the keeper was started without " + EnvRootSocket + ", the root helper's socket"
|
||||
|
||||
func (noRoot) Transaction(kind, run string, ignore []string) error {
|
||||
return errors.New(noRootSaid + ": no transaction can start")
|
||||
}
|
||||
|
||||
func (noRoot) Journal(unit string, n int) ([]string, error) {
|
||||
return nil, errors.New(noRootSaid + ": no journal can be read")
|
||||
}
|
||||
|
||||
func (noRoot) Rollback(run string, plan []RollbackItem) error {
|
||||
return errors.New(noRootSaid + ": no rollback can start")
|
||||
}
|
||||
|
||||
func (noRoot) Maps() ([]Mapped, error) {
|
||||
return nil, errors.New(noRootSaid + ": no process's maps can be read")
|
||||
}
|
||||
|
||||
// Keeper is the module's process on one node.
|
||||
type Keeper struct {
|
||||
Node string
|
||||
Config KeeperConfig
|
||||
Machine *Machine
|
||||
State func(name string) Store
|
||||
// Events emits the keeper's events; package.changed carries the id `installed.<node>.<revision>`, the bus's
|
||||
// dedup id, so the same change said twice is dropped.
|
||||
Events EventSink
|
||||
Clock Clock
|
||||
Watcher Watcher
|
||||
// Locked says pacman's lock is held now: a released lock followed at once by another transaction is read
|
||||
// at that one's end.
|
||||
Locked func() bool
|
||||
// DBChanged is when the local package database last changed, the time of a value read at the start.
|
||||
DBChanged func() time.Time
|
||||
Logf func(format string, a ...any)
|
||||
// Hooks are the parts joining the loop; MakeHooks, when set, builds them once the keeper is connected
|
||||
// (its State and Events set), replacing Hooks.
|
||||
Hooks Hooks
|
||||
MakeHooks func(k *Keeper) Hooks
|
||||
// AskRecord reads the operator channel's record of one of the module's asks (process.Conn.Ask).
|
||||
AskRecord func(id string) (*stdio.StateEntry, error)
|
||||
// Root carries out the acts that need root; nil is the root helper's client (newRootActs). Check is the
|
||||
// after-upgrade check; nil is the machine's (AfterCheck). A test sets both.
|
||||
Root RootActs
|
||||
Check func(run Run, record *Record) Check
|
||||
|
||||
upgrades *Upgrades
|
||||
|
||||
started time.Time
|
||||
|
||||
nextUpdates time.Time
|
||||
wakes []time.Time
|
||||
}
|
||||
|
||||
func (k *Keeper) logf(format string, a ...any) {
|
||||
if k.Logf != nil {
|
||||
k.Logf(format, a...)
|
||||
}
|
||||
}
|
||||
|
||||
type change struct {
|
||||
hook StateHook
|
||||
c stdio.StateChange
|
||||
}
|
||||
|
||||
// Run reads the machine into the bus and then follows it until ctx ends or the watch fails.
|
||||
func (k *Keeper) Run(ctx context.Context) error {
|
||||
if k.MakeHooks != nil {
|
||||
k.Hooks = k.MakeHooks(k)
|
||||
}
|
||||
k.started = k.Clock.Now()
|
||||
at := k.started
|
||||
if k.DBChanged != nil {
|
||||
if t := k.DBChanged(); !t.IsZero() {
|
||||
at = t
|
||||
}
|
||||
}
|
||||
k.refreshInstalled(at)
|
||||
k.refreshUpdates()
|
||||
for _, h := range k.Hooks.Start {
|
||||
if err := h(k); err != nil {
|
||||
k.logf("[pacman] keeper: a start hook: %v", err)
|
||||
}
|
||||
}
|
||||
changes := make(chan change, 256)
|
||||
for _, h := range k.Hooks.Changes {
|
||||
h := h
|
||||
err := k.State(h.State).Watch(h.Key, func(c stdio.StateChange) error {
|
||||
if c.Current {
|
||||
// the current values are handed over inside Watch, on this goroutine.
|
||||
k.handle(h, c)
|
||||
return nil
|
||||
}
|
||||
select {
|
||||
case changes <- change{h, c}:
|
||||
return nil
|
||||
case <-ctx.Done():
|
||||
return ctx.Err()
|
||||
}
|
||||
})
|
||||
if err != nil {
|
||||
return fmt.Errorf("watching %s %q: %w", h.State, h.Key, err)
|
||||
}
|
||||
}
|
||||
now := k.started
|
||||
k.wakes = make([]time.Time, len(k.Hooks.Wakes))
|
||||
for i, w := range k.Hooks.Wakes {
|
||||
k.wakes[i] = w(k, now)
|
||||
}
|
||||
signals := k.Watcher.Signals()
|
||||
for {
|
||||
now = k.Clock.Now()
|
||||
next := k.nextUpdates
|
||||
for _, w := range k.wakes {
|
||||
if !w.IsZero() && w.Before(next) {
|
||||
next = w
|
||||
}
|
||||
}
|
||||
timer := k.Clock.After(next.Sub(now))
|
||||
select {
|
||||
case <-ctx.Done():
|
||||
return ctx.Err()
|
||||
case s, open := <-signals:
|
||||
if !open {
|
||||
return errors.New("the machine's watch ended")
|
||||
}
|
||||
k.signal(s)
|
||||
case c := <-changes:
|
||||
k.handle(c.hook, c.c)
|
||||
case <-timer:
|
||||
k.due()
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// wakeAt sets when wake i is called next, from a change a hook handled; before the wakes are first called
|
||||
// (the current values handed at a watch's start) there is nothing to set.
|
||||
func (k *Keeper) wakeAt(i int, at time.Time) {
|
||||
if i < len(k.wakes) {
|
||||
k.wakes[i] = at
|
||||
}
|
||||
}
|
||||
|
||||
func (k *Keeper) handle(h StateHook, c stdio.StateChange) {
|
||||
if err := h.Handle(k, c); err != nil {
|
||||
k.logf("[pacman] keeper: handling %s %s: %v", h.State, c.Key, err)
|
||||
}
|
||||
}
|
||||
|
||||
// signal acts on what the machine's watch saw.
|
||||
func (k *Keeper) signal(s Signal) {
|
||||
switch s {
|
||||
case LockReleased:
|
||||
if k.Locked != nil && k.Locked() {
|
||||
return
|
||||
}
|
||||
at := k.Clock.Now()
|
||||
changed := k.refreshInstalled(at)
|
||||
k.refreshUpdates()
|
||||
for _, h := range k.Hooks.AfterTransaction {
|
||||
if err := h(k, Transaction{At: at, Changed: changed}); err != nil {
|
||||
k.logf("[pacman] keeper: after a transaction: %v", err)
|
||||
}
|
||||
}
|
||||
case MirrorsRewritten:
|
||||
k.refreshUpdates()
|
||||
}
|
||||
}
|
||||
|
||||
// due runs what the clock has reached: the daily read of `updates`, and every wake whose time came.
|
||||
func (k *Keeper) due() {
|
||||
now := k.Clock.Now()
|
||||
if !now.Before(k.nextUpdates) {
|
||||
k.refreshUpdates()
|
||||
}
|
||||
for i, w := range k.wakes {
|
||||
if !w.IsZero() && !now.Before(w) {
|
||||
k.wakes[i] = k.Hooks.Wakes[i](k, now)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// refreshInstalled reads the node's packages and writes `installed.<node>` when they differ from the key,
|
||||
// emitting `package.changed` with what changed. It answers that change, nil when nothing was written.
|
||||
func (k *Keeper) refreshInstalled(at time.Time) *PackageChanged {
|
||||
pkgs, skipped, err := k.Machine.ReadInstalled()
|
||||
if err != nil {
|
||||
k.logf("[pacman] keeper: reading the installed packages: %v", err)
|
||||
return nil
|
||||
}
|
||||
if skipped > 0 {
|
||||
k.logf("[pacman] keeper: %d lines of pacman's list were not a package and were left out", skipped)
|
||||
}
|
||||
st := k.State(StateInstalled)
|
||||
var before InstalledValue
|
||||
entry, err := st.Get(k.Node)
|
||||
if err != nil {
|
||||
k.logf("[pacman] keeper: reading installed.%s: %v", k.Node, err)
|
||||
return nil
|
||||
}
|
||||
if entry != nil {
|
||||
if err := json.Unmarshal(entry.Value, &before); err != nil {
|
||||
k.logf("[pacman] keeper: installed.%s holds no package list, so it is written whole: %v", k.Node, err)
|
||||
entry = nil
|
||||
}
|
||||
}
|
||||
if entry != nil && samePackages(before.Packages, pkgs) {
|
||||
k.current()
|
||||
return nil
|
||||
}
|
||||
value := InstalledValue{At: at.UTC(), Packages: pkgs}
|
||||
data, err := json.Marshal(value)
|
||||
if err != nil {
|
||||
k.logf("[pacman] keeper: %v", err)
|
||||
return nil
|
||||
}
|
||||
if len(data) > ValueCap {
|
||||
k.notCurrent(len(data), at)
|
||||
return nil
|
||||
}
|
||||
rev, err := st.Put(k.Node, value)
|
||||
if err != nil {
|
||||
k.logf("[pacman] keeper: writing installed.%s: %v", k.Node, err)
|
||||
return nil
|
||||
}
|
||||
k.current()
|
||||
c := DiffInstalled(before.Packages, pkgs)
|
||||
if c.Empty() {
|
||||
return nil
|
||||
}
|
||||
ev := &PackageChanged{Node: k.Node, At: at.UTC(), Added: c.Added, Removed: c.Removed, Changed: c.Changed}
|
||||
id := fmt.Sprintf("%s.%s.%d", StateInstalled, k.Node, rev)
|
||||
if err := k.Events.Emit(EventPackageChanged, *ev, stdio.EmitOptions{MessageID: id}); err != nil {
|
||||
k.logf("[pacman] keeper: emitting %s for installed.%s at revision %d: %v", EventPackageChanged, k.Node, rev, err)
|
||||
}
|
||||
return ev
|
||||
}
|
||||
|
||||
func samePackages(a, b []InstalledPackage) bool {
|
||||
if len(a) != len(b) {
|
||||
return false
|
||||
}
|
||||
for i := range a {
|
||||
if a[i] != b[i] {
|
||||
return false
|
||||
}
|
||||
}
|
||||
return true
|
||||
}
|
||||
|
||||
// notCurrent marks the node's packages as not current, once: the first size and moment stay until it is current
|
||||
// again.
|
||||
func (k *Keeper) notCurrent(size int, at time.Time) {
|
||||
st := k.State(StateInstalled)
|
||||
key := NotCurrentKey(k.Node)
|
||||
k.logf("[pacman] keeper: the node's packages come to %d bytes, over the bus's cap of %d: installed.%s keeps its last value", size, ValueCap, k.Node)
|
||||
if e, err := st.Get(key); err != nil || e != nil {
|
||||
if err != nil {
|
||||
k.logf("[pacman] keeper: reading %s: %v", key, err)
|
||||
}
|
||||
return
|
||||
}
|
||||
if _, err := st.Put(key, NotCurrent{Bytes: size, Since: at.UTC()}); err != nil {
|
||||
k.logf("[pacman] keeper: writing %s: %v", key, err)
|
||||
}
|
||||
}
|
||||
|
||||
// current deletes the not-current mark, when there is one: the key holds the node's packages again.
|
||||
func (k *Keeper) current() {
|
||||
st := k.State(StateInstalled)
|
||||
key := NotCurrentKey(k.Node)
|
||||
e, err := st.Get(key)
|
||||
if err != nil {
|
||||
k.logf("[pacman] keeper: reading %s: %v", key, err)
|
||||
return
|
||||
}
|
||||
if e == nil {
|
||||
return
|
||||
}
|
||||
if err := st.Delete(key); err != nil {
|
||||
k.logf("[pacman] keeper: deleting %s: %v", key, err)
|
||||
}
|
||||
}
|
||||
|
||||
// refreshUpdates reads what a full upgrade would change and writes `updates.<node>` when the list or reachable
|
||||
// changed. A failed read keeps the last list and says reachable false, never an empty list in its place.
|
||||
func (k *Keeper) refreshUpdates() {
|
||||
k.nextUpdates = k.Clock.Now().Add(UpdatesEvery)
|
||||
list, err := k.Machine.UpdateList()
|
||||
st := k.State(StateUpdates)
|
||||
entry, gerr := st.Get(k.Node)
|
||||
if gerr != nil {
|
||||
k.logf("[pacman] keeper: reading updates.%s: %v", k.Node, gerr)
|
||||
return
|
||||
}
|
||||
var before UpdatesValue
|
||||
if entry != nil {
|
||||
if uerr := json.Unmarshal(entry.Value, &before); uerr != nil {
|
||||
entry = nil
|
||||
}
|
||||
}
|
||||
next := UpdatesValue{Reachable: true, Packages: list}
|
||||
if err != nil {
|
||||
k.logf("[pacman] keeper: reading what an upgrade would change: %v", err)
|
||||
next = UpdatesValue{Reachable: false, Packages: before.Packages}
|
||||
}
|
||||
if next.Packages == nil {
|
||||
next.Packages = []VersionChange{}
|
||||
}
|
||||
if entry != nil && sameUpdates(before, next) {
|
||||
return
|
||||
}
|
||||
if _, err := st.Put(k.Node, next); err != nil {
|
||||
k.logf("[pacman] keeper: writing updates.%s: %v", k.Node, err)
|
||||
}
|
||||
}
|
||||
|
||||
func sameUpdates(a, b UpdatesValue) bool {
|
||||
if a.Reachable != b.Reachable || len(a.Packages) != len(b.Packages) {
|
||||
return false
|
||||
}
|
||||
for i := range a.Packages {
|
||||
if a.Packages[i] != b.Packages[i] {
|
||||
return false
|
||||
}
|
||||
}
|
||||
return true
|
||||
}
|
||||
|
||||
// The keeper as a process.
|
||||
|
||||
// ReadKeeperConfig reads the keeper's config file, which the mesh writes from the module's config values.
|
||||
func ReadKeeperConfig(path string) (KeeperConfig, error) {
|
||||
var c KeeperConfig
|
||||
if path == "" {
|
||||
return c, fmt.Errorf("the keeper was started without %s, its config file", EnvConfig)
|
||||
}
|
||||
data, err := os.ReadFile(path)
|
||||
if err != nil {
|
||||
return c, fmt.Errorf("reading the keeper's config: %w", err)
|
||||
}
|
||||
if err := json.Unmarshal(data, &c); err != nil {
|
||||
return c, fmt.Errorf("the keeper's config at %s is not its JSON: %w", path, err)
|
||||
}
|
||||
return c, nil
|
||||
}
|
||||
|
||||
// keeper is `pacman-tools keeper`: it reads its bus account and its config, watches the machine, connects and
|
||||
// runs until stopped. It refuses to start naming what is missing, and never logs a password.
|
||||
func keeper() int {
|
||||
logf := func(f string, a ...any) { fmt.Fprintf(os.Stderr, f+"\n", a...) }
|
||||
brokerFile := os.Getenv(EnvBroker)
|
||||
if brokerFile == "" {
|
||||
logf("[pacman] keeper was started without %s, its bus account's file", EnvBroker)
|
||||
return 2
|
||||
}
|
||||
account, err := process.ReadAccount(brokerFile)
|
||||
if err != nil {
|
||||
logf("[pacman] keeper: %v", err)
|
||||
return 1
|
||||
}
|
||||
cfg, err := ReadKeeperConfig(os.Getenv(EnvConfig))
|
||||
if err != nil {
|
||||
logf("[pacman] keeper: %v", err)
|
||||
return 1
|
||||
}
|
||||
// the watch starts before the first read, so a transaction between the two is not missed.
|
||||
w, err := WatchMachine(PacmanDBDir, LockName, MirrorDir, MirrorName)
|
||||
if err != nil {
|
||||
logf("[pacman] keeper: watching the machine: %v", err)
|
||||
return 1
|
||||
}
|
||||
defer w.Close()
|
||||
ctx, stop := signal.NotifyContext(context.Background(), os.Interrupt, syscall.SIGTERM)
|
||||
defer stop()
|
||||
k := &Keeper{
|
||||
Node: account.Node, Config: cfg, Machine: ThisMachine(), Clock: realClock{}, Watcher: w, Logf: logf,
|
||||
Locked: func() bool { _, err := os.Stat(filepath.Join(PacmanDBDir, LockName)); return err == nil },
|
||||
DBChanged: func() time.Time { return modified(filepath.Join(PacmanDBDir, "local")) },
|
||||
MakeHooks: keeperHooks,
|
||||
}
|
||||
logf("[pacman] keeper of %s is starting as %s", account.Node, account.User)
|
||||
if err := runKeeper(ctx, account, k); err != nil && ctx.Err() == nil {
|
||||
logf("[pacman] keeper stopped: %v", err)
|
||||
return 1
|
||||
}
|
||||
return 0
|
||||
}
|
||||
|
||||
func modified(path string) time.Time {
|
||||
if fi, err := os.Stat(path); err == nil {
|
||||
return fi.ModTime()
|
||||
}
|
||||
return time.Time{}
|
||||
}
|
||||
|
||||
// runKeeper connects as the keeper's own account and runs k on that connection until ctx ends. Apart from keeper
|
||||
// so a test can drive it over a bus of its own.
|
||||
func runKeeper(ctx context.Context, account process.Account, k *Keeper) error {
|
||||
conn, err := process.Connect(account)
|
||||
if err != nil {
|
||||
return fmt.Errorf("the bus cannot be reached: %w", err)
|
||||
}
|
||||
defer conn.Close()
|
||||
if k.Node == "" {
|
||||
k.Node = account.Node
|
||||
}
|
||||
k.State = func(name string) Store { return conn.State(name) }
|
||||
k.Events = conn
|
||||
k.AskRecord = func(id string) (*stdio.StateEntry, error) { return conn.Ask(MessengerAsksBucket, id) }
|
||||
return k.Run(ctx)
|
||||
}
|
||||
@@ -0,0 +1,97 @@
|
||||
package main
|
||||
|
||||
// What the keeper puts on the bus is only its named fields (novox/hq ADR 0507 §1; acceptance 1840/AC7): pacman's
|
||||
// output holding command lines, environment variables and messages leaves none of them in a value or an event,
|
||||
// and no command the keeper runs names a package, a foreign one least of all. (busshapes_test.go holds every
|
||||
// shape to its named fields.)
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
)
|
||||
|
||||
func TestPacmanOutputHoldingCommandLinesAndEnvironmentPutsNeitherOnTheBus(t *testing.T) {
|
||||
dir := filepath.Join("testdata", "pacman-output-with-command-lines-and-environment")
|
||||
read := func(name string) string {
|
||||
b, err := os.ReadFile(filepath.Join(dir, name))
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
return string(b)
|
||||
}
|
||||
db := &pacmanDB{sl: read("pacman-Sl"), checkup: Ran{Stdout: read("checkupdates")}}
|
||||
db.set(read("pacman-Q"), read("pacman-Qeq"))
|
||||
k, states, ev, clock, w := newTestKeeper(db)
|
||||
ctx, cancel := context.WithCancel(context.Background())
|
||||
done := make(chan error, 1)
|
||||
go func() { done <- k.Run(ctx) }()
|
||||
for deadline := time.Now().Add(5 * time.Second); time.Now().Before(deadline); time.Sleep(time.Millisecond) {
|
||||
if e, _ := states.store(StateUpdates).Get("node-a"); e != nil {
|
||||
break
|
||||
}
|
||||
}
|
||||
clock.Advance(time.Minute)
|
||||
w.c <- LockReleased
|
||||
w.c <- MirrorsRewritten
|
||||
time.Sleep(20 * time.Millisecond)
|
||||
cancel()
|
||||
<-done
|
||||
|
||||
var onBus []string
|
||||
for _, name := range []string{StateInstalled, StateUpdates} {
|
||||
s := states.store(name)
|
||||
keys, _ := s.Keys()
|
||||
for _, key := range keys {
|
||||
e, _ := s.Get(key)
|
||||
onBus = append(onBus, name+"."+key+" "+string(e.Value))
|
||||
}
|
||||
}
|
||||
events := ev.Take()
|
||||
if len(events) == 0 {
|
||||
t.Fatal("the first write said nothing")
|
||||
}
|
||||
for _, e := range events {
|
||||
body, _ := json.Marshal(e.body)
|
||||
onBus = append(onBus, e.name+" "+string(body))
|
||||
}
|
||||
if len(onBus) < 3 {
|
||||
t.Fatalf("on the bus: %v", onBus)
|
||||
}
|
||||
whole := strings.Join(onBus, "\n")
|
||||
for _, never := range []string{"=", "sudo", "--noconfirm", "/", "rm", "warning", "hunter2", "Synchronizing", "HOME", "TOKEN", "SECRET"} {
|
||||
if strings.Contains(whole, never) {
|
||||
t.Errorf("%q reached the bus:\n%s", never, whole)
|
||||
}
|
||||
}
|
||||
var inst InstalledValue
|
||||
e, _ := states.store(StateInstalled).Get("node-a")
|
||||
json.Unmarshal(e.Value, &inst)
|
||||
if !samePackages(inst.Packages, []InstalledPackage{{"glibc", "2.42-1", false, "core"}, {"yay", "12.0-1", true, RepoForeign}, {"zsh", "5.9-1", true, "extra"}}) {
|
||||
t.Errorf("the packages are read, the rest left out: %+v", inst.Packages)
|
||||
}
|
||||
var up UpdatesValue
|
||||
e, _ = states.store(StateUpdates).Get("node-a")
|
||||
json.Unmarshal(e.Value, &up)
|
||||
if len(up.Packages) != 1 || up.Packages[0] != (VersionChange{"glibc", "2.42-1", "2.42-2"}) {
|
||||
t.Errorf("updates: %+v", up)
|
||||
}
|
||||
|
||||
// no command the keeper ran names a package; the foreign one least of all.
|
||||
db.mu.Lock()
|
||||
defer db.mu.Unlock()
|
||||
for _, c := range db.commands {
|
||||
switch c {
|
||||
case "pacman -Q", "pacman -Qeq", "pacman -Sl", "checkupdates":
|
||||
default:
|
||||
t.Errorf("the keeper ran %q", c)
|
||||
}
|
||||
if strings.Contains(c, "yay") {
|
||||
t.Errorf("a command named the foreign package: %q", c)
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,172 @@
|
||||
package main
|
||||
|
||||
// The upgrade parts joined to the keeper's loop (keeperHooks): the log read into runs and old records pruned at
|
||||
// the start, an ask for this node taken from `asks`, and the records pruned again a day later.
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
)
|
||||
|
||||
func TestTheKeeperRunsTheUpgradePartsAtItsStartOnItsAsksAndDaily(t *testing.T) {
|
||||
db := &pacmanDB{sl: syncDBs, checkup: Ran{Status: 2}}
|
||||
db.set("glibc 2.42-1\n", "")
|
||||
k, states, _, clock, _ := newTestKeeper(db)
|
||||
log := capturedLog(t)
|
||||
k.Machine.ReadFile = func(path string) ([]byte, error) {
|
||||
if path == PacmanLog {
|
||||
return []byte(log), nil
|
||||
}
|
||||
return nil, errNoFile
|
||||
}
|
||||
k.MakeHooks = keeperHooks
|
||||
k.Logf = t.Logf
|
||||
k.Config = KeeperConfig{RecordDays: 180}
|
||||
records := states.store(StateRecords)
|
||||
records.Put(NodeKey("node-a", "20250101T040000Z"), Record{Node: "node-a", Run: "20250101T040000Z"})
|
||||
records.Put(NodeKey("node-b", "20250101T040000Z"), Record{Node: "node-b", Run: "20250101T040000Z"})
|
||||
done := Run{Node: "node-a", Run: "20261003T040000Z", Kind: RunUpgrade, From: FromAsk, Ask: "ask-1",
|
||||
Steps: []Step{{Step: StepRefused, At: t0, Why: "a test"}}}
|
||||
|
||||
ctx, cancel := context.WithCancel(context.Background())
|
||||
defer cancel()
|
||||
ended := make(chan error, 1)
|
||||
go func() { ended <- k.Run(ctx) }()
|
||||
until := func(what string, ok func() bool) {
|
||||
t.Helper()
|
||||
for deadline := time.Now().Add(5 * time.Second); time.Now().Before(deadline); time.Sleep(time.Millisecond) {
|
||||
if ok() {
|
||||
return
|
||||
}
|
||||
}
|
||||
t.Fatalf("never: %s", what)
|
||||
}
|
||||
until("the log read into runs", func() bool {
|
||||
keys, _ := states.store(StateRuns).Keys()
|
||||
return len(keys) > 1
|
||||
})
|
||||
until("the old record pruned at the start", func() bool {
|
||||
e, _ := records.Get(NodeKey("node-a", "20250101T040000Z"))
|
||||
return e == nil
|
||||
})
|
||||
if e, _ := records.Get(NodeKey("node-b", "20250101T040000Z")); e == nil {
|
||||
t.Error("another node's record was pruned")
|
||||
}
|
||||
|
||||
// a run an ask started, ended; its ask left behind by a keeper that went away.
|
||||
states.store(StateRuns).Put(NodeKey("node-a", done.Run), done)
|
||||
asks := states.store(StateAsks)
|
||||
asks.Put(NodeKey("node-a", "ask-1"), Ask{ID: "ask-1", Node: "node-a", Act: ActUpgrade, At: t0})
|
||||
until("the ask a run names deleted", func() bool {
|
||||
e, _ := asks.Get(NodeKey("node-a", "ask-1"))
|
||||
return e == nil
|
||||
})
|
||||
|
||||
records.Put(NodeKey("node-a", "20250102T040000Z"), Record{Node: "node-a", Run: "20250102T040000Z"})
|
||||
time.Sleep(20 * time.Millisecond)
|
||||
if e, _ := records.Get(NodeKey("node-a", "20250102T040000Z")); e == nil {
|
||||
t.Fatal("pruned before a day passed")
|
||||
}
|
||||
clock.Advance(UpdatesEvery)
|
||||
until("the old record pruned a day later", func() bool {
|
||||
e, _ := records.Get(NodeKey("node-a", "20250102T040000Z"))
|
||||
return e == nil
|
||||
})
|
||||
cancel()
|
||||
if err := <-ended; err == nil || !strings.Contains(err.Error(), "canceled") {
|
||||
t.Errorf("Run: %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
// roundKeeper is a keeper of node in a round of the made-up nodes order, its machine's time the keeper's clock.
|
||||
func roundKeeper(t *testing.T, node string, order []string) (*Keeper, *fakeStates, *fakeClock) {
|
||||
t.Helper()
|
||||
db := &pacmanDB{sl: syncDBs, checkup: Ran{Status: 2}}
|
||||
db.set("glibc 2.42-1\n", "")
|
||||
k, states, _, clock, _ := newTestKeeper(db)
|
||||
k.Node = node
|
||||
k.Machine.Now = clock.Now
|
||||
k.MakeHooks = keeperHooks
|
||||
k.Logf = t.Logf
|
||||
k.Config = KeeperConfig{RecordDays: 180, RoundStartMinutes: 120, UpgradeOrder: order}
|
||||
k.AskRecord = func(string) (*stdio.StateEntry, error) { return nil, nil }
|
||||
return k, states, clock
|
||||
}
|
||||
|
||||
func until(t *testing.T, what string, ok func() bool) {
|
||||
t.Helper()
|
||||
for deadline := time.Now().Add(5 * time.Second); time.Now().Before(deadline); time.Sleep(time.Millisecond) {
|
||||
if ok() {
|
||||
return
|
||||
}
|
||||
}
|
||||
t.Fatalf("never: %s", what)
|
||||
}
|
||||
|
||||
func roundNodeOf(s *fakeStore, round, node string) RoundNode {
|
||||
var rn RoundNode
|
||||
if e, _ := s.Get(RoundNodeKey(round, node)); e != nil {
|
||||
_ = json.Unmarshal(e.Value, &rn)
|
||||
}
|
||||
return rn
|
||||
}
|
||||
|
||||
// 1841/AC6 through the keeper's loop: the first of the order opens the week's round at its start on Saturday
|
||||
// 04:00 and starts its run in it; the round's parts get the operator channel's reader and the rounds state.
|
||||
func TestTheFirstNodesKeeperOpensTheRoundAndTakesItsTurn(t *testing.T) {
|
||||
k, states, _ := roundKeeper(t, "node-a", []string{"node-a", "node-b"})
|
||||
ctx, cancel := context.WithCancel(context.Background())
|
||||
defer cancel()
|
||||
go func() { _ = k.Run(ctx) }()
|
||||
rounds := states.store(StateRounds)
|
||||
round := RoundID(t0)
|
||||
until(t, "the round opened", func() bool { e, _ := rounds.Get(round); return e != nil })
|
||||
until(t, "node-a's run named in the round", func() bool { return roundNodeOf(rounds, round, "node-a").Run != "" })
|
||||
if rn := roundNodeOf(rounds, round, "node-b"); rn.Run != "" {
|
||||
t.Errorf("node-b started in node-a's keeper: %+v", rn)
|
||||
}
|
||||
}
|
||||
|
||||
// 1841/AC6 through the keeper's loop: a node waits until the node before it passed its check, and starts on
|
||||
// the change of `runs` that says so, with no wait for a timer.
|
||||
func TestTheNextNodesKeeperStartsWhenTheRunBeforeItPassed(t *testing.T) {
|
||||
k, states, clock := roundKeeper(t, "node-b", []string{"node-a", "node-b"})
|
||||
round := RoundID(t0)
|
||||
rounds, runs := states.store(StateRounds), states.store(StateRuns)
|
||||
runA := RunID(t0)
|
||||
rounds.Put(round, Round{Round: round, Opened: t0, OpenedBy: "node-a", Order: []string{"node-a", "node-b"}})
|
||||
rounds.Put(RoundNodeKey(round, "node-a"), RoundNode{Round: round, Node: "node-a", Run: runA})
|
||||
a := Run{Node: "node-a", Run: runA, Kind: RunUpgrade, From: FromRound, Round: round,
|
||||
Steps: []Step{{Step: StepRecorded, At: t0}, {Step: StepStarted, At: t0}}}
|
||||
runs.Put(NodeKey("node-a", runA), a)
|
||||
clock.Advance(10 * time.Minute)
|
||||
|
||||
ctx, cancel := context.WithCancel(context.Background())
|
||||
defer cancel()
|
||||
go func() { _ = k.Run(ctx) }()
|
||||
time.Sleep(30 * time.Millisecond)
|
||||
if rn := roundNodeOf(rounds, round, "node-b"); rn.Run != "" {
|
||||
t.Fatalf("node-b started before node-a passed: %+v", rn)
|
||||
}
|
||||
a.Steps = append(a.Steps, Step{Step: StepFinished, At: clock.Now()}, Step{Step: StepCheckedAfter, At: clock.Now(), Check: &Check{Passed: true}})
|
||||
runs.Put(NodeKey("node-a", runA), a)
|
||||
until(t, "node-b's run named in the round", func() bool { return roundNodeOf(rounds, round, "node-b").Run != "" })
|
||||
}
|
||||
|
||||
// The pass of a node is answered from the operator channel's record: the keeper hands its reader on.
|
||||
func TestTheKeeperHandsTheOperatorChannelsReaderToTheRound(t *testing.T) {
|
||||
k, states, _ := roundKeeper(t, "node-a", nil)
|
||||
k.State = states.State
|
||||
h := keeperHooks(k)
|
||||
if k.upgrades == nil || k.upgrades.AskRecord == nil || k.upgrades.Rounds == nil {
|
||||
t.Fatalf("the round's parts lack the channel's reader or the rounds state: %+v", k.upgrades)
|
||||
}
|
||||
if len(h.Wakes) < 2 {
|
||||
t.Fatalf("no wake for the round: %d wakes", len(h.Wakes))
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,216 @@
|
||||
//go:build integration
|
||||
|
||||
package main
|
||||
|
||||
// The keeper whole over a real bus (novox/hq ADR 0507 §2, §3, §6; acceptance 1840/AC10; the shape of brother's
|
||||
// device_integration_test.go): a JetStream server of the test's own, the module's state buckets and its event
|
||||
// stream, and a fake machine. At its start and after a released lock the node's key and package.changed are on
|
||||
// the bus, and pacman_packages on another instance reads them from there. Compiled by the merge gate's
|
||||
// `go vet -tags integration`; run by the merge check, never by hand.
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
"git.novox.be/novox/mesh-sdk/go/process"
|
||||
"github.com/nats-io/nats-server/v2/server"
|
||||
"github.com/nats-io/nats.go"
|
||||
"github.com/nats-io/nats.go/jetstream"
|
||||
)
|
||||
|
||||
func newTestBus(t *testing.T) string {
|
||||
t.Helper()
|
||||
opts := &server.Options{Host: "127.0.0.1", Port: server.RANDOM_PORT, JetStream: true, StoreDir: t.TempDir(),
|
||||
NoLog: true, NoSigs: true}
|
||||
s, err := server.NewServer(opts)
|
||||
if err != nil {
|
||||
t.Fatalf("a bus for this test could not be made: %v", err)
|
||||
}
|
||||
go s.Start()
|
||||
if !s.ReadyForConnections(30 * time.Second) {
|
||||
s.Shutdown()
|
||||
t.Fatal("a bus for this test did not come up within 30s")
|
||||
}
|
||||
t.Cleanup(func() { s.Shutdown(); s.WaitForShutdown() })
|
||||
return s.ClientURL()
|
||||
}
|
||||
|
||||
// kvReader reads a bucket as a tools bundle's state does (stdio.State): what another instance's verbs see.
|
||||
type kvReader struct{ kv jetstream.KeyValue }
|
||||
|
||||
func (r kvReader) Get(key string) (*stdio.StateEntry, error) {
|
||||
e, err := r.kv.Get(context.Background(), key)
|
||||
if errors.Is(err, jetstream.ErrKeyNotFound) {
|
||||
return nil, nil
|
||||
}
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return &stdio.StateEntry{Key: key, Value: json.RawMessage(e.Value()), Revision: e.Revision()}, nil
|
||||
}
|
||||
|
||||
func (r kvReader) Keys() ([]string, error) {
|
||||
keys, err := r.kv.Keys(context.Background())
|
||||
if errors.Is(err, jetstream.ErrNoKeysFound) {
|
||||
return []string{}, nil
|
||||
}
|
||||
return keys, err
|
||||
}
|
||||
|
||||
func (r kvReader) History(key string) ([]stdio.StateRevision, error) {
|
||||
h, err := r.kv.History(context.Background(), key)
|
||||
if errors.Is(err, jetstream.ErrKeyNotFound) {
|
||||
return []stdio.StateRevision{}, nil
|
||||
}
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
out := []stdio.StateRevision{}
|
||||
for _, e := range h {
|
||||
op := "put"
|
||||
if e.Operation() != jetstream.KeyValuePut {
|
||||
op = "delete"
|
||||
}
|
||||
out = append(out, stdio.StateRevision{Key: key, Value: e.Value(), Revision: e.Revision(), Op: op, At: e.Created()})
|
||||
}
|
||||
return out, nil
|
||||
}
|
||||
|
||||
func TestTheKeeperPutsTheNodesPackagesOnARealBusAndAnotherInstanceReadsThem(t *testing.T) {
|
||||
url := newTestBus(t)
|
||||
ctx := context.Background()
|
||||
nc, err := nats.Connect(url)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
defer nc.Close()
|
||||
js, err := jetstream.New(nc)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
account := process.Account{URL: url, Node: "node-a", Module: "pacman", User: "node-a.pacman.keeper", Password: "pw"}
|
||||
buckets := map[string]jetstream.KeyValue{}
|
||||
for _, name := range States {
|
||||
kv, err := js.CreateKeyValue(ctx, jetstream.KeyValueConfig{Bucket: account.Module + "_" + name, History: 8})
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
buckets[name] = kv
|
||||
}
|
||||
declared, err := js.CreateKeyValue(ctx, jetstream.KeyValueConfig{Bucket: "mesh-controller_declared-packages"})
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
declared.Put(ctx, "node-a", []byte(`{"node":"node-a","packages":[{"package":"zsh","modules":["shell"],"absent":false}]}`))
|
||||
stream, err := js.CreateStream(ctx, jetstream.StreamConfig{Name: "pacman-events", Subjects: []string{"mesh.mod.pacman.event.>"}})
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
|
||||
db := &pacmanDB{sl: syncDBs, checkup: Ran{Stdout: "zsh 5.9-1 -> 5.9-2\n"}}
|
||||
db.set("glibc 2.42-1\nzsh 5.9-1\n", "zsh\n")
|
||||
signals := make(chan Signal)
|
||||
k := &Keeper{Config: KeeperConfig{}, Machine: machine(db.runner(), 1000), Clock: realClock{},
|
||||
Watcher: fakeWatcher{signals}, Logf: t.Logf, MakeHooks: keeperHooks}
|
||||
runCtx, cancel := context.WithCancel(ctx)
|
||||
done := make(chan error, 1)
|
||||
go func() { done <- runKeeper(runCtx, account, k) }()
|
||||
|
||||
waitKey := func(kv jetstream.KeyValue, key, holds string) string {
|
||||
t.Helper()
|
||||
for deadline := time.Now().Add(15 * time.Second); time.Now().Before(deadline); time.Sleep(50 * time.Millisecond) {
|
||||
if e, err := kv.Get(ctx, key); err == nil && strings.Contains(string(e.Value()), holds) {
|
||||
return string(e.Value())
|
||||
}
|
||||
}
|
||||
t.Fatalf("%s never held %q", key, holds)
|
||||
return ""
|
||||
}
|
||||
waitKey(buckets[StateInstalled], "node-a", `"zsh"`)
|
||||
waitKey(buckets[StateUpdates], "node-a", `"reachable":true`)
|
||||
|
||||
// a transaction: tmux added, zsh upgraded; the lock released.
|
||||
db.set("glibc 2.42-1\ntmux 3.5-1\nzsh 5.9-2\n", "tmux\nzsh\n")
|
||||
db.setUpdates(Ran{Status: 2})
|
||||
signals <- LockReleased
|
||||
waitKey(buckets[StateInstalled], "node-a", `"tmux"`)
|
||||
waitKey(buckets[StateUpdates], "node-a", `"packages":[]`)
|
||||
|
||||
var changes []PackageChanged
|
||||
for deadline := time.Now().Add(15 * time.Second); len(changes) < 2 && time.Now().Before(deadline); time.Sleep(50 * time.Millisecond) {
|
||||
changes = nil
|
||||
info, err := stream.Info(ctx)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
for seq := uint64(1); seq <= info.State.LastSeq; seq++ {
|
||||
msg, err := stream.GetMsg(ctx, seq)
|
||||
if err != nil || msg.Subject != "mesh.mod.pacman.event."+EventPackageChanged {
|
||||
continue
|
||||
}
|
||||
var pc PackageChanged
|
||||
if err := json.Unmarshal(msg.Data, &pc); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
changes = append(changes, pc)
|
||||
}
|
||||
}
|
||||
if len(changes) != 2 {
|
||||
t.Fatalf("package.changed at the start and after the transaction: %+v", changes)
|
||||
}
|
||||
last := changes[1]
|
||||
if last.Node != "node-a" || len(last.Added) != 1 || last.Added[0].Name != "tmux" || len(last.Changed) != 1 ||
|
||||
last.Changed[0] != (VersionChange{"zsh", "5.9-1", "5.9-2"}) || len(last.Removed) != 0 {
|
||||
t.Errorf("the transaction's package.changed: %+v", last)
|
||||
}
|
||||
|
||||
// another instance's verbs read what the keeper wrote, asking no machine.
|
||||
other := machine(func(_ context.Context, name string, args ...string) Ran {
|
||||
t.Errorf("another instance asked a machine: %s %v", name, args)
|
||||
return Ran{Status: 99}
|
||||
}, 1000)
|
||||
other.Node = "node-b"
|
||||
other.States = func(name string) StateReader {
|
||||
if name == DeclaredPackagesState {
|
||||
return kvReader{declared}
|
||||
}
|
||||
return kvReader{buckets[name]}
|
||||
}
|
||||
r, err := other.Packages("zsh", 500)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
on := r["nodes"].([]PackageOnNode)
|
||||
if len(on) != 1 || on[0].Node != "node-a" || on[0].Version != "5.9-2" || strings.Join(on[0].DeclaredBy, ",") != "shell" {
|
||||
t.Errorf("pacman_packages zsh from another instance: %+v", on)
|
||||
}
|
||||
r, err = other.Packages("", 500)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if v := r["nodes"].([]NodeView); len(v) != 1 || v[0].Packages != 3 || len(v[0].Undeclared) != 1 || v[0].Undeclared[0].Name != "tmux" {
|
||||
t.Errorf("the view from another instance: %+v", v)
|
||||
}
|
||||
u, err := other.UpdatesFor("node-a")
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if v := u["nodes"].([]UpdatesView); len(v) != 1 || !v[0].Reachable || v[0].Count != 0 || v[0].Changed == nil {
|
||||
t.Errorf("pacman_updates from another instance: %+v", v)
|
||||
}
|
||||
|
||||
cancel()
|
||||
select {
|
||||
case err := <-done:
|
||||
if !errors.Is(err, context.Canceled) {
|
||||
t.Errorf("runKeeper ended with %v, want its context's end", err)
|
||||
}
|
||||
case <-time.After(10 * time.Second):
|
||||
t.Fatal("runKeeper did not stop once its context was cancelled")
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,213 @@
|
||||
package main
|
||||
|
||||
// The manifest of the keeper (novox/hq ADR 0507 §1-§3, §5; ADR 0508 §1, §2, §5, §6, §8; acceptance 1841/AC12):
|
||||
// the module's background work is a process of its own under an account the module makes, with a bus account of
|
||||
// its own, the module's six states and the nine events it emits; the module's config values land in the
|
||||
// keeper's config file, which restarts it; the weekly timer and its service are gone (a dropped file or service
|
||||
// is removed by the node-engine); `upgrade-order` falls back to the mesh-level key of that name.
|
||||
|
||||
import (
|
||||
"bytes"
|
||||
"encoding/json"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"reflect"
|
||||
"strings"
|
||||
"testing"
|
||||
)
|
||||
|
||||
// rawManifest is the manifest as written, every field kept.
|
||||
func rawManifest(t *testing.T) map[string]any {
|
||||
t.Helper()
|
||||
raw, err := os.ReadFile(filepath.Join(moduleDir, "module.json"))
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
var m map[string]any
|
||||
if err := json.Unmarshal(raw, &m); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
return m
|
||||
}
|
||||
|
||||
func strs(v any) []string {
|
||||
var out []string
|
||||
l, _ := v.([]any)
|
||||
for _, x := range l {
|
||||
s, _ := x.(string)
|
||||
out = append(out, s)
|
||||
}
|
||||
return out
|
||||
}
|
||||
|
||||
func TestTheWeeklyTimerAndItsServiceAreGone(t *testing.T) {
|
||||
m := manifest(t)
|
||||
for _, r := range m.Resources {
|
||||
id, _ := r["id"].(string)
|
||||
path, _ := r["path"].(string)
|
||||
unit, _ := r["unit"].(string)
|
||||
if id == "weekly-upgrade" || id == "upgrade-timer" || id == "upgrade-service" ||
|
||||
strings.Contains(path, "weekly-upgrade") || strings.Contains(unit, "weekly-upgrade") {
|
||||
t.Errorf("the weekly upgrade is replaced by the round (ADR 0508 §5), and %v still declares it", id)
|
||||
}
|
||||
}
|
||||
for _, name := range []string{"mesh-pacman-weekly-upgrade.timer", "mesh-pacman-weekly-upgrade.service"} {
|
||||
if _, err := os.Stat(filepath.Join(moduleDir, "files", name)); err == nil {
|
||||
t.Errorf("files/%s is still shipped", name)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheKeeperIsAProcessUnderItsOwnAccountWithABusAccountOfItsOwn(t *testing.T) {
|
||||
m := manifest(t)
|
||||
user := m.resource(t, "keeper-account")
|
||||
if user["type"] != "user" || user["name"] != "pacman-keeper" || user["shell"] != "/usr/bin/nologin" {
|
||||
t.Fatalf("the keeper's account: %v", user)
|
||||
}
|
||||
k := m.resource(t, "keeper")
|
||||
if k["type"] != "process" || k["name"] != "pacman-keeper" || k["artifact"] != "tools" ||
|
||||
strings.Join(strs(k["run"]), " ") != "./"+binaryName+" keeper" {
|
||||
t.Fatalf("the keeper: %v", k)
|
||||
}
|
||||
if k["user"] != "pacman-keeper" {
|
||||
t.Errorf("the keeper runs as its own account, never root (ADR 0508 §8): user %v", k["user"])
|
||||
}
|
||||
env, _ := k["env"].(map[string]any)
|
||||
if len(env) != 3 || env[EnvBroker] != "${dir:state}/broker-keeper.secret" || env[EnvConfig] != "${dir:state}/keeper.json" ||
|
||||
env[EnvRootSocket] != RootSocketPath {
|
||||
t.Errorf("the keeper's environment names its account file, its config file and the root helper's socket: %v", env)
|
||||
}
|
||||
secrets, _ := rawManifest(t)["own-secrets"].(map[string]any)
|
||||
broker, _ := secrets["broker-keeper"].(map[string]any)
|
||||
if broker["path"] != env[EnvBroker] || broker["owner"] != "pacman-keeper" || len(secrets) != 1 {
|
||||
t.Errorf("the bus account is the keeper's own secret, owned by its account: %v", secrets)
|
||||
}
|
||||
state := m.resource(t, "state")
|
||||
if state["type"] != "directory" || state["place"] != "." || state["unknown"] == nil {
|
||||
t.Errorf("the state directory: %v", state)
|
||||
}
|
||||
order := map[string]int{}
|
||||
for i, r := range m.Resources {
|
||||
order[r["id"].(string)] = i
|
||||
}
|
||||
if order["keeper-account"] > order["keeper"] || order["state"] > order["keeper-config"] || order["keeper-config"] > order["keeper"] {
|
||||
t.Errorf("the account, the directory and the config file come before the process: %v", order)
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheKeeperDeclaresTheModulesStatesAndEvents(t *testing.T) {
|
||||
raw := rawManifest(t)
|
||||
k := manifest(t).resource(t, "keeper")
|
||||
bus, _ := k["bus"].(map[string]any)
|
||||
var events []string
|
||||
for _, e := range Events {
|
||||
events = append(events, e.Name)
|
||||
}
|
||||
if got := strs(bus["state"]); !reflect.DeepEqual(got, States) {
|
||||
t.Errorf("the keeper's states: %v, want %v", got, States)
|
||||
}
|
||||
if got := strs(bus["emits"]); !reflect.DeepEqual(got, events) {
|
||||
t.Errorf("the keeper's events: %v, want %v", got, events)
|
||||
}
|
||||
for key := range bus {
|
||||
if key != "state" && key != "emits" && key != "asks" {
|
||||
t.Errorf("the keeper's bus declares %s: it keeps its own states, emits its events and reads its asks' answers, nothing else", key)
|
||||
}
|
||||
}
|
||||
// A pass is answered on the operator's channel: the tools ask, the keeper reads the channel's record of its own
|
||||
// asks (ADR 0508 §6, ADR 0259, ADR 0436 §1).
|
||||
if bus["asks"] != true || strings.Join(strs(raw["uses"]), " ") != "operator-channel" {
|
||||
t.Errorf("the keeper reads its asks' answers and the module uses the operator's channel: %v %v", bus["asks"], raw["uses"])
|
||||
}
|
||||
// The tools read round-start-minutes from the keeper's config file.
|
||||
for _, a := range manifest(t).Build.Artifacts {
|
||||
if env, _ := a["env"].(map[string]any); a["name"] == "tools" && env[EnvConfig] != "${dir:state}/keeper.json" {
|
||||
t.Errorf("the tools bundle is not named the keeper's config file: %v", a["env"])
|
||||
}
|
||||
}
|
||||
if got := strs(raw["state"]); !reflect.DeepEqual(got, States) {
|
||||
t.Errorf("the module's states: %v, want %v", got, States)
|
||||
}
|
||||
if got := strs(raw["emits"]); !reflect.DeepEqual(got, events) {
|
||||
t.Errorf("the module's events: %v, want %v", got, events)
|
||||
}
|
||||
// The verbs read which packages each node's modules declare; the process reads no other module's state
|
||||
// (ADR 0507 §5, ADR 0436 §1).
|
||||
if got := strs(raw["reads"]); !reflect.DeepEqual(got, []string{"mesh-controller.declared-packages"}) {
|
||||
t.Errorf("reads: %v", got)
|
||||
}
|
||||
}
|
||||
|
||||
// configValue is the module's config value of that key.
|
||||
func configValue(t *testing.T, key string) map[string]any {
|
||||
t.Helper()
|
||||
config, _ := rawManifest(t)["config"].(map[string]any)
|
||||
v, _ := config[key].(map[string]any)
|
||||
if v == nil {
|
||||
t.Fatalf("no config value %s", key)
|
||||
}
|
||||
if d, _ := v["description"].(string); len(d) < 40 {
|
||||
t.Errorf("%s says what it means: %q", key, d)
|
||||
}
|
||||
return v
|
||||
}
|
||||
|
||||
func TestTheConfigValuesLandInTheKeepersConfigFileWhichRestartsIt(t *testing.T) {
|
||||
order := configValue(t, "upgrade-order")
|
||||
if order["type"] != "list" || order["trusted"] != true || order["predefined"] != "upgrade-order" ||
|
||||
!reflect.DeepEqual(order["default"], []any{}) || order["required"] != nil {
|
||||
t.Errorf("upgrade-order is a trusted list falling back to the mesh-level key, empty by default: %v", order)
|
||||
}
|
||||
for key, def := range map[string]float64{"record-days": 180, "round-start-minutes": 120} {
|
||||
v := configValue(t, key)
|
||||
if v["type"] != "number" || v["default"] != def || v["minimum"] != float64(1) {
|
||||
t.Errorf("%s is a number of at least 1, %v by default: %v", key, def, v)
|
||||
}
|
||||
}
|
||||
config, _ := rawManifest(t)["config"].(map[string]any)
|
||||
if len(config) != 3 {
|
||||
t.Errorf("config: %v", config)
|
||||
}
|
||||
m := manifest(t)
|
||||
f := m.resource(t, "keeper-config")
|
||||
if f["type"] != "file" || f["path"] != "${dir:state}/keeper.json" || f["merge"] != "json" || f["trusted"] != true || f["mode"] != "0644" {
|
||||
t.Fatalf("the keeper's config file: %v", f)
|
||||
}
|
||||
if got := strs(f["takes-config"]); strings.Join(got, " ") != "upgrade-order record-days round-start-minutes" {
|
||||
t.Errorf("the file takes every config value: %v", got)
|
||||
}
|
||||
if got := strs(m.resource(t, "keeper")["restart-on"]); strings.Join(got, " ") != "keeper-config" {
|
||||
t.Errorf("a changed config file restarts the keeper, which reads it at its start: %v", got)
|
||||
}
|
||||
// The template is the file a node starts with: the defaults, read as the keeper reads them.
|
||||
dec := json.NewDecoder(bytes.NewReader([]byte(templateText(t, f))))
|
||||
dec.DisallowUnknownFields()
|
||||
var c KeeperConfig
|
||||
if err := dec.Decode(&c); err != nil {
|
||||
t.Fatalf("the keeper cannot read its config file: %v", err)
|
||||
}
|
||||
if c.UpgradeOrder == nil || len(c.UpgradeOrder) != 0 || c.RecordDays != 180 || c.RoundStartMinutes != 120 {
|
||||
t.Errorf("the template holds the defaults: %+v", c)
|
||||
}
|
||||
}
|
||||
|
||||
func TestUpgradeOrderIsATrustedListOfTheMesh(t *testing.T) {
|
||||
raw, err := os.ReadFile(filepath.Join(moduleDir, "..", "..", "predefined.json"))
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
var keys map[string]map[string]any
|
||||
if err := json.Unmarshal(raw, &keys); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
k := keys["upgrade-order"]
|
||||
if k["type"] != "list" || k["trusted"] != true {
|
||||
t.Fatalf("predefined upgrade-order: %v", k)
|
||||
}
|
||||
if d, _ := k["description"].(string); !strings.Contains(d, "order") || !strings.Contains(d, "round") {
|
||||
t.Errorf("its description says the nodes' order and the round: %q", d)
|
||||
}
|
||||
if k["type"] != configValue(t, "upgrade-order")["type"] {
|
||||
t.Error("a value falls back only to a key of its own type (ADR 0348 §4)")
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,630 @@
|
||||
package main
|
||||
|
||||
// The keeper's start and its lock watch (novox/hq ADR 0507 §2, §3; acceptance 1840/AC1-AC3), over a fake state,
|
||||
// a fake machine, a fake clock and a fake watch.
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"strings"
|
||||
"sync"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
)
|
||||
|
||||
// fakeStore is one state in memory, recording every write.
|
||||
type fakeStore struct {
|
||||
mu sync.Mutex
|
||||
name string
|
||||
values map[string]json.RawMessage
|
||||
revs map[string]uint64
|
||||
at map[string]time.Time
|
||||
rev uint64
|
||||
writes []string // "put <key>", "delete <key>"
|
||||
watchers []fakeWatch
|
||||
now func() time.Time
|
||||
}
|
||||
|
||||
type fakeWatch struct {
|
||||
key string
|
||||
handle func(stdio.StateChange) error
|
||||
}
|
||||
|
||||
func newFakeStore(name string) *fakeStore {
|
||||
return &fakeStore{name: name, values: map[string]json.RawMessage{}, revs: map[string]uint64{}, at: map[string]time.Time{}}
|
||||
}
|
||||
|
||||
func (s *fakeStore) Get(key string) (*stdio.StateEntry, error) {
|
||||
s.mu.Lock()
|
||||
defer s.mu.Unlock()
|
||||
v, ok := s.values[key]
|
||||
if !ok {
|
||||
return nil, nil
|
||||
}
|
||||
return &stdio.StateEntry{Key: key, Value: v, Revision: s.revs[key]}, nil
|
||||
}
|
||||
|
||||
func (s *fakeStore) write(key string, value any, op string) (uint64, error) {
|
||||
data, err := json.Marshal(value)
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
s.mu.Lock()
|
||||
s.rev++
|
||||
s.values[key], s.revs[key], s.at[key] = data, s.rev, s.clock()
|
||||
s.writes = append(s.writes, op+" "+key)
|
||||
rev, ws := s.rev, append([]fakeWatch(nil), s.watchers...)
|
||||
s.mu.Unlock()
|
||||
for _, w := range ws {
|
||||
if matchKey(w.key, key) {
|
||||
w.handle(stdio.StateChange{State: s.name, Key: key, Op: "put", Value: data, Revision: rev})
|
||||
}
|
||||
}
|
||||
return rev, nil
|
||||
}
|
||||
|
||||
// clock is when a write lands: the store's own time, which a test sets.
|
||||
func (s *fakeStore) clock() time.Time {
|
||||
if s.now != nil {
|
||||
return s.now()
|
||||
}
|
||||
return t0
|
||||
}
|
||||
|
||||
func (s *fakeStore) Put(key string, value any) (uint64, error) { return s.write(key, value, "put") }
|
||||
|
||||
func (s *fakeStore) Create(key string, value any) (uint64, error) {
|
||||
if e, _ := s.Get(key); e != nil {
|
||||
return 0, fmt.Errorf("%s: %w", key, stdio.ErrChanged)
|
||||
}
|
||||
return s.write(key, value, "create")
|
||||
}
|
||||
|
||||
func (s *fakeStore) Update(key string, value any, revision uint64) (uint64, error) {
|
||||
if e, _ := s.Get(key); e == nil || e.Revision != revision {
|
||||
return 0, fmt.Errorf("%s: %w", key, stdio.ErrChanged)
|
||||
}
|
||||
return s.write(key, value, "update")
|
||||
}
|
||||
|
||||
func (s *fakeStore) Delete(key string) error {
|
||||
s.mu.Lock()
|
||||
delete(s.values, key)
|
||||
delete(s.revs, key)
|
||||
s.rev++
|
||||
s.writes = append(s.writes, "delete "+key)
|
||||
rev, ws := s.rev, append([]fakeWatch(nil), s.watchers...)
|
||||
s.mu.Unlock()
|
||||
for _, w := range ws {
|
||||
if matchKey(w.key, key) {
|
||||
w.handle(stdio.StateChange{State: s.name, Key: key, Op: "delete", Revision: rev})
|
||||
}
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
func (s *fakeStore) Keys() ([]string, error) {
|
||||
s.mu.Lock()
|
||||
defer s.mu.Unlock()
|
||||
keys := []string{}
|
||||
for k := range s.values {
|
||||
keys = append(keys, k)
|
||||
}
|
||||
sortStrings(keys)
|
||||
return keys, nil
|
||||
}
|
||||
|
||||
func (s *fakeStore) Watch(key string, handle func(stdio.StateChange) error) error {
|
||||
keys, _ := s.Keys()
|
||||
for _, k := range keys {
|
||||
if !matchKey(key, k) {
|
||||
continue
|
||||
}
|
||||
e, _ := s.Get(k)
|
||||
if err := handle(stdio.StateChange{State: s.name, Key: k, Op: "put", Value: e.Value, Revision: e.Revision, Current: true}); err != nil {
|
||||
return err
|
||||
}
|
||||
}
|
||||
s.mu.Lock()
|
||||
s.watchers = append(s.watchers, fakeWatch{key, handle})
|
||||
s.mu.Unlock()
|
||||
return nil
|
||||
}
|
||||
|
||||
// Writes is what was written, in order, and forgets it.
|
||||
func (s *fakeStore) Writes() []string {
|
||||
s.mu.Lock()
|
||||
defer s.mu.Unlock()
|
||||
w := s.writes
|
||||
s.writes = nil
|
||||
return w
|
||||
}
|
||||
|
||||
// matchKey narrows as the bus does: `*` one name, `**` the rest, empty every key.
|
||||
func matchKey(pattern, key string) bool {
|
||||
if pattern == "" {
|
||||
return true
|
||||
}
|
||||
p, k := strings.Split(pattern, "."), strings.Split(key, ".")
|
||||
for i := range p {
|
||||
if p[i] == "**" {
|
||||
return len(k) > i
|
||||
}
|
||||
if i >= len(k) || (p[i] != "*" && p[i] != k[i]) {
|
||||
return false
|
||||
}
|
||||
}
|
||||
return len(p) == len(k)
|
||||
}
|
||||
|
||||
func sortStrings(s []string) {
|
||||
for i := 1; i < len(s); i++ {
|
||||
for j := i; j > 0 && s[j] < s[j-1]; j-- {
|
||||
s[j], s[j-1] = s[j-1], s[j]
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// fakeStates is the module's states in memory.
|
||||
type fakeStates struct {
|
||||
mu sync.Mutex
|
||||
states map[string]*fakeStore
|
||||
}
|
||||
|
||||
func (f *fakeStates) State(name string) Store { return f.store(name) }
|
||||
|
||||
func (f *fakeStates) store(name string) *fakeStore {
|
||||
f.mu.Lock()
|
||||
defer f.mu.Unlock()
|
||||
if f.states == nil {
|
||||
f.states = map[string]*fakeStore{}
|
||||
}
|
||||
if f.states[name] == nil {
|
||||
f.states[name] = newFakeStore(name)
|
||||
}
|
||||
return f.states[name]
|
||||
}
|
||||
|
||||
// emitted is the events a keeper emitted.
|
||||
type emitted struct {
|
||||
mu sync.Mutex
|
||||
events []emittedEvent
|
||||
}
|
||||
|
||||
type emittedEvent struct {
|
||||
name string
|
||||
body any
|
||||
id string
|
||||
}
|
||||
|
||||
func (e *emitted) Emit(name string, body any, opts ...stdio.EmitOptions) error {
|
||||
e.mu.Lock()
|
||||
defer e.mu.Unlock()
|
||||
id := ""
|
||||
for _, o := range opts {
|
||||
id = o.MessageID
|
||||
}
|
||||
e.events = append(e.events, emittedEvent{name, body, id})
|
||||
return nil
|
||||
}
|
||||
|
||||
func (e *emitted) Take() []emittedEvent {
|
||||
e.mu.Lock()
|
||||
defer e.mu.Unlock()
|
||||
out := e.events
|
||||
e.events = nil
|
||||
return out
|
||||
}
|
||||
|
||||
// fakeClock moves only when the test moves it; After fires once the clock passes it.
|
||||
type fakeClock struct {
|
||||
mu sync.Mutex
|
||||
now time.Time
|
||||
timers []fakeTimer
|
||||
}
|
||||
|
||||
type fakeTimer struct {
|
||||
at time.Time
|
||||
c chan time.Time
|
||||
}
|
||||
|
||||
func (c *fakeClock) Now() time.Time {
|
||||
c.mu.Lock()
|
||||
defer c.mu.Unlock()
|
||||
return c.now
|
||||
}
|
||||
|
||||
func (c *fakeClock) After(d time.Duration) <-chan time.Time {
|
||||
c.mu.Lock()
|
||||
defer c.mu.Unlock()
|
||||
t := fakeTimer{c.now.Add(d), make(chan time.Time, 1)}
|
||||
if d <= 0 {
|
||||
t.c <- c.now
|
||||
return t.c
|
||||
}
|
||||
c.timers = append(c.timers, t)
|
||||
return t.c
|
||||
}
|
||||
|
||||
func (c *fakeClock) Advance(d time.Duration) {
|
||||
c.mu.Lock()
|
||||
defer c.mu.Unlock()
|
||||
c.now = c.now.Add(d)
|
||||
kept := c.timers[:0]
|
||||
for _, t := range c.timers {
|
||||
if !t.at.After(c.now) {
|
||||
t.c <- c.now
|
||||
continue
|
||||
}
|
||||
kept = append(kept, t)
|
||||
}
|
||||
c.timers = kept
|
||||
}
|
||||
|
||||
// fakeWatcher hands the test's signals to the keeper.
|
||||
type fakeWatcher struct{ c chan Signal }
|
||||
|
||||
func (w fakeWatcher) Signals() <-chan Signal { return w.c }
|
||||
func (w fakeWatcher) Close() error { return nil }
|
||||
|
||||
// pacmanDB is a node's package database as pacman answers it: the test changes it between reads.
|
||||
type pacmanDB struct {
|
||||
mu sync.Mutex
|
||||
q, qe string
|
||||
sl string
|
||||
checkup Ran
|
||||
commands []string
|
||||
}
|
||||
|
||||
func (d *pacmanDB) set(q, qe string) {
|
||||
d.mu.Lock()
|
||||
defer d.mu.Unlock()
|
||||
d.q, d.qe = q, qe
|
||||
}
|
||||
|
||||
func (d *pacmanDB) setUpdates(r Ran) {
|
||||
d.mu.Lock()
|
||||
defer d.mu.Unlock()
|
||||
d.checkup = r
|
||||
}
|
||||
|
||||
func (d *pacmanDB) runner() Runner {
|
||||
return func(_ context.Context, name string, args ...string) Ran {
|
||||
d.mu.Lock()
|
||||
defer d.mu.Unlock()
|
||||
line := strings.TrimSpace(name + " " + strings.Join(args, " "))
|
||||
d.commands = append(d.commands, line)
|
||||
switch line {
|
||||
case "pacman -Q":
|
||||
return Ran{Stdout: d.q}
|
||||
case "pacman -Qeq":
|
||||
return Ran{Stdout: d.qe}
|
||||
case "pacman -Sl":
|
||||
return Ran{Stdout: d.sl}
|
||||
case "checkupdates":
|
||||
return d.checkup
|
||||
}
|
||||
return Ran{Status: 99, Stderr: "unexpected command: " + line}
|
||||
}
|
||||
}
|
||||
|
||||
var syncDBs = "core glibc 2.42-1 [installed]\ncore linux 6.17.1-1 [installed]\nextra zsh 5.9-2 [installed: 5.9-1]\n" +
|
||||
"extra zsh-completions 0.35-1\nextra tmux 3.5-1\nextra-testing zsh 5.9-3\n"
|
||||
|
||||
var t0 = time.Date(2026, 10, 24, 4, 0, 0, 0, time.UTC)
|
||||
|
||||
func newTestKeeper(db *pacmanDB) (*Keeper, *fakeStates, *emitted, *fakeClock, fakeWatcher) {
|
||||
states, ev, clock := &fakeStates{}, &emitted{}, &fakeClock{now: t0}
|
||||
w := fakeWatcher{make(chan Signal)}
|
||||
m := machine(db.runner(), 1000)
|
||||
k := &Keeper{Node: "node-a", Machine: m, State: states.State, Events: ev, Clock: clock, Watcher: w,
|
||||
Logf: func(string, ...any) {}}
|
||||
return k, states, ev, clock, w
|
||||
}
|
||||
|
||||
func installedOf(t *testing.T, s *fakeStore, key string) InstalledValue {
|
||||
t.Helper()
|
||||
e, _ := s.Get(key)
|
||||
if e == nil {
|
||||
t.Fatalf("no %s", key)
|
||||
}
|
||||
var v InstalledValue
|
||||
if err := json.Unmarshal(e.Value, &v); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
return v
|
||||
}
|
||||
|
||||
// 1840/AC1: at the start every package is written with its version, its reason and its repository or foreign,
|
||||
// and written again only when it differs from the key.
|
||||
func TestAtItsStartTheKeeperWritesEveryPackageOnlyWhenItDiffersFromTheKey(t *testing.T) {
|
||||
db := &pacmanDB{sl: syncDBs, checkup: Ran{Status: 2}}
|
||||
db.set("glibc 2.42-1\nlinux 6.17.1-1\nyay 12.0-1\nzsh 5.9-1\n", "linux\nyay\nzsh\n")
|
||||
k, states, ev, _, _ := newTestKeeper(db)
|
||||
k.DBChanged = func() time.Time { return t0.Add(-time.Hour) }
|
||||
k.refreshInstalled(k.DBChanged())
|
||||
inst := states.store(StateInstalled)
|
||||
if w := inst.Writes(); strings.Join(w, ",") != "put node-a" {
|
||||
t.Fatalf("writes: %v", w)
|
||||
}
|
||||
v := installedOf(t, inst, "node-a")
|
||||
want := []InstalledPackage{
|
||||
{"glibc", "2.42-1", false, "core"}, {"linux", "6.17.1-1", true, "core"},
|
||||
{"yay", "12.0-1", true, RepoForeign}, {"zsh", "5.9-1", true, "extra"},
|
||||
}
|
||||
if !samePackages(v.Packages, want) || !v.At.Equal(t0.Add(-time.Hour)) {
|
||||
t.Fatalf("installed.node-a: %+v", v)
|
||||
}
|
||||
if e := ev.Take(); len(e) != 1 || len(e[0].body.(PackageChanged).Added) != 4 {
|
||||
t.Fatalf("the first write adds every package: %+v", e)
|
||||
}
|
||||
// a second start over the same database writes nothing and says nothing, whatever the time.
|
||||
k.refreshInstalled(t0)
|
||||
if w := inst.Writes(); len(w) != 0 {
|
||||
t.Errorf("the same packages were written again: %v", w)
|
||||
}
|
||||
if e := ev.Take(); len(e) != 0 {
|
||||
t.Errorf("the same packages were said again: %+v", e)
|
||||
}
|
||||
}
|
||||
|
||||
// 1840/AC2: a transaction that adds, removes and changes packages is one package.changed naming the node and
|
||||
// exactly those; one that changes nothing writes and emits nothing; a lock held again waits for its release.
|
||||
func TestAReleasedLockSaysExactlyWhatTheTransactionChangedOnce(t *testing.T) {
|
||||
db := &pacmanDB{sl: syncDBs, checkup: Ran{Status: 2}}
|
||||
db.set("glibc 2.42-1\nlinux 6.17.1-1\nzsh 5.9-1\n", "linux\nzsh\n")
|
||||
k, states, ev, clock, _ := newTestKeeper(db)
|
||||
k.refreshInstalled(t0)
|
||||
ev.Take()
|
||||
states.store(StateInstalled).Writes()
|
||||
|
||||
held := false
|
||||
k.Locked = func() bool { return held }
|
||||
db.set("glibc 2.42-1\nlinux 6.17.2-1\ntmux 3.5-1\n", "linux\ntmux\n")
|
||||
held = true
|
||||
k.signal(LockReleased)
|
||||
if w := states.store(StateInstalled).Writes(); len(w) != 0 {
|
||||
t.Fatalf("read while the lock was held again: %v", w)
|
||||
}
|
||||
held = false
|
||||
clock.Advance(time.Minute)
|
||||
k.signal(LockReleased)
|
||||
if w := states.store(StateInstalled).Writes(); strings.Join(w, ",") != "put node-a" {
|
||||
t.Fatalf("writes: %v", w)
|
||||
}
|
||||
e := ev.Take()
|
||||
if len(e) != 1 || e[0].name != EventPackageChanged {
|
||||
t.Fatalf("events: %+v", e)
|
||||
}
|
||||
got := e[0].body.(PackageChanged)
|
||||
if got.Node != "node-a" || !got.At.Equal(t0.Add(time.Minute)) ||
|
||||
fmt.Sprint(got.Added) != "[{tmux 3.5-1}]" || fmt.Sprint(got.Removed) != "[{zsh 5.9-1}]" ||
|
||||
fmt.Sprint(got.Changed) != "[{linux 6.17.1-1 6.17.2-1}]" {
|
||||
t.Fatalf("package.changed: %+v", got)
|
||||
}
|
||||
if !strings.HasPrefix(e[0].id, "installed.node-a.") {
|
||||
t.Errorf("the event's id is the key's revision, so a second emit is dropped: %q", e[0].id)
|
||||
}
|
||||
// a transaction that changed nothing (a database sync alone).
|
||||
k.signal(LockReleased)
|
||||
if w := states.store(StateInstalled).Writes(); len(w) != 0 {
|
||||
t.Errorf("nothing changed and the key was written: %v", w)
|
||||
}
|
||||
if e := ev.Take(); len(e) != 0 {
|
||||
t.Errorf("nothing changed and it was said: %+v", e)
|
||||
}
|
||||
// a change of reason alone rewrites the key and is no package.changed.
|
||||
db.set("glibc 2.42-1\nlinux 6.17.2-1\ntmux 3.5-1\n", "glibc\nlinux\ntmux\n")
|
||||
k.signal(LockReleased)
|
||||
if w := states.store(StateInstalled).Writes(); len(w) != 1 {
|
||||
t.Errorf("a changed reason is a changed key: %v", w)
|
||||
}
|
||||
if e := ev.Take(); len(e) != 0 {
|
||||
t.Errorf("a changed reason is no change of a version: %+v", e)
|
||||
}
|
||||
}
|
||||
|
||||
// bigDB is a package database whose value is over the bus's cap.
|
||||
func bigDB(n int) string {
|
||||
var b strings.Builder
|
||||
for i := 0; i < n; i++ {
|
||||
fmt.Fprintf(&b, "package-with-a-long-name-%05d 1.0.0-1\n", i)
|
||||
}
|
||||
return b.String()
|
||||
}
|
||||
|
||||
// 1840/AC3: a value over the bus's cap is not written: the key keeps its last value and `<node>.not-current`
|
||||
// holds the size and since when, written once; the next write that fits deletes it.
|
||||
func TestAValueOverTheCapKeepsTheLastAndMarksTheNodeNotCurrent(t *testing.T) {
|
||||
db := &pacmanDB{sl: syncDBs, checkup: Ran{Status: 2}}
|
||||
db.set("glibc 2.42-1\n", "")
|
||||
k, states, ev, clock, _ := newTestKeeper(db)
|
||||
k.refreshInstalled(t0)
|
||||
inst := states.store(StateInstalled)
|
||||
inst.Writes()
|
||||
ev.Take()
|
||||
|
||||
db.set(bigDB(5000), "")
|
||||
clock.Advance(time.Minute)
|
||||
k.signal(LockReleased)
|
||||
if w := inst.Writes(); strings.Join(w, ",") != "put node-a.not-current" {
|
||||
t.Fatalf("writes: %v", w)
|
||||
}
|
||||
if v := installedOf(t, inst, "node-a"); len(v.Packages) != 1 {
|
||||
t.Fatalf("the last value is kept: %d packages", len(v.Packages))
|
||||
}
|
||||
if e := ev.Take(); len(e) != 0 {
|
||||
t.Errorf("nothing was written and it was said: %+v", e)
|
||||
}
|
||||
e, _ := inst.Get("node-a.not-current")
|
||||
var nc NotCurrent
|
||||
json.Unmarshal(e.Value, &nc)
|
||||
if nc.Bytes <= ValueCap || !nc.Since.Equal(t0.Add(time.Minute)) {
|
||||
t.Fatalf("not-current: %+v", nc)
|
||||
}
|
||||
// still over: the mark is written once.
|
||||
db.set(bigDB(5001), "")
|
||||
clock.Advance(time.Minute)
|
||||
k.signal(LockReleased)
|
||||
if w := inst.Writes(); len(w) != 0 {
|
||||
t.Errorf("the mark was written again: %v", w)
|
||||
}
|
||||
// back under the cap: written, and the mark deleted.
|
||||
db.set("glibc 2.42-1\nzsh 5.9-1\n", "")
|
||||
k.signal(LockReleased)
|
||||
if w := inst.Writes(); strings.Join(w, ",") != "put node-a,delete node-a.not-current" {
|
||||
t.Errorf("writes: %v", w)
|
||||
}
|
||||
}
|
||||
|
||||
// The mark goes too when the node's packages come back to what the key holds: the key is current again.
|
||||
func TestTheMarkGoesWhenThePackagesComeBackToTheKey(t *testing.T) {
|
||||
db := &pacmanDB{sl: syncDBs, checkup: Ran{Status: 2}}
|
||||
db.set("glibc 2.42-1\n", "")
|
||||
k, states, _, _, _ := newTestKeeper(db)
|
||||
k.refreshInstalled(t0)
|
||||
db.set(bigDB(5000), "")
|
||||
k.signal(LockReleased)
|
||||
inst := states.store(StateInstalled)
|
||||
inst.Writes()
|
||||
db.set("glibc 2.42-1\n", "")
|
||||
k.signal(LockReleased)
|
||||
if w := inst.Writes(); strings.Join(w, ",") != "delete node-a.not-current" {
|
||||
t.Errorf("writes: %v", w)
|
||||
}
|
||||
}
|
||||
|
||||
// The loop: a released lock is read, a hook runs after it with what it changed, a watched state's change reaches
|
||||
// its hook, and a wake comes at the time it asked.
|
||||
func TestTheLoopRunsTheHooksOnItsOneGoroutine(t *testing.T) {
|
||||
db := &pacmanDB{sl: syncDBs, checkup: Ran{Status: 2}}
|
||||
db.set("glibc 2.42-1\n", "")
|
||||
k, states, _, clock, w := newTestKeeper(db)
|
||||
var mu sync.Mutex
|
||||
var seen []string
|
||||
say := func(s string) { mu.Lock(); seen = append(seen, s); mu.Unlock() }
|
||||
k.Hooks = Hooks{
|
||||
Start: []func(*Keeper) error{func(*Keeper) error { say("start"); return nil }},
|
||||
AfterTransaction: []func(*Keeper, Transaction) error{func(_ *Keeper, tr Transaction) error {
|
||||
say(fmt.Sprintf("transaction changed=%v", tr.Changed != nil))
|
||||
return nil
|
||||
}},
|
||||
Changes: []StateHook{{State: StateAsks, Key: "node-a.*", Handle: func(_ *Keeper, c stdio.StateChange) error {
|
||||
say(fmt.Sprintf("%s %s current=%v", c.Op, c.Key, c.Current))
|
||||
return nil
|
||||
}}},
|
||||
Wakes: []Wake{func(_ *Keeper, now time.Time) time.Time {
|
||||
if now.Equal(t0) {
|
||||
return t0.Add(time.Hour)
|
||||
}
|
||||
say("wake " + now.Sub(t0).String())
|
||||
return time.Time{}
|
||||
}},
|
||||
}
|
||||
states.store(StateAsks).Put("node-a.ask-1", Ask{ID: "ask-1"})
|
||||
states.store(StateAsks).Put("node-b.ask-2", Ask{ID: "ask-2"})
|
||||
ctx, cancel := context.WithCancel(context.Background())
|
||||
done := make(chan error, 1)
|
||||
go func() { done <- k.Run(ctx) }()
|
||||
waitFor(t, &mu, &seen, "start")
|
||||
db.set("glibc 2.42-1\nzsh 5.9-1\n", "")
|
||||
w.c <- LockReleased
|
||||
waitFor(t, &mu, &seen, "transaction changed=true")
|
||||
states.store(StateAsks).Put("node-a.ask-3", Ask{ID: "ask-3"})
|
||||
waitFor(t, &mu, &seen, "put node-a.ask-3 current=false")
|
||||
clock.Advance(time.Hour)
|
||||
waitFor(t, &mu, &seen, "wake 1h0m0s")
|
||||
cancel()
|
||||
if err := <-done; err != context.Canceled {
|
||||
t.Errorf("Run ended with %v", err)
|
||||
}
|
||||
mu.Lock()
|
||||
defer mu.Unlock()
|
||||
if strings.Join(seen, "|") != "put node-a.ask-1 current=true|start|transaction changed=true|put node-a.ask-3 current=false|wake 1h0m0s" &&
|
||||
strings.Join(seen, "|") != "start|put node-a.ask-1 current=true|transaction changed=true|put node-a.ask-3 current=false|wake 1h0m0s" {
|
||||
t.Errorf("seen: %v", seen)
|
||||
}
|
||||
}
|
||||
|
||||
func waitFor(t *testing.T, mu *sync.Mutex, seen *[]string, want string) {
|
||||
t.Helper()
|
||||
for deadline := time.Now().Add(5 * time.Second); time.Now().Before(deadline); time.Sleep(time.Millisecond) {
|
||||
mu.Lock()
|
||||
for _, s := range *seen {
|
||||
if s == want {
|
||||
mu.Unlock()
|
||||
return
|
||||
}
|
||||
}
|
||||
mu.Unlock()
|
||||
}
|
||||
t.Fatalf("never saw %q", want)
|
||||
}
|
||||
|
||||
// The watch ending ends the keeper, loudly.
|
||||
func TestTheKeeperStopsWhenTheWatchEnds(t *testing.T) {
|
||||
db := &pacmanDB{sl: syncDBs, checkup: Ran{Status: 2}}
|
||||
k, _, _, _, w := newTestKeeper(db)
|
||||
close(w.c)
|
||||
if err := k.Run(context.Background()); err == nil || !strings.Contains(err.Error(), "watch ended") {
|
||||
t.Errorf("Run: %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
// The machine's watch: the lock deleted and the mirror list written or moved in are each a signal; anything else
|
||||
// in those directories is none.
|
||||
func TestTheWatchSeesTheLockReleasedAndTheMirrorListRewritten(t *testing.T) {
|
||||
lockDir, mirrorDir := t.TempDir(), t.TempDir()
|
||||
w, err := WatchMachine(lockDir, LockName, mirrorDir, MirrorName)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
next := func() Signal {
|
||||
select {
|
||||
case s := <-w.Signals():
|
||||
return s
|
||||
case <-time.After(200 * time.Millisecond):
|
||||
return 0
|
||||
}
|
||||
}
|
||||
lock := filepath.Join(lockDir, LockName)
|
||||
os.WriteFile(lock, nil, 0o644)
|
||||
os.WriteFile(filepath.Join(lockDir, "other"), nil, 0o644)
|
||||
os.Remove(filepath.Join(lockDir, "other"))
|
||||
if s := next(); s != 0 {
|
||||
t.Fatalf("a lock taken, and another file, are no signal: %v", s)
|
||||
}
|
||||
os.Remove(lock)
|
||||
if s := next(); s != LockReleased {
|
||||
t.Fatalf("the lock deleted: %v", s)
|
||||
}
|
||||
os.WriteFile(filepath.Join(mirrorDir, MirrorName), []byte("Server = x\n"), 0o644)
|
||||
if s := next(); s != MirrorsRewritten {
|
||||
t.Fatalf("the list written: %v", s)
|
||||
}
|
||||
tmp := filepath.Join(mirrorDir, "mirrorlist.new")
|
||||
os.WriteFile(tmp, []byte("Server = y\n"), 0o644)
|
||||
if s := next(); s != 0 {
|
||||
t.Fatalf("another file written: %v", s)
|
||||
}
|
||||
os.Rename(tmp, filepath.Join(mirrorDir, MirrorName))
|
||||
if s := next(); s != MirrorsRewritten {
|
||||
t.Fatalf("the list moved in: %v", s)
|
||||
}
|
||||
w.Close()
|
||||
for deadline := time.After(2 * time.Second); ; {
|
||||
select {
|
||||
case _, open := <-w.Signals():
|
||||
if !open {
|
||||
return
|
||||
}
|
||||
case <-deadline:
|
||||
t.Fatal("closing the watch did not end its signals")
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -22,6 +22,8 @@ import (
|
||||
"os/exec"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
)
|
||||
|
||||
// Ran is what one command did: its output, its exit status, and why it never ran to an answer.
|
||||
@@ -105,6 +107,17 @@ type Machine struct {
|
||||
ReadFile func(path string) ([]byte, error)
|
||||
Now func() time.Time
|
||||
Sleep func(time.Duration)
|
||||
// Node is the node this bundle runs on (the runtime's MESH_NODE), and States reads the module's states and
|
||||
// those it reads (the manifest's `reads`), for the verbs that answer from the bus (novox/hq ADR 0507 §6).
|
||||
Node string
|
||||
States func(name string) StateReader
|
||||
}
|
||||
|
||||
// StateReader is a state as a verb reads it: stdio.State, or a fake in a test.
|
||||
type StateReader interface {
|
||||
Get(key string) (*stdio.StateEntry, error)
|
||||
Keys() ([]string, error)
|
||||
History(key string) ([]stdio.StateRevision, error)
|
||||
}
|
||||
|
||||
// ThisMachine is the machine the runtime launched this bundle on.
|
||||
@@ -117,7 +130,8 @@ func ThisMachine() *Machine {
|
||||
if account == "" {
|
||||
account = user
|
||||
}
|
||||
return &Machine{Run: ExecRunner, UID: os.Getuid(), User: user, Account: account, ReadFile: os.ReadFile, Now: time.Now, Sleep: time.Sleep}
|
||||
return &Machine{Run: ExecRunner, UID: os.Getuid(), User: user, Account: account, ReadFile: os.ReadFile, Now: time.Now, Sleep: time.Sleep,
|
||||
Node: os.Getenv("MESH_NODE"), States: func(name string) StateReader { return stdio.State(name) }}
|
||||
}
|
||||
|
||||
// Out runs a command that only reads, and fails with what went wrong named.
|
||||
|
||||
@@ -10,6 +10,7 @@ import (
|
||||
"context"
|
||||
"fmt"
|
||||
"os"
|
||||
"regexp"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
)
|
||||
@@ -20,6 +21,19 @@ const binaryName = "pacman-tools"
|
||||
func bg() context.Context { return context.Background() }
|
||||
|
||||
func main() {
|
||||
// `pacman-tools keeper` is the module's process (keeper.go), `root` its root helper (root.go); without a
|
||||
// word it serves the tools.
|
||||
if len(os.Args) > 1 {
|
||||
switch os.Args[1] {
|
||||
case "keeper":
|
||||
os.Exit(keeper())
|
||||
case "root":
|
||||
// the root helper (root.go, ADR 0510), and the command of a rollback's unit it starts
|
||||
os.Exit(rootHelper())
|
||||
case "root-rollback":
|
||||
os.Exit(rootRollback(os.Args[2:]))
|
||||
}
|
||||
}
|
||||
// An empty name serves as the module the runtime names (MESH_SERVED_MODULE): pacman.
|
||||
if err := stdio.Serve("", tools(ThisMachine())); err != nil {
|
||||
fmt.Fprintln(os.Stderr, err)
|
||||
@@ -33,8 +47,22 @@ var (
|
||||
return map[string]any{"type": "integer", "description": fmt.Sprintf("at most this many (default %d, at most %d)", def, most)}
|
||||
}
|
||||
sinceArg = map[string]any{"type": "string", "description": "a day, YYYY-MM-DD"}
|
||||
nodeArg = map[string]any{"type": "string", "description": "a node's name"}
|
||||
nodeName = regexp.MustCompile(`^[A-Za-z0-9_-]+$`)
|
||||
)
|
||||
|
||||
// nodeOf is the node argument, empty when absent; a node's name is one key token.
|
||||
func nodeOf(args map[string]any) (string, error) {
|
||||
n, err := text(args, "node", false)
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
if n != "" && !nodeName.MatchString(n) {
|
||||
return "", fmt.Errorf("node %q is not a node's name", n)
|
||||
}
|
||||
return n, nil
|
||||
}
|
||||
|
||||
func tools(m *Machine) []stdio.Tool {
|
||||
return []stdio.Tool{
|
||||
{
|
||||
@@ -66,15 +94,22 @@ func tools(m *Machine) []stdio.Tool {
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "pacman_installed",
|
||||
Description: "Installed packages with version, why each is installed (explicit or dependency) and whether it is foreign (in no repository); narrowed by name, reason or foreign; with totals.",
|
||||
Name: "pacman_installed",
|
||||
Description: "A node's installed packages, from the bus (ADR 0507 §2): version, why each is installed (explicit or dependency), its repository, " +
|
||||
"or foreign (in no repository); narrowed by name, reason or foreign; with totals, the time of the transaction they describe and whether they are current. " +
|
||||
"This node's by default. (r)",
|
||||
Input: schema(map[string]any{
|
||||
"node": nodeArg,
|
||||
"match": map[string]any{"type": "string", "description": "only names holding this"},
|
||||
"reason": map[string]any{"type": "string", "enum": []string{"explicit", "dependency"}},
|
||||
"foreign": map[string]any{"type": "boolean", "description": "only foreign packages"},
|
||||
"limit": limitArg(5000, 20000),
|
||||
}),
|
||||
Run: func(args map[string]any) (any, error) {
|
||||
node, err := nodeOf(args)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
match, err := text(args, "match", false)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
@@ -94,7 +129,29 @@ func tools(m *Machine) []stdio.Tool {
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return m.Installed(match, reason, foreign, n)
|
||||
return m.InstalledFrom(node, match, reason, foreign, n)
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "pacman_packages",
|
||||
Description: "Every node's packages at once, from the bus (ADR 0507 §6): with package, that package on every node (version, reason, repository " +
|
||||
"or foreign, the modules declaring it); without, per node its counts, the packages installed explicitly that no module assigned to it declares, " +
|
||||
"and its foreign packages apart, then the packages whose versions differ between nodes and the packages on some nodes only. " +
|
||||
"A node whose packages are over the bus's cap is marked not current. No machine is asked. (r)",
|
||||
Input: schema(map[string]any{
|
||||
"package": map[string]any{"type": "string", "description": "one package, on every node"},
|
||||
"limit": limitArg(500, 20000),
|
||||
}),
|
||||
Run: func(args map[string]any) (any, error) {
|
||||
p, err := text(args, "package", false)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
n, err := whole(args, "limit", 500, 1, 20000)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return m.Packages(p, n)
|
||||
},
|
||||
},
|
||||
{
|
||||
@@ -126,18 +183,29 @@ func tools(m *Machine) []stdio.Tool {
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "pacman_updates",
|
||||
Description: "What a full upgrade would change, each package from and to (checkupdates: the repositories are asked into a copy of their databases, so asking never sets up a partial upgrade). Needs the network.",
|
||||
Input: schema(map[string]any{}),
|
||||
Run: func(map[string]any) (any, error) { return m.Updates() },
|
||||
Name: "pacman_updates",
|
||||
Description: "What a full upgrade would change, from the bus (ADR 0507 §4), on one node or on every node: each package from and to, whether the " +
|
||||
"node reached its mirrors (false: the list is the last it could confirm), and when the node's list last changed. The keeper reads it with " +
|
||||
"checkupdates at its start, after every transaction, after every mirror refresh and once a day. (r)",
|
||||
Input: schema(map[string]any{"node": nodeArg}),
|
||||
Run: func(args map[string]any) (any, error) {
|
||||
node, err := nodeOf(args)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return m.UpdatesFor(node)
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "pacman_upgrade",
|
||||
Description: "Start a full system upgrade (pacman -Syu --noconfirm, through sudo -n) as a transient unit of its own that outlives the call, " +
|
||||
"answering at once with the unit and the distribution's news since the last upgrade — read the news first. Given that unit, " +
|
||||
"answer whether it is running, finished and succeeded, with the tail of its log.",
|
||||
Description: "Ask for a full system upgrade of a node (this one unless named): the ask is written to the bus, where that node's keeper " +
|
||||
"takes it and runs it as a run: the package record, the kernel check, the transaction as a unit of its own, and the after-upgrade " +
|
||||
"check (ADR 0508 §8). Answers at once with the ask and the distribution's news since the last upgrade — read the news first; " +
|
||||
"pacman_upgrades with the node shows the run that names the ask. Given a transaction unit, answers whether it is running, " +
|
||||
"finished and succeeded, with the tail of its log.",
|
||||
Input: schema(map[string]any{
|
||||
"unit": map[string]any{"type": "string", "description": "a unit this tool started, to read how it goes (optional)"},
|
||||
"node": map[string]any{"type": "string", "description": "the node to upgrade (default: this one)"},
|
||||
"unit": map[string]any{"type": "string", "description": "a transaction unit, to read how it goes (optional)"},
|
||||
"lines": limitArg(60, 400),
|
||||
}),
|
||||
Run: func(args map[string]any) (any, error) {
|
||||
@@ -149,15 +217,127 @@ func tools(m *Machine) []stdio.Tool {
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return m.Upgrade(unit, n)
|
||||
if unit != "" {
|
||||
return m.UpgradeStatus(unit, n)
|
||||
}
|
||||
node, err := text(args, "node", false)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
if node == "" {
|
||||
node = thisNode()
|
||||
}
|
||||
a, err := AskUpgrade(upgradeState(StateAsks), node, m.Now())
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return map[string]any{"asked": a, "follow": "pacman_upgrades with node " + node + " lists the run that names this ask", "news": m.News("")}, nil
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "pacman_upgrade_health",
|
||||
Description: "Whether the weekly full upgrade (Saturday 04:00 local time, ADR 0477) last passed, as the module's health check: unhealthy " +
|
||||
"with the exit and the last error of the journal when it failed, healthy while it runs, after it passed and before it ever ran. (r)",
|
||||
Description: "This node's upgrades, read from the bus, as the module's health check (ADR 0508 §11): unhealthy while no upgrade-order " +
|
||||
"is set, while the node's latest run is refused, failed or did not pass its after-upgrade check, hung past two hours, or held a " +
|
||||
"kernel, and while the week's round stands still at this node or at the node it waits for, naming it and since when. (r)",
|
||||
Input: schema(map[string]any{}),
|
||||
Run: func(map[string]any) (any, error) { return m.UpgradeHealth(), nil },
|
||||
Run: func(map[string]any) (any, error) {
|
||||
return m.UpgradeHealth(), nil
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "pacman_upgrades",
|
||||
Description: "Every full upgrade and rollback of every node, from the bus (ADR 0508 §1): each run's node, kind, what started it, " +
|
||||
"its round, when it began and ended, its outcome (refused, finished, failed or running) and whether its after-upgrade check passed; " +
|
||||
"narrowed to a node and to days, newest first. With a node and a run, that run whole, every step with its time.",
|
||||
Input: schema(map[string]any{
|
||||
"node": map[string]any{"type": "string", "description": "only this node's runs"},
|
||||
"from": sinceArg,
|
||||
"until": map[string]any{"type": "string", "description": "a day, YYYY-MM-DD, the last one listed"},
|
||||
"run": map[string]any{"type": "string", "description": "one run, by its id, with node"},
|
||||
"limit": limitArg(50, 1000),
|
||||
}),
|
||||
Run: func(args map[string]any) (any, error) {
|
||||
var v [4]string
|
||||
for i, k := range []string{"node", "from", "until", "run"} {
|
||||
s, err := text(args, k, false)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
v[i] = s
|
||||
}
|
||||
n, err := whole(args, "limit", 50, 1, 1000)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return ListUpgrades(upgradeState(StateRuns), v[0], v[1], v[2], v[3], n)
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "pacman_rollback",
|
||||
Description: "Return a node to the package record a run took (ADR 0508 §7). Without apply it says what that would change: each package " +
|
||||
"from its version now to the record's and where that version comes from (the node's package cache, the Arch Linux Archive's copy " +
|
||||
"of that exact version, or the archive's repository on the record's date), the packages installed since (removed only with " +
|
||||
"remove_added), and no foreign package. With apply it asks the node's keeper to carry it out as a run of its own, followed by " +
|
||||
"the after-upgrade check; pacman_upgrades follows it. This node's by default.",
|
||||
Input: schema(map[string]any{
|
||||
"node": nodeArg,
|
||||
"run": map[string]any{"type": "string", "description": "the run whose record to return to (pacman_upgrades lists them)"},
|
||||
"remove_added": map[string]any{"type": "boolean", "description": "also remove the packages installed since the record"},
|
||||
"apply": map[string]any{"type": "boolean", "description": "carry it out"},
|
||||
}, "run"),
|
||||
Run: func(args map[string]any) (any, error) {
|
||||
node, err := nodeOf(args)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
run, err := text(args, "run", true)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
if !nodeName.MatchString(run) {
|
||||
return nil, fmt.Errorf("run %q is not a run's id", run)
|
||||
}
|
||||
remove, err := flag(args, "remove_added")
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
apply, err := flag(args, "apply")
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return m.Rollback(node, run, remove, apply)
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "pacman_round",
|
||||
Description: "Where the week's upgrade rounds stand (ADR 0508 §5, §6): each round's order and opening, and every node's state in it: " +
|
||||
"waiting (for the node named), turn, stalled (its turn came longer than round-start-minutes ago), running, passed, failed " +
|
||||
"(the round stops there) or refused, with since when. With pass, a node and a reason, asks the operator on their channel to let " +
|
||||
"the round go on past that node; it goes on only on the operator's Approve.",
|
||||
Input: schema(map[string]any{
|
||||
"round": map[string]any{"type": "string", "description": "one round, as 2026-W43 (default: the newest)"},
|
||||
"pass": map[string]any{"type": "string", "description": "a node to ask the operator to pass"},
|
||||
"reason": map[string]any{"type": "string", "description": "why the node should be passed: required with pass, and shown to the operator"},
|
||||
"limit": limitArg(4, 52),
|
||||
}),
|
||||
Run: func(args map[string]any) (any, error) {
|
||||
var v [3]string
|
||||
for i, k := range []string{"round", "pass", "reason"} {
|
||||
s, err := text(args, k, false)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
v[i] = s
|
||||
}
|
||||
if v[1] != "" {
|
||||
return AskPass(upgradeState(StateRounds), upgradeState(StateRuns), upgradeState(StateAsks), v[0], v[1], v[2], m.Now())
|
||||
}
|
||||
n, err := whole(args, "limit", 4, 1, 52)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return ListRounds(upgradeState(StateRounds), upgradeState(StateRuns), v[0], m.Now(), roundStartMinutes(), n)
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "pacman_orphans",
|
||||
@@ -258,10 +438,17 @@ func tools(m *Machine) []stdio.Tool {
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "pacman_foreign",
|
||||
Description: "Installed packages that no repository this machine syncs carries (pacman -Qm): built from the AUR or by hand, which the mesh cannot install.",
|
||||
Input: schema(map[string]any{}),
|
||||
Run: func(map[string]any) (any, error) { return m.Foreign() },
|
||||
Name: "pacman_foreign",
|
||||
Description: "A node's packages that no repository it syncs carries, from the bus: built from the AUR or by hand, which the mesh neither " +
|
||||
"upgrades nor removes (ADR 0507 §7). This node's by default. (r)",
|
||||
Input: schema(map[string]any{"node": nodeArg}),
|
||||
Run: func(args map[string]any) (any, error) {
|
||||
node, err := nodeOf(args)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return m.ForeignFrom(node)
|
||||
},
|
||||
},
|
||||
{
|
||||
Name: "pacman_news",
|
||||
|
||||
@@ -101,53 +101,3 @@ func TestTheReflectorPackageIsDeclaredBeforeTheFileItShips(t *testing.T) {
|
||||
t.Fatalf("a package's file and timer come after the package: %v", order)
|
||||
}
|
||||
}
|
||||
|
||||
// The weekly upgrade (issue 1574, ADR 0477 clause 5): a timer on Saturday 04:00 local time, persistent so a
|
||||
// node that was off upgrades at its next start, the same moment on every node (no random delay), starting a
|
||||
// oneshot service that runs the very upgrade pacman_upgrade runs.
|
||||
func TestTheWeeklyUpgradeIsATimerOnSaturdayFourPersistentRunningTheToolsUpgrade(t *testing.T) {
|
||||
m := manifest(t)
|
||||
timer := m.resource(t, "upgrade-timer")
|
||||
service := m.resource(t, "upgrade-service")
|
||||
if timer["path"] != "/etc/systemd/system/mesh-pacman-weekly-upgrade.timer" || service["path"] != "/etc/systemd/system/"+WeeklyUnit {
|
||||
t.Fatalf("paths: %v %v", timer["path"], service["path"])
|
||||
}
|
||||
tt, st := templateText(t, timer), templateText(t, service)
|
||||
for _, want := range []string{"\nOnCalendar=Sat *-*-* 04:00:00\n", "\nPersistent=true\n", "\nWantedBy=timers.target\n"} {
|
||||
if !strings.Contains(tt, want) {
|
||||
t.Errorf("the timer lacks %q:\n%s", strings.TrimSpace(want), tt)
|
||||
}
|
||||
}
|
||||
for _, not := range []string{"RandomizedDelaySec", "OnBootSec", "OnUnitActiveSec", "UTC"} {
|
||||
if strings.Contains(tt, not) {
|
||||
t.Errorf("the timer says %s: the same local moment on every node, and no other moment", not)
|
||||
}
|
||||
}
|
||||
if strings.Count(tt, "OnCalendar=") != 1 {
|
||||
t.Errorf("one moment only:\n%s", tt)
|
||||
}
|
||||
if !strings.Contains(st, "\nType=oneshot\n") {
|
||||
t.Errorf("the service is a oneshot:\n%s", st)
|
||||
}
|
||||
if want := "\nExecStart=/usr/bin/pacman " + strings.Join(UpgradeArgs, " ") + "\n"; !strings.Contains(st, want) {
|
||||
t.Errorf("the service does not run the upgrade pacman_upgrade runs (%q):\n%s", want, st)
|
||||
}
|
||||
if strings.Contains(st, "reboot") || strings.Contains(tt, "reboot") || strings.Contains(st, "systemctl") {
|
||||
t.Error("the mesh restarts no machine for an upgrade")
|
||||
}
|
||||
r := m.resource(t, "weekly-upgrade")
|
||||
if r["type"] != "service" || r["unit"] != "mesh-pacman-weekly-upgrade.timer" || r["state"] != "running" || r["boot"] != "enabled" {
|
||||
t.Fatalf("the timer runs and is enabled: %v", r)
|
||||
}
|
||||
h, _ := r["health"].(map[string]any)
|
||||
if h["kind"] != "tool" || h["tool"] != "pacman_upgrade_health" {
|
||||
t.Errorf("the upgrade's outcome is the resource's health: %v", r["health"])
|
||||
}
|
||||
order := map[string]int{}
|
||||
for i, x := range m.Resources {
|
||||
order[x["id"].(string)] = i
|
||||
}
|
||||
if order["upgrade-service"] > order["weekly-upgrade"] || order["upgrade-timer"] > order["weekly-upgrade"] || order["package"] > order["upgrade-service"] {
|
||||
t.Errorf("the units are placed before the timer is started: %v", order)
|
||||
}
|
||||
}
|
||||
|
||||
@@ -0,0 +1,173 @@
|
||||
package main
|
||||
|
||||
// The after-upgrade check, its second part (novox/hq ADR 0508 §4): the processes still mapping a file the upgrade
|
||||
// replaced, named by their unit or container and their executable's path, never their command line or
|
||||
// environment; and whether every DKMS module is built for every installed kernel. Reading another process's
|
||||
// mappings needs root, so the root helper reads them (ADR 0510's `maps`), with MappedDeleted below.
|
||||
|
||||
import (
|
||||
"os"
|
||||
"path"
|
||||
"regexp"
|
||||
"sort"
|
||||
"strconv"
|
||||
"strings"
|
||||
)
|
||||
|
||||
// ProcFS is the process table as MappedDeleted reads it: the machine's /proc, or a stand-in in a test.
|
||||
type ProcFS interface {
|
||||
ReadDir(dir string) ([]string, error)
|
||||
ReadFile(name string) ([]byte, error)
|
||||
Readlink(name string) (string, error)
|
||||
}
|
||||
|
||||
// osProc is the machine's own /proc.
|
||||
type osProc struct{}
|
||||
|
||||
func (osProc) ReadDir(dir string) ([]string, error) {
|
||||
entries, err := os.ReadDir(dir)
|
||||
var names []string
|
||||
for _, e := range entries {
|
||||
names = append(names, e.Name())
|
||||
}
|
||||
return names, err
|
||||
}
|
||||
func (osProc) ReadFile(name string) ([]byte, error) { return os.ReadFile(name) }
|
||||
func (osProc) Readlink(name string) (string, error) { return os.Readlink(name) }
|
||||
|
||||
// replacedPrefixes are where a package's files are: a mapping deleted elsewhere (a temporary file, shared
|
||||
// memory) is no file an upgrade replaced.
|
||||
var replacedPrefixes = []string{"/usr/", "/opt/"}
|
||||
|
||||
var containerScope = regexp.MustCompile(`(?:docker|libpod|crio|cri-containerd)-([0-9a-f]{12})[0-9a-f]*\.scope$`)
|
||||
|
||||
// cgroupOwner names a process's unit or container from its cgroup line `0::/system.slice/x.service`.
|
||||
func cgroupOwner(cgroup string) (unit, container string) {
|
||||
for _, l := range strings.Split(cgroup, "\n") {
|
||||
_, p, ok := strings.Cut(l, "::")
|
||||
if !ok {
|
||||
continue
|
||||
}
|
||||
parts := strings.Split(strings.Trim(p, "/"), "/")
|
||||
for i := len(parts) - 1; i >= 0; i-- {
|
||||
if g := containerScope.FindStringSubmatch(parts[i]); g != nil {
|
||||
return "", g[1]
|
||||
}
|
||||
if i > 0 && parts[i-1] == "docker" && len(parts[i]) >= 12 {
|
||||
return "", parts[i][:12]
|
||||
}
|
||||
}
|
||||
for i := len(parts) - 1; i >= 0; i-- {
|
||||
if strings.HasSuffix(parts[i], ".service") || strings.HasSuffix(parts[i], ".scope") {
|
||||
return parts[i], ""
|
||||
}
|
||||
}
|
||||
}
|
||||
return "", ""
|
||||
}
|
||||
|
||||
// MappedDeleted reads the process table under procRoot for processes mapping a deleted package file, each named
|
||||
// once by its unit or container and its executable, sorted.
|
||||
func MappedDeleted(fs ProcFS, procRoot string) ([]Mapped, error) {
|
||||
names, err := fs.ReadDir(procRoot)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
seen := map[Mapped]bool{}
|
||||
out := []Mapped{}
|
||||
for _, name := range names {
|
||||
if _, err := strconv.Atoi(name); err != nil {
|
||||
continue
|
||||
}
|
||||
dir := path.Join(procRoot, name)
|
||||
maps, err := fs.ReadFile(dir + "/maps")
|
||||
if err != nil {
|
||||
continue // the process ended meanwhile
|
||||
}
|
||||
replaced := false
|
||||
for _, l := range strings.Split(string(maps), "\n") {
|
||||
if !strings.HasSuffix(l, " (deleted)") {
|
||||
continue
|
||||
}
|
||||
f := strings.Fields(strings.TrimSuffix(l, " (deleted)"))
|
||||
if len(f) < 6 {
|
||||
continue
|
||||
}
|
||||
for _, p := range replacedPrefixes {
|
||||
if strings.HasPrefix(f[5], p) {
|
||||
replaced = true
|
||||
}
|
||||
}
|
||||
}
|
||||
if !replaced {
|
||||
continue
|
||||
}
|
||||
exe, _ := fs.Readlink(dir + "/exe")
|
||||
cgroup, _ := fs.ReadFile(dir + "/cgroup")
|
||||
unit, container := cgroupOwner(string(cgroup))
|
||||
m := Mapped{Unit: unit, Container: container, Executable: strings.TrimSuffix(exe, " (deleted)")}
|
||||
if !seen[m] {
|
||||
seen[m] = true
|
||||
out = append(out, m)
|
||||
}
|
||||
}
|
||||
sort.Slice(out, func(i, j int) bool {
|
||||
a, b := out[i], out[j]
|
||||
if a.Unit+a.Container != b.Unit+b.Container {
|
||||
return a.Unit+a.Container < b.Unit+b.Container
|
||||
}
|
||||
return a.Executable < b.Executable
|
||||
})
|
||||
return out, nil
|
||||
}
|
||||
|
||||
// unbuiltDKMS is every DKMS module built for some kernel and not for an installed one its own bound allows.
|
||||
func (m *Machine) unbuiltDKMS() ([]DKMSBuild, error) {
|
||||
modules, err := m.DKMSModules()
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
kernels, err := m.installedKernels()
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
out := []DKMSBuild{}
|
||||
for _, mod := range modules {
|
||||
if len(mod.Kernels) == 0 {
|
||||
continue // added and never built: not a module this machine uses
|
||||
}
|
||||
bound := m.dkmsBound(mod)
|
||||
for _, k := range kernels {
|
||||
if contains(mod.Kernels, k) {
|
||||
continue
|
||||
}
|
||||
if held, _ := heldBy(bound, k); held != "" {
|
||||
continue
|
||||
}
|
||||
out = append(out, DKMSBuild{Module: mod.Module, Version: mod.Version, Kernel: k})
|
||||
}
|
||||
}
|
||||
return out, nil
|
||||
}
|
||||
|
||||
// afterCheckII fills the check's second part and judges it: a DKMS module not built for a kernel fails it;
|
||||
// a process still mapping a replaced file is said and does not.
|
||||
func (u *Upgrades) afterCheckII(c *Check) {
|
||||
if u.Root == nil {
|
||||
c.MappedError = "no root helper to read the processes' mapped files"
|
||||
} else if mapped, err := u.Root.Maps(); err != nil {
|
||||
c.MappedError = err.Error()
|
||||
} else {
|
||||
c.Mapped = mapped
|
||||
}
|
||||
unbuilt, err := u.Machine.unbuiltDKMS()
|
||||
if err != nil {
|
||||
c.DKMSError = err.Error()
|
||||
c.Passed = false
|
||||
return
|
||||
}
|
||||
c.Unbuilt = unbuilt
|
||||
if len(unbuilt) > 0 {
|
||||
c.Passed = false
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,131 @@
|
||||
package main
|
||||
|
||||
// The after-upgrade check, its second part (novox/hq ADR 0508 §4; acceptance 1841/AC5): processes still mapping a
|
||||
// replaced file, by unit or container and executable, never a command line or environment; DKMS built for every
|
||||
// installed kernel.
|
||||
|
||||
import (
|
||||
"errors"
|
||||
"strings"
|
||||
"testing"
|
||||
)
|
||||
|
||||
// standInProc is a process table in memory; it records every file read, so nothing else is read.
|
||||
type standInProc struct {
|
||||
files map[string]string
|
||||
links map[string]string
|
||||
read []string
|
||||
}
|
||||
|
||||
func (p *standInProc) ReadDir(dir string) ([]string, error) {
|
||||
seen := map[string]bool{}
|
||||
var names []string
|
||||
for f := range p.files {
|
||||
if rest, ok := strings.CutPrefix(f, dir+"/"); ok {
|
||||
n, _, _ := strings.Cut(rest, "/")
|
||||
if !seen[n] {
|
||||
seen[n] = true
|
||||
names = append(names, n)
|
||||
}
|
||||
}
|
||||
}
|
||||
return names, nil
|
||||
}
|
||||
|
||||
func (p *standInProc) ReadFile(name string) ([]byte, error) {
|
||||
p.read = append(p.read, name)
|
||||
if s, ok := p.files[name]; ok {
|
||||
return []byte(s), nil
|
||||
}
|
||||
return nil, errNoFile
|
||||
}
|
||||
|
||||
func (p *standInProc) Readlink(name string) (string, error) {
|
||||
if s, ok := p.links[name]; ok {
|
||||
return s, nil
|
||||
}
|
||||
return "", errNoFile
|
||||
}
|
||||
|
||||
const libcDeleted = "7f2a1c000000-7f2a1c028000 r--p 00000000 103:02 5513 /usr/lib/libc.so.6 (deleted)\n"
|
||||
|
||||
func TestAProcessMappingAReplacedFileIsNamedByItsUnitOrContainerAndExecutableOnly(t *testing.T) {
|
||||
p := &standInProc{files: map[string]string{
|
||||
"/proc/1/maps": libcDeleted,
|
||||
"/proc/1/cgroup": "0::/init.scope\n",
|
||||
"/proc/200/maps": "5600-5700 r-xp 00000000 103:02 77 /usr/bin/sshd\n" + strings.Replace(libcDeleted, "libc.so.6", "libcrypto.so.3", 1),
|
||||
"/proc/200/cgroup": "0::/system.slice/sshd.service\n",
|
||||
"/proc/201/maps": libcDeleted,
|
||||
"/proc/201/cgroup": "0::/system.slice/sshd.service\n",
|
||||
"/proc/300/maps": libcDeleted,
|
||||
"/proc/300/cgroup": "0::/system.slice/docker-0123456789abcdef0123456789abcdef.scope\n",
|
||||
"/proc/400/maps": "7f00-7f10 rw-s 00000000 00:1a 9 /dev/shm/made-up (deleted)\n",
|
||||
"/proc/400/cgroup": "0::/system.slice/made-up.service\n",
|
||||
"/proc/500/maps": "5600-5700 r-xp 00000000 103:02 77 /usr/bin/made-up\n",
|
||||
"/proc/500/cmdline": "made-up --password=secret",
|
||||
"/proc/500/environ": "TOKEN=secret",
|
||||
"/proc/self/maps": libcDeleted,
|
||||
}, links: map[string]string{
|
||||
"/proc/1/exe": "/usr/lib/systemd/systemd", "/proc/200/exe": "/usr/bin/sshd (deleted)", "/proc/201/exe": "/usr/bin/sshd (deleted)",
|
||||
"/proc/300/exe": "/usr/bin/made-up-server",
|
||||
}}
|
||||
got, err := MappedDeleted(p, "/proc")
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
var said []string
|
||||
for _, m := range got {
|
||||
said = append(said, m.Unit+"|"+m.Container+"|"+m.Executable)
|
||||
}
|
||||
if strings.Join(said, " ") != "|0123456789ab|/usr/bin/made-up-server init.scope||/usr/lib/systemd/systemd sshd.service||/usr/bin/sshd" {
|
||||
t.Errorf("%v", said)
|
||||
}
|
||||
for _, r := range p.read {
|
||||
if strings.HasSuffix(r, "/cmdline") || strings.HasSuffix(r, "/environ") {
|
||||
t.Errorf("read %s: a process is named by its unit and executable, never its command line or environment", r)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestADKMSModuleNotBuiltForAnInstalledKernelFailsTheCheck(t *testing.T) {
|
||||
m := checkMachine("", "7.2.9-arch1-1")
|
||||
table := map[string]Ran{
|
||||
"systemctl list-units --state=failed --plain --no-legend --no-pager": {},
|
||||
"pacman -Qq": {Stdout: "linux\nlinux-lts\n"},
|
||||
"pacman -Qlq -- linux": {Stdout: "/usr/lib/modules/7.2.9-arch1-1/vmlinuz\n"},
|
||||
"pacman -Qlq -- linux-lts": {Stdout: "/usr/lib/modules/6.18.54-2-lts/vmlinuz\n"},
|
||||
"dkms status": {Stdout: "zfs/2.4.4, 6.18.54-2-lts, x86_64: installed\nzfs/2.4.4, 7.2.8-arch1-2, x86_64: installed\n" +
|
||||
"made-up-driver/1.0, 6.18.54-2-lts, x86_64: installed\nxpad/0.4: added\n"},
|
||||
}
|
||||
m.Run = byLine(table, nil)
|
||||
read := m.ReadFile
|
||||
m.ReadFile = func(path string) ([]byte, error) {
|
||||
if path == "/usr/src/made-up-driver-1.0/dkms.conf" {
|
||||
return []byte("BUILD_EXCLUSIVE_KERNEL=\"^6\\.\"\n"), nil
|
||||
}
|
||||
return read(path)
|
||||
}
|
||||
u := checkUpgrades(t, m)
|
||||
u.Root = &fakeRoot{mapped: []Mapped{{Unit: "sshd.service", Executable: "/usr/bin/sshd"}}}
|
||||
c := u.AfterCheck(checkedRun(StepStarted, StepFinished), &Record{})
|
||||
if len(c.Unbuilt) != 1 || c.Unbuilt[0] != (DKMSBuild{Module: "zfs", Version: "2.4.4", Kernel: "7.2.9-arch1-1"}) || c.Passed {
|
||||
t.Errorf("zfs not built for the new kernel fails the check (made-up-driver excludes it, xpad was never built): %+v", c)
|
||||
}
|
||||
if len(c.Mapped) != 1 || c.Mapped[0].Unit != "sshd.service" {
|
||||
t.Errorf("mapped: %+v", c.Mapped)
|
||||
}
|
||||
|
||||
table["dkms status"] = Ran{Stdout: "zfs/2.4.4, 6.18.54-2-lts, x86_64: installed\nzfs/2.4.4, 7.2.9-arch1-1, x86_64: installed\n"}
|
||||
if c := u.AfterCheck(checkedRun(StepStarted, StepFinished), &Record{}); !c.Passed || len(c.Mapped) != 1 {
|
||||
t.Errorf("built for every kernel passes; a mapped file is said and does not fail it: %+v", c)
|
||||
}
|
||||
|
||||
u.Root = &fakeRoot{mapsErr: errors.New("the root helper's socket is not there")}
|
||||
if c := u.AfterCheck(checkedRun(StepStarted, StepFinished), &Record{}); !strings.Contains(c.MappedError, "socket") {
|
||||
t.Errorf("unread mappings are said: %+v", c)
|
||||
}
|
||||
table["dkms status"] = Ran{Status: 1, Stderr: "dkms: permission denied\n"}
|
||||
if c := u.AfterCheck(checkedRun(StepStarted, StepFinished), &Record{}); c.Passed || c.DKMSError == "" {
|
||||
t.Errorf("unread DKMS builds do not pass: %+v", c)
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,144 @@
|
||||
package main
|
||||
|
||||
// What the keeper reads from the machine into the bus (novox/hq ADR 0507 §2, §4, §7): the local package
|
||||
// database (`pacman -Q`, `pacman -Qeq`) with each package's repository from the sync databases (`pacman -Sl`),
|
||||
// and what a full upgrade would change (`checkupdates`). Every line is read against the shape of a package name
|
||||
// and a version, and a line that is not one is left out, so nothing but names, versions and repositories ever
|
||||
// reaches a value on the bus (ADR 0507 §1): never a command line, an environment or a message pacman printed.
|
||||
// None of these commands names a package, so none names a foreign one either.
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"regexp"
|
||||
"sort"
|
||||
"strings"
|
||||
)
|
||||
|
||||
var (
|
||||
// packageName is what pacman allows in a package's name (makepkg's lint: alphanumerics and @._+-).
|
||||
packageName = regexp.MustCompile(`^[a-z0-9@._+][a-z0-9@._+-]*$`)
|
||||
// packageVersion is an epoch, a version and a release: alphanumerics and ._+:~-.
|
||||
packageVersion = regexp.MustCompile(`^[A-Za-z0-9._+~][A-Za-z0-9._+:~-]*$`)
|
||||
// repoName is a repository's section name in pacman.conf.
|
||||
repoName = regexp.MustCompile(`^[A-Za-z0-9_-][A-Za-z0-9._-]*$`)
|
||||
)
|
||||
|
||||
// nameVersion reads a `<name> <version>` line, refusing anything else.
|
||||
func nameVersion(l string) (string, string, bool) {
|
||||
f := strings.Fields(l)
|
||||
if len(f) != 2 || !packageName.MatchString(f[0]) || !packageVersion.MatchString(f[1]) {
|
||||
return "", "", false
|
||||
}
|
||||
return f[0], f[1], true
|
||||
}
|
||||
|
||||
// ReadInstalled is the node's installed packages, sorted by name, each with its version, whether it was installed
|
||||
// explicitly, and its repository or RepoForeign; and how many lines of pacman's output were not a package and
|
||||
// were left out.
|
||||
func (m *Machine) ReadInstalled() ([]InstalledPackage, int, error) {
|
||||
all, err := m.Out("pacman", "-Q")
|
||||
if err != nil {
|
||||
return nil, 0, err
|
||||
}
|
||||
explicit, err := m.query("-Qeq")
|
||||
if err != nil {
|
||||
return nil, 0, err
|
||||
}
|
||||
synced, err := m.Out("pacman", "-Sl")
|
||||
if err != nil {
|
||||
return nil, 0, err
|
||||
}
|
||||
skipped := 0
|
||||
ex := map[string]bool{}
|
||||
for _, l := range lines(explicit) {
|
||||
if n := strings.TrimSpace(l); packageName.MatchString(n) {
|
||||
ex[n] = true
|
||||
}
|
||||
}
|
||||
// A package in two repositories is the first's: `pacman -Sl` lists them in pacman.conf's order.
|
||||
repo := map[string]string{}
|
||||
for _, l := range lines(synced) {
|
||||
f := strings.Fields(l)
|
||||
if len(f) < 3 || !repoName.MatchString(f[0]) || !packageName.MatchString(f[1]) {
|
||||
continue
|
||||
}
|
||||
if _, seen := repo[f[1]]; !seen {
|
||||
repo[f[1]] = f[0]
|
||||
}
|
||||
}
|
||||
pkgs := []InstalledPackage{}
|
||||
for _, l := range lines(all) {
|
||||
name, version, ok := nameVersion(l)
|
||||
if !ok {
|
||||
skipped++
|
||||
continue
|
||||
}
|
||||
p := InstalledPackage{Name: name, Version: version, Explicit: ex[name], Repo: repo[name]}
|
||||
if p.Repo == "" {
|
||||
p.Repo = RepoForeign
|
||||
}
|
||||
pkgs = append(pkgs, p)
|
||||
}
|
||||
sort.Slice(pkgs, func(i, j int) bool { return pkgs[i].Name < pkgs[j].Name })
|
||||
return pkgs, skipped, nil
|
||||
}
|
||||
|
||||
var updateLine = regexp.MustCompile(`^(\S+) (\S+) -> (\S+)$`)
|
||||
|
||||
// UpdateList is what a full upgrade would change, sorted by name, from checkupdates: a copy of the sync
|
||||
// databases refreshed apart from the machine's own, so asking never makes a partial upgrade possible. Its
|
||||
// status 2 is nothing to do; any other failure is an error, never an empty list.
|
||||
func (m *Machine) UpdateList() ([]VersionChange, error) {
|
||||
r := m.Run(bg(), "checkupdates")
|
||||
switch {
|
||||
case r.Err == "ENOENT":
|
||||
return nil, fmt.Errorf("checkupdates is not installed: it comes with pacman-contrib, which this module declares")
|
||||
case r.Err == "" && r.Status == 2:
|
||||
return []VersionChange{}, nil
|
||||
case r.Err != "" || r.Status != 0:
|
||||
return nil, failure("checkupdates", "checkupdates", r)
|
||||
}
|
||||
ups := []VersionChange{}
|
||||
for _, l := range lines(r.Stdout) {
|
||||
u := updateLine.FindStringSubmatch(strings.TrimSpace(l))
|
||||
if u == nil || !packageName.MatchString(u[1]) || !packageVersion.MatchString(u[2]) || !packageVersion.MatchString(u[3]) {
|
||||
continue
|
||||
}
|
||||
ups = append(ups, VersionChange{Name: u[1], From: u[2], To: u[3]})
|
||||
}
|
||||
sort.Slice(ups, func(i, j int) bool { return ups[i].Name < ups[j].Name })
|
||||
return ups, nil
|
||||
}
|
||||
|
||||
// DiffInstalled is what changed from one package list to the next, each list sorted by name: a package added,
|
||||
// removed, or at another version. A change of reason or repository alone is no change of a version.
|
||||
func DiffInstalled(before, after []InstalledPackage) Changes {
|
||||
c := Changes{Added: []PackageVersion{}, Removed: []PackageVersion{}, Changed: []VersionChange{}}
|
||||
was := map[string]string{}
|
||||
for _, p := range before {
|
||||
was[p.Name] = p.Version
|
||||
}
|
||||
is := map[string]bool{}
|
||||
for _, p := range after {
|
||||
is[p.Name] = true
|
||||
v, ok := was[p.Name]
|
||||
switch {
|
||||
case !ok:
|
||||
c.Added = append(c.Added, PackageVersion{p.Name, p.Version})
|
||||
case v != p.Version:
|
||||
c.Changed = append(c.Changed, VersionChange{p.Name, v, p.Version})
|
||||
}
|
||||
}
|
||||
for _, p := range before {
|
||||
if !is[p.Name] {
|
||||
c.Removed = append(c.Removed, PackageVersion{p.Name, p.Version})
|
||||
}
|
||||
}
|
||||
sort.Slice(c.Added, func(i, j int) bool { return c.Added[i].Name < c.Added[j].Name })
|
||||
sort.Slice(c.Removed, func(i, j int) bool { return c.Removed[i].Name < c.Removed[j].Name })
|
||||
sort.Slice(c.Changed, func(i, j int) bool { return c.Changed[i].Name < c.Changed[j].Name })
|
||||
return c
|
||||
}
|
||||
|
||||
// Empty says a change changed nothing.
|
||||
func (c Changes) Empty() bool { return len(c.Added)+len(c.Removed)+len(c.Changed) == 0 }
|
||||
@@ -1,7 +1,6 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"errors"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
@@ -78,31 +77,20 @@ func TestInfoReadsListsAsListsAndFallsBackToTheRepositories(t *testing.T) {
|
||||
}
|
||||
}
|
||||
|
||||
func TestInstalledSaysWhyAndWhatIsForeign(t *testing.T) {
|
||||
func TestReadInstalledSaysWhyAndWhichRepositoryOrForeign(t *testing.T) {
|
||||
m := machine(byLine(map[string]Ran{
|
||||
"pacman -Q": {Stdout: "glibc 2.42-1\nyay 12.0-1\nzsh 5.9-1\n"},
|
||||
"pacman -Q": {Stdout: "zsh 5.9-1\nglibc 2.42-1\nyay 12.0-1\n"},
|
||||
"pacman -Qeq": {Stdout: "yay\nzsh\n"},
|
||||
"pacman -Qmq": {Stdout: "yay\n"},
|
||||
"pacman -Sl": {Stdout: "core glibc 2.42-1 [installed]\nextra zsh 5.9-1 [installed]\nextra-testing zsh 5.9-2\n"},
|
||||
}, nil), 1000)
|
||||
r, err := m.Installed("", "", false, 10)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
pk, skipped, err := m.ReadInstalled()
|
||||
if err != nil || skipped != 0 {
|
||||
t.Fatal(err, skipped)
|
||||
}
|
||||
pk := r["packages"].([]Package)
|
||||
if pk[0].Reason != "dependency" || pk[1].Reason != "explicit" || !pk[1].Foreign || pk[2].Foreign {
|
||||
want := []InstalledPackage{{"glibc", "2.42-1", false, "core"}, {"yay", "12.0-1", true, RepoForeign}, {"zsh", "5.9-1", true, "extra"}}
|
||||
if !samePackages(pk, want) {
|
||||
t.Fatalf("%+v", pk)
|
||||
}
|
||||
if tot := r["totals"].(map[string]int); tot["explicit"] != 2 || tot["dependency"] != 1 || tot["foreign"] != 1 {
|
||||
t.Fatalf("%v", tot)
|
||||
}
|
||||
r, _ = m.Installed("", "", true, 10)
|
||||
if r["count"] != 1 {
|
||||
t.Fatalf("foreign only: %v", r)
|
||||
}
|
||||
r, _ = m.Installed("", "explicit", false, 1)
|
||||
if r["count"] != 2 || r["truncated"] != true {
|
||||
t.Fatalf("bounded: %v", r)
|
||||
}
|
||||
}
|
||||
|
||||
func TestOwnsAnswersNoOwnerAsAnAnswer(t *testing.T) {
|
||||
@@ -125,18 +113,18 @@ func TestOwnsAnswersNoOwnerAsAnAnswer(t *testing.T) {
|
||||
}
|
||||
}
|
||||
|
||||
func TestUpdatesReadsCheckupdatesAndItsNothingToDo(t *testing.T) {
|
||||
m := machine(byLine(map[string]Ran{"checkupdates": {Stdout: "linux 6.1-1 -> 6.2-1\nzsh 5.9-1 -> 5.9-2\n"}}, nil), 1000)
|
||||
r, err := m.Updates()
|
||||
if err != nil || r["count"] != 2 || r["updates"].([]Update)[0] != (Update{"linux", "6.1-1", "6.2-1"}) {
|
||||
func TestUpdateListReadsCheckupdatesAndItsNothingToDo(t *testing.T) {
|
||||
m := machine(byLine(map[string]Ran{"checkupdates": {Stdout: "zsh 5.9-1 -> 5.9-2\nlinux 6.1-1 -> 6.2-1\n"}}, nil), 1000)
|
||||
r, err := m.UpdateList()
|
||||
if err != nil || len(r) != 2 || r[0] != (VersionChange{"linux", "6.1-1", "6.2-1"}) {
|
||||
t.Fatalf("%v %v", r, err)
|
||||
}
|
||||
m = machine(byLine(map[string]Ran{"checkupdates": {Status: 2}}, nil), 1000)
|
||||
if r, err := m.Updates(); err != nil || r["count"] != 0 {
|
||||
if r, err := m.UpdateList(); err != nil || r == nil || len(r) != 0 {
|
||||
t.Fatalf("%v %v", r, err)
|
||||
}
|
||||
m = machine(byLine(map[string]Ran{"checkupdates": {Status: 1, Stderr: "==> ERROR: Cannot fetch updates\n"}}, nil), 1000)
|
||||
if _, err := m.Updates(); err == nil || !strings.Contains(err.Error(), "Cannot fetch updates") {
|
||||
if _, err := m.UpdateList(); err == nil || !strings.Contains(err.Error(), "Cannot fetch updates") {
|
||||
t.Fatalf("%v", err)
|
||||
}
|
||||
}
|
||||
@@ -146,60 +134,17 @@ func lockless(m *Machine) *Machine {
|
||||
return m
|
||||
}
|
||||
|
||||
func TestAnUpgradeRunsAsAUnitOfItsOwnThroughSudoAndBringsTheNews(t *testing.T) {
|
||||
fetch = func(string) ([]byte, error) { return nil, errors.New("no network") }
|
||||
var calls []call
|
||||
m := lockless(machine(fake(func(c call) Ran { return Ran{} }, &calls), 1000))
|
||||
r, err := m.Upgrade("", 60)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
unit := r["started"].(string)
|
||||
if unit != "mesh-pacman-upgrade-1791115200" {
|
||||
t.Fatalf("unit: %s", unit)
|
||||
}
|
||||
last := calls[len(calls)-1]
|
||||
want := "sudo -n systemd-run --unit=" + unit
|
||||
if !strings.HasPrefix(last.String(), want) || !strings.HasSuffix(last.String(), "--quiet pacman -Syu --noconfirm") || strings.Contains(last.String(), "--wait") {
|
||||
t.Fatalf("started as: %s", last)
|
||||
}
|
||||
news := r["news"].(map[string]any)
|
||||
if news["reachable"] != false || !strings.Contains(news["error"].(string), "no network") {
|
||||
t.Fatalf("no network is an answer: %v", news)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAnUpgradeIsRefusedWhileTheDatabaseIsLocked(t *testing.T) {
|
||||
fetch = func(string) ([]byte, error) { return nil, errors.New("offline") }
|
||||
var calls []call
|
||||
m := machine(fake(func(c call) Ran { return Ran{} }, &calls), 1000)
|
||||
m.ReadFile = func(p string) ([]byte, error) {
|
||||
if p == DBLock {
|
||||
return []byte{}, nil
|
||||
}
|
||||
return nil, errNoFile
|
||||
}
|
||||
if _, err := m.Upgrade("", 60); err == nil || !strings.Contains(err.Error(), "another pacman holds") {
|
||||
t.Fatalf("%v", err)
|
||||
}
|
||||
for _, c := range calls {
|
||||
if c.name == "sudo" {
|
||||
t.Fatal("started while locked")
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestAnUpgradesUnitIsReadBack(t *testing.T) {
|
||||
unit := "mesh-pacman-upgrade-1791115200"
|
||||
m := machine(byLine(map[string]Ran{
|
||||
"systemctl show " + unit + " --no-pager --property=LoadState --property=ActiveState --property=SubState --property=Result --property=ExecMainStatus": {Stdout: "LoadState=loaded\nActiveState=failed\nSubState=failed\nResult=exit-code\nExecMainStatus=1\n"},
|
||||
"sudo -n journalctl --no-pager -o cat -n 60 -u " + unit: {Stdout: "error: failed to commit transaction (conflicting files)\n"},
|
||||
}, nil), 1000)
|
||||
r, err := m.Upgrade(unit, 60)
|
||||
r, err := m.UpgradeStatus(unit, 60)
|
||||
if err != nil || r["finished"] != true || r["succeeded"] != false || r["exit_status"] != "1" || len(r["log"].([]string)) != 1 {
|
||||
t.Fatalf("%v %v", r, err)
|
||||
}
|
||||
if _, err := m.Upgrade("sshd.service", 60); err == nil {
|
||||
if _, err := m.UpgradeStatus("sshd.service", 60); err == nil {
|
||||
t.Fatal("a unit the tools did not start was read")
|
||||
}
|
||||
}
|
||||
@@ -374,57 +319,3 @@ func TestConfigIsReadAsPacmanConfResolvesIt(t *testing.T) {
|
||||
t.Fatalf("%+v", c)
|
||||
}
|
||||
}
|
||||
|
||||
// The weekly upgrade's outcome is the module's health (issue 1574, ADR 0477 clause 5).
|
||||
func upgradeHealthMachine(props, journal string) *Machine {
|
||||
return machine(byLine(map[string]Ran{
|
||||
"systemctl show " + WeeklyUnit + " --no-pager --property=LoadState --property=ActiveState --property=Result --property=ExecMainStatus --property=ExecMainExitTimestamp --property=ExecMainStartTimestamp": {Stdout: props},
|
||||
"sudo -n journalctl --no-pager -o cat -n 40 -u " + WeeklyUnit: {Stdout: journal},
|
||||
"sudo -n journalctl --no-pager -o cat -n 1 -u " + WeeklyUnit: {Stdout: journal},
|
||||
}, nil), 1000)
|
||||
}
|
||||
|
||||
func TestAFailedWeeklyUpgradeIsUnhealthyNamingTheError(t *testing.T) {
|
||||
h := upgradeHealthMachine("LoadState=loaded\nActiveState=failed\nResult=exit-code\nExecMainStatus=1\nExecMainExitTimestamp=Sat 2026-10-24 04:03:11 CEST\n",
|
||||
":: Synchronizing package databases...\nerror: failed to commit transaction (conflicting files)\nErrors occurred, no packages were upgraded.\n").UpgradeHealth()
|
||||
if h.Healthy || !strings.Contains(h.Why, "failed to commit transaction (conflicting files)") || !strings.Contains(h.Why, "exit-code") || !strings.Contains(h.Why, "Sat 2026-10-24 04:03:11 CEST") {
|
||||
t.Fatalf("%+v", h)
|
||||
}
|
||||
h = upgradeHealthMachine("LoadState=loaded\nActiveState=failed\nResult=exit-code\nExecMainStatus=1\nExecMainExitTimestamp=\n", "").UpgradeHealth()
|
||||
if h.Healthy || !strings.Contains(h.Why, "status 1") {
|
||||
t.Fatalf("a failure without a journal is still unhealthy: %+v", h)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAPassedRunningOrNeverRunWeeklyUpgradeIsHealthy(t *testing.T) {
|
||||
for name, props := range map[string]string{
|
||||
"passed": "LoadState=loaded\nActiveState=inactive\nResult=success\nExecMainStatus=0\nExecMainExitTimestamp=Sat 2026-10-24 04:09:00 CEST\n",
|
||||
"running": "LoadState=loaded\nActiveState=activating\nResult=success\nExecMainStatus=0\nExecMainExitTimestamp=\nExecMainStartTimestamp=Wed 2026-10-04 11:50:00 UTC\n",
|
||||
"never run": "LoadState=loaded\nActiveState=inactive\nResult=success\nExecMainStatus=0\nExecMainExitTimestamp=\n",
|
||||
} {
|
||||
if h := upgradeHealthMachine(props, "").UpgradeHealth(); !h.Healthy {
|
||||
t.Errorf("%s: %+v", name, h)
|
||||
}
|
||||
}
|
||||
if h := machine(fake(func(call) Ran { return Ran{Status: 1, Stderr: "boom"} }, nil), 1000).UpgradeHealth(); h.Healthy {
|
||||
t.Errorf("a unit that cannot be read is not healthy: %+v", h)
|
||||
}
|
||||
}
|
||||
|
||||
// A run that goes on past RunLimit is hung and holds pacman's lock: unhealthy, naming its start (issue 1574).
|
||||
func TestAHungWeeklyUpgradeIsUnhealthyNamingItsStart(t *testing.T) {
|
||||
// the test machine's clock is 2026-10-04 12:00 UTC
|
||||
h := upgradeHealthMachine("LoadState=loaded\nActiveState=activating\nResult=success\nExecMainStatus=0\nExecMainStartTimestamp=Wed 2026-10-04 09:00:00 UTC\n",
|
||||
"downloading linux-6.99...\n").UpgradeHealth()
|
||||
if h.Healthy || !strings.Contains(h.Why, "Wed 2026-10-04 09:00:00 UTC") || !strings.Contains(h.Why, "downloading linux-6.99") {
|
||||
t.Fatalf("started 3 h ago: %+v", h)
|
||||
}
|
||||
h = upgradeHealthMachine("LoadState=loaded\nActiveState=activating\nResult=success\nExecMainStatus=0\nExecMainStartTimestamp=Wed 2026-10-04 11:50:00 UTC\n", "").UpgradeHealth()
|
||||
if !h.Healthy {
|
||||
t.Fatalf("started 10 min ago: %+v", h)
|
||||
}
|
||||
h = upgradeHealthMachine("LoadState=loaded\nActiveState=activating\nResult=success\nExecMainStatus=0\nExecMainStartTimestamp=\n", "").UpgradeHealth()
|
||||
if h.Healthy {
|
||||
t.Fatalf("a start that cannot be read is not healthy: %+v", h)
|
||||
}
|
||||
}
|
||||
|
||||
@@ -8,7 +8,6 @@ import (
|
||||
"fmt"
|
||||
"path"
|
||||
"regexp"
|
||||
"sort"
|
||||
"strings"
|
||||
)
|
||||
|
||||
@@ -177,10 +176,11 @@ func (m *Machine) Info(name string) (map[string]any, error) {
|
||||
|
||||
// Package is an installed package and why it is installed.
|
||||
type Package struct {
|
||||
Name string `json:"name"`
|
||||
Version string `json:"version"`
|
||||
Reason string `json:"reason"`
|
||||
Foreign bool `json:"foreign"`
|
||||
Name string `json:"name"`
|
||||
Version string `json:"version"`
|
||||
Reason string `json:"reason"`
|
||||
Repository string `json:"repository,omitempty"`
|
||||
Foreign bool `json:"foreign"`
|
||||
}
|
||||
|
||||
func nameVersions(out string) [][2]string {
|
||||
@@ -193,51 +193,6 @@ func nameVersions(out string) [][2]string {
|
||||
return nv
|
||||
}
|
||||
|
||||
func nameSet(out string) map[string]bool {
|
||||
s := map[string]bool{}
|
||||
for _, l := range lines(out) {
|
||||
s[strings.TrimSpace(l)] = true
|
||||
}
|
||||
return s
|
||||
}
|
||||
|
||||
// Installed is every installed package with its version, whether it was installed explicitly or as
|
||||
// a dependency, and whether it is foreign (in no repository this machine syncs).
|
||||
func (m *Machine) Installed(match, reason string, foreignOnly bool, limit int) (map[string]any, error) {
|
||||
all, err := m.query("-Q")
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
explicit, err := m.query("-Qeq")
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
foreign, err := m.query("-Qmq")
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
ex, fo := nameSet(explicit), nameSet(foreign)
|
||||
pkgs := []Package{}
|
||||
counts := map[string]int{"explicit": 0, "dependency": 0, "foreign": 0}
|
||||
for _, nv := range nameVersions(all) {
|
||||
p := Package{Name: nv[0], Version: nv[1], Reason: "dependency", Foreign: fo[nv[0]]}
|
||||
if ex[p.Name] {
|
||||
p.Reason = "explicit"
|
||||
}
|
||||
counts[p.Reason]++
|
||||
if p.Foreign {
|
||||
counts["foreign"]++
|
||||
}
|
||||
if match != "" && !strings.Contains(p.Name, match) || reason != "" && p.Reason != reason || foreignOnly && !p.Foreign {
|
||||
continue
|
||||
}
|
||||
pkgs = append(pkgs, p)
|
||||
}
|
||||
out := bound("packages", pkgs, limit)
|
||||
out["totals"] = counts
|
||||
return out, nil
|
||||
}
|
||||
|
||||
var ownedBy = regexp.MustCompile(`^(.*) is owned by (\S+) (\S+)$`)
|
||||
|
||||
// Owns is which package owns a path.
|
||||
@@ -292,51 +247,6 @@ func (m *Machine) Orphans() (map[string]any, error) {
|
||||
return map[string]any{"count": len(pkgs), "orphans": pkgs}, nil
|
||||
}
|
||||
|
||||
// Foreign is every installed package no repository this machine syncs carries: built from the AUR
|
||||
// or by hand, which the host's `package` shape cannot install (research 027, question 1).
|
||||
func (m *Machine) Foreign() (map[string]any, error) {
|
||||
out, err := m.query("-Qm")
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
pkgs := []map[string]string{}
|
||||
for _, nv := range nameVersions(out) {
|
||||
pkgs = append(pkgs, map[string]string{"name": nv[0], "version": nv[1]})
|
||||
}
|
||||
sort.Slice(pkgs, func(i, j int) bool { return pkgs[i]["name"] < pkgs[j]["name"] })
|
||||
return map[string]any{"count": len(pkgs), "packages": pkgs}, nil
|
||||
}
|
||||
|
||||
// Update is one package an upgrade would change.
|
||||
type Update struct {
|
||||
Name string `json:"name"`
|
||||
From string `json:"from"`
|
||||
To string `json:"to"`
|
||||
}
|
||||
|
||||
var updateLine = regexp.MustCompile(`^(\S+) (\S+) -> (\S+)`)
|
||||
|
||||
// Updates is what a full upgrade would change, from checkupdates: a copy of the sync databases
|
||||
// refreshed apart from the machine's own, so asking never makes a partial upgrade possible.
|
||||
func (m *Machine) Updates() (map[string]any, error) {
|
||||
r := m.Run(bg(), "checkupdates")
|
||||
switch {
|
||||
case r.Err == "ENOENT":
|
||||
return nil, fmt.Errorf("checkupdates is not installed: it comes with pacman-contrib, which this module declares")
|
||||
case r.Err == "" && r.Status == 2:
|
||||
return map[string]any{"count": 0, "updates": []Update{}}, nil
|
||||
case r.Err != "" || r.Status != 0:
|
||||
return nil, failure("checkupdates", "checkupdates", r)
|
||||
}
|
||||
ups := []Update{}
|
||||
for _, l := range lines(r.Stdout) {
|
||||
if u := updateLine.FindStringSubmatch(strings.TrimSpace(l)); u != nil {
|
||||
ups = append(ups, Update{u[1], u[2], u[3]})
|
||||
}
|
||||
}
|
||||
return map[string]any{"count": len(ups), "updates": ups}, nil
|
||||
}
|
||||
|
||||
// Config is the configuration pacman runs with, as pacman-conf resolves it.
|
||||
type Config struct {
|
||||
Options map[string][]string `json:"options"`
|
||||
|
||||
@@ -0,0 +1,183 @@
|
||||
package main
|
||||
|
||||
// The package record taken before every upgrade (novox/hq ADR 0508 §2): a copy of the node's `installed` key,
|
||||
// the running kernel, the units failed at that moment and the installed DKMS modules, written to `records` at
|
||||
// `<node>.<run>` and read back. A run does not start, and says why, when the node's packages are marked not
|
||||
// current or the record cannot be written and read back equal. Records older than `record-days` are deleted.
|
||||
|
||||
import (
|
||||
"bytes"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"sort"
|
||||
"strings"
|
||||
"time"
|
||||
)
|
||||
|
||||
// OSRelease is the running kernel's release.
|
||||
const OSRelease = "/proc/sys/kernel/osrelease"
|
||||
|
||||
// defaultRecordDays is record-days' default in the manifest, for a config file that holds none.
|
||||
const defaultRecordDays = 180
|
||||
|
||||
// FailedUnits are the units of the service manager failed now, by name.
|
||||
func (m *Machine) FailedUnits() ([]string, error) {
|
||||
out, err := m.Out("systemctl", "list-units", "--state=failed", "--plain", "--no-legend", "--no-pager")
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
units := []string{}
|
||||
for _, l := range lines(out) {
|
||||
if f := strings.Fields(l); len(f) > 0 {
|
||||
units = append(units, f[0])
|
||||
}
|
||||
}
|
||||
sort.Strings(units)
|
||||
return units, nil
|
||||
}
|
||||
|
||||
// ParseDKMSStatus reads `dkms status`: each module at its version and the kernels it is installed for. A module
|
||||
// only added (not built for any kernel) is listed with none. Both forms dkms has written are read:
|
||||
// `name/version, kernel, arch: state` and the older `name, version, kernel, arch: state`.
|
||||
func ParseDKMSStatus(out string) []DKMSModule {
|
||||
byModule := map[string]*DKMSModule{}
|
||||
var order []string
|
||||
for _, l := range lines(out) {
|
||||
head, state, ok := strings.Cut(l, ": ")
|
||||
if !ok {
|
||||
continue
|
||||
}
|
||||
parts := strings.Split(head, ", ")
|
||||
var name, version, kernel string
|
||||
if n, v, slash := strings.Cut(parts[0], "/"); slash {
|
||||
name, version = n, v
|
||||
if len(parts) > 1 {
|
||||
kernel = parts[1]
|
||||
}
|
||||
} else if len(parts) >= 2 {
|
||||
name, version = parts[0], parts[1]
|
||||
if len(parts) > 2 {
|
||||
kernel = parts[2]
|
||||
}
|
||||
} else {
|
||||
continue
|
||||
}
|
||||
key := name + "/" + version
|
||||
m := byModule[key]
|
||||
if m == nil {
|
||||
m = &DKMSModule{Module: name, Version: version, Kernels: []string{}}
|
||||
byModule[key] = m
|
||||
order = append(order, key)
|
||||
}
|
||||
if kernel != "" && strings.HasPrefix(strings.TrimSpace(state), "installed") {
|
||||
m.Kernels = append(m.Kernels, kernel)
|
||||
}
|
||||
}
|
||||
sort.Strings(order)
|
||||
modules := []DKMSModule{}
|
||||
for _, k := range order {
|
||||
sort.Strings(byModule[k].Kernels)
|
||||
modules = append(modules, *byModule[k])
|
||||
}
|
||||
return modules
|
||||
}
|
||||
|
||||
// DKMSModules are the DKMS modules installed on the machine; none when dkms is not installed.
|
||||
func (m *Machine) DKMSModules() ([]DKMSModule, error) {
|
||||
r := m.Run(bg(), "dkms", "status")
|
||||
if r.Err == "ENOENT" {
|
||||
return []DKMSModule{}, nil
|
||||
}
|
||||
if r.Status != 0 || r.Err != "" {
|
||||
return nil, failure("dkms", "dkms", r)
|
||||
}
|
||||
return ParseDKMSStatus(r.Stdout), nil
|
||||
}
|
||||
|
||||
// TakeRecord takes the node's record for a run, writes it and reads it back. It answers the record, or why
|
||||
// it could not be taken: a run that has a why does not start.
|
||||
func (u *Upgrades) TakeRecord(run string) (*Record, string) {
|
||||
nc, err := u.Installed.Get(NotCurrentKey(u.Node))
|
||||
if err != nil {
|
||||
return nil, "the node's packages could not be read from the bus: " + err.Error()
|
||||
}
|
||||
if nc != nil {
|
||||
var mark NotCurrent
|
||||
_ = json.Unmarshal(nc.Value, &mark)
|
||||
return nil, fmt.Sprintf("the node's packages on the bus are not current since %s (%d bytes, over the bus's cap on a value): a record of them would not be what is installed",
|
||||
mark.Since.Format(time.RFC3339), mark.Bytes)
|
||||
}
|
||||
e, err := u.Installed.Get(u.Node)
|
||||
if err != nil {
|
||||
return nil, "the node's packages could not be read from the bus: " + err.Error()
|
||||
}
|
||||
if e == nil {
|
||||
return nil, "the node's packages are not on the bus yet: there is nothing to record"
|
||||
}
|
||||
var installed InstalledValue
|
||||
if err := json.Unmarshal(e.Value, &installed); err != nil {
|
||||
return nil, "the node's packages on the bus are not readable: " + err.Error()
|
||||
}
|
||||
kernel, err := u.Machine.ReadFile(OSRelease)
|
||||
if err != nil {
|
||||
return nil, "the running kernel could not be read: " + err.Error()
|
||||
}
|
||||
failed, err := u.Machine.FailedUnits()
|
||||
if err != nil {
|
||||
return nil, "the failed units could not be read: " + err.Error()
|
||||
}
|
||||
dkms, err := u.Machine.DKMSModules()
|
||||
if err != nil {
|
||||
return nil, "the DKMS modules could not be read: " + err.Error()
|
||||
}
|
||||
rec := Record{Node: u.Node, Run: run, At: u.Machine.Now().UTC(), Installed: installed,
|
||||
Kernel: strings.TrimSpace(string(kernel)), FailedUnits: failed, DKMS: dkms}
|
||||
key := NodeKey(u.Node, run)
|
||||
if _, err := u.Records.Put(key, rec); err != nil {
|
||||
return nil, "the record could not be written: " + err.Error()
|
||||
}
|
||||
back, err := u.Records.Get(key)
|
||||
if err != nil {
|
||||
return nil, "the record could not be read back: " + err.Error()
|
||||
}
|
||||
if back == nil {
|
||||
return nil, "the record was written and is not there when read back"
|
||||
}
|
||||
var read Record
|
||||
want, _ := json.Marshal(rec)
|
||||
if err := json.Unmarshal(back.Value, &read); err != nil {
|
||||
return nil, "the record read back is not readable: " + err.Error()
|
||||
}
|
||||
if got, _ := json.Marshal(read); !bytes.Equal(got, want) {
|
||||
return nil, "the record read back differs from the one written"
|
||||
}
|
||||
return &rec, ""
|
||||
}
|
||||
|
||||
// PruneRecords deletes the node's records older than record-days, judged by the time of their run, and
|
||||
// answers how many it deleted.
|
||||
func (u *Upgrades) PruneRecords() (int, error) {
|
||||
days := u.Config.RecordDays
|
||||
if days < 1 {
|
||||
days = defaultRecordDays
|
||||
}
|
||||
cutoff := u.Machine.Now().AddDate(0, 0, -days)
|
||||
keys, err := u.Records.Keys()
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("listing records: %w", err)
|
||||
}
|
||||
n := 0
|
||||
for _, k := range keys {
|
||||
node, id, ok := strings.Cut(k, ".")
|
||||
if !ok || node != u.Node {
|
||||
continue
|
||||
}
|
||||
if t, ok := runTime(id); ok && t.Before(cutoff) {
|
||||
if err := u.Records.Delete(k); err != nil {
|
||||
return n, fmt.Errorf("deleting record %s: %w", k, err)
|
||||
}
|
||||
n++
|
||||
}
|
||||
}
|
||||
return n, nil
|
||||
}
|
||||
@@ -0,0 +1,152 @@
|
||||
package main
|
||||
|
||||
// The record before every upgrade (novox/hq ADR 0508 §2; acceptance 1841/AC3).
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"os"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
)
|
||||
|
||||
func capturedDKMSStatus(t *testing.T) string {
|
||||
t.Helper()
|
||||
raw, err := os.ReadFile("testdata/dkms-status-installed-and-added/status")
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
return string(raw)
|
||||
}
|
||||
|
||||
// recordMachine is a machine running one kernel, with one unit failed and the captured DKMS modules.
|
||||
func recordMachine(t *testing.T) *Machine {
|
||||
m := machine(byLine(map[string]Ran{
|
||||
"systemctl list-units --state=failed --plain --no-legend --no-pager": {Stdout: "backup.service loaded failed failed The backup\n"},
|
||||
"dkms status": {Stdout: capturedDKMSStatus(t)},
|
||||
}, nil), 1000)
|
||||
m.ReadFile = func(path string) ([]byte, error) {
|
||||
if path == OSRelease {
|
||||
return []byte("7.2.8-arch1-2\n"), nil
|
||||
}
|
||||
return nil, errNoFile
|
||||
}
|
||||
return m
|
||||
}
|
||||
|
||||
var nodeAPackages = InstalledValue{At: time.Date(2026, 10, 3, 9, 0, 0, 0, time.UTC), Packages: []InstalledPackage{
|
||||
{Name: "linux", Version: "7.2.8.arch1-2", Explicit: true, Repo: "core"},
|
||||
{Name: "zfs-dkms", Version: "2.4.4-1", Explicit: true, Repo: RepoForeign},
|
||||
}}
|
||||
|
||||
func recordUpgrades(t *testing.T) (*Upgrades, *memBucket, *memBucket) {
|
||||
installed, records := newBucket(), newBucket()
|
||||
if _, err := installed.Put("node-a", nodeAPackages); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
return &Upgrades{Node: "node-a", Machine: recordMachine(t), Installed: installed, Records: records}, installed, records
|
||||
}
|
||||
|
||||
func TestDKMSStatusInBothFormsIsEachModuleWithTheKernelsItIsBuiltFor(t *testing.T) {
|
||||
mods := ParseDKMSStatus(capturedDKMSStatus(t) + "zfs, 2.4.4, 7.2.8-arch1-2, x86_64: installed\nzfs, 2.4.4, 6.18.54-2-lts, x86_64: installed\n")
|
||||
var got []string
|
||||
for _, m := range mods {
|
||||
got = append(got, m.Module+"/"+m.Version+"="+strings.Join(m.Kernels, "+"))
|
||||
}
|
||||
want := "openrazer-driver/3.12.4=7.2.8-arch1-2 vboxhost/7.2.20_OSE=7.2.8-arch1-2 xone/0.5.8=7.2.8-arch1-2 xpad/0.4= zfs/2.4.4=6.18.54-2-lts+7.2.8-arch1-2"
|
||||
if strings.Join(got, " ") != want {
|
||||
t.Errorf("%v", got)
|
||||
}
|
||||
}
|
||||
|
||||
func TestARecordIsTheNodesPackagesKernelFailedUnitsAndDKMSWrittenAndReadBack(t *testing.T) {
|
||||
u, _, records := recordUpgrades(t)
|
||||
rec, why := u.TakeRecord("20261004T120000Z")
|
||||
if why != "" || rec == nil {
|
||||
t.Fatalf("refused: %s", why)
|
||||
}
|
||||
var kept Record
|
||||
if !records.value("node-a.20261004T120000Z", &kept) {
|
||||
t.Fatal("no record at <node>.<run>")
|
||||
}
|
||||
a, _ := json.Marshal(kept.Installed)
|
||||
b, _ := json.Marshal(nodeAPackages)
|
||||
if string(a) != string(b) {
|
||||
t.Errorf("the record copies installed.<node>: %s", a)
|
||||
}
|
||||
if kept.Kernel != "7.2.8-arch1-2" || strings.Join(kept.FailedUnits, " ") != "backup.service" || len(kept.DKMS) != 4 ||
|
||||
kept.Node != "node-a" || kept.Run != "20261004T120000Z" {
|
||||
t.Errorf("record: %+v", kept)
|
||||
}
|
||||
if w := strings.Join(records.writes, " "); w != "put node-a.20261004T120000Z" {
|
||||
t.Errorf("writes: %s", w)
|
||||
}
|
||||
}
|
||||
|
||||
func TestNoRecordAndNoRunWhenThePackagesAreNotCurrentOrTheRecordIsNotReadBack(t *testing.T) {
|
||||
u, installed, records := recordUpgrades(t)
|
||||
if _, err := installed.Put(NotCurrentKey("node-a"), NotCurrent{Bytes: 300000, Since: time.Date(2026, 10, 3, 0, 0, 0, 0, time.UTC)}); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if rec, why := u.TakeRecord("r1"); rec != nil || !strings.Contains(why, "not current") || !strings.Contains(why, "300000") {
|
||||
t.Errorf("not current: %v %q", rec, why)
|
||||
}
|
||||
if len(records.writes) != 0 {
|
||||
t.Errorf("a record of packages not current was written: %v", records.writes)
|
||||
}
|
||||
|
||||
u, installed, _ = recordUpgrades(t)
|
||||
_ = installed.Delete("node-a")
|
||||
if _, why := u.TakeRecord("r1"); !strings.Contains(why, "not on the bus") {
|
||||
t.Errorf("no installed key: %q", why)
|
||||
}
|
||||
|
||||
u, _, records = recordUpgrades(t)
|
||||
records.fail["put node-a.r1"] = errors.New("value over the bucket's limit")
|
||||
if _, why := u.TakeRecord("r1"); !strings.Contains(why, "could not be written") || !strings.Contains(why, "over the bucket's limit") {
|
||||
t.Errorf("not written: %q", why)
|
||||
}
|
||||
|
||||
u, _, _ = recordUpgrades(t)
|
||||
u.Records = &mangling{newBucket()}
|
||||
if _, why := u.TakeRecord("r1"); !strings.Contains(why, "differs") {
|
||||
t.Errorf("read back unequal: %q", why)
|
||||
}
|
||||
}
|
||||
|
||||
// mangling is a state that reads back other than what was written.
|
||||
type mangling struct{ *memBucket }
|
||||
|
||||
func (m *mangling) Get(key string) (*stdio.StateEntry, error) {
|
||||
e, err := m.memBucket.Get(key)
|
||||
if e != nil {
|
||||
e.Value = []byte(strings.Replace(string(e.Value), "7.2.8-arch1-2", "7.2.7-arch1-1", 1))
|
||||
}
|
||||
return e, err
|
||||
}
|
||||
|
||||
func TestRecordsOlderThanRecordDaysAreDeletedAndYoungerOnesKept(t *testing.T) {
|
||||
u, _, records := recordUpgrades(t)
|
||||
u.Config.RecordDays = 180
|
||||
now := u.Machine.Now() // 2026-10-04
|
||||
for _, k := range []string{
|
||||
NodeKey("node-a", RunID(now.AddDate(0, 0, -181))),
|
||||
NodeKey("node-a", RunID(now.AddDate(0, 0, -179))),
|
||||
NodeKey("node-b", RunID(now.AddDate(0, 0, -400))),
|
||||
} {
|
||||
if _, err := records.Put(k, Record{}); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
}
|
||||
n, err := u.PruneRecords()
|
||||
if err != nil || n != 1 {
|
||||
t.Fatalf("%d deleted, %v", n, err)
|
||||
}
|
||||
keys, _ := records.Keys()
|
||||
if strings.Join(keys, " ") != NodeKey("node-a", RunID(now.AddDate(0, 0, -179)))+" "+NodeKey("node-b", RunID(now.AddDate(0, 0, -400))) {
|
||||
t.Errorf("kept: %v (the younger record, and another node's, which its own keeper prunes)", keys)
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,300 @@
|
||||
package main
|
||||
|
||||
// A rollback (novox/hq ADR 0508 §7; acceptance 1841/AC8): a node is returned to the record a run took, and says
|
||||
// first what that would change. The plan compares the record's packages with the node's `installed` now: each
|
||||
// package whose version differs, or that is gone since, is brought back to the record's version, from the node's
|
||||
// package cache when it holds that file, else from the Arch Linux Archive's copy of that exact version, else from
|
||||
// the archive's repository as it stood on the record's date. The packages installed since are listed, and are
|
||||
// removed only when asked. A foreign package, in the record or now, is never in the plan (ADR 0507 §7).
|
||||
//
|
||||
// The archive keeps every version its repositories published, so a version is taken from it by its file's name,
|
||||
// which needs the package's architecture: the architecture of any of the package's files in the cache, else of
|
||||
// its installed version (the local database). A package whose architecture is not known on this machine is
|
||||
// taken from the dated repository, where pacman finds it by name. Nothing is fetched to make the plan; a file
|
||||
// the archive does not hold fails the transaction, which the run then says.
|
||||
//
|
||||
// With apply, pacman_rollback writes an ask; the node's keeper takes it (TakeAsk), makes the plan again on the
|
||||
// node itself (its cache is the one that counts), and runs it as a run of kind rollback: one transaction started
|
||||
// by the root helper with the plan in the request (RootActs.Rollback), `rollback.started` and
|
||||
// `rollback.finished`, then the after-upgrade check against the record (Follow, checkAfter).
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"fmt"
|
||||
"path"
|
||||
"regexp"
|
||||
"sort"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
)
|
||||
|
||||
// ArchiveURL is the Arch Linux Archive.
|
||||
const ArchiveURL = "https://archive.archlinux.org"
|
||||
|
||||
// SourceRemoved marks a plan item that removes a package installed since the record (to "" and no location):
|
||||
// only when the rollback was asked to remove what the upgrade added.
|
||||
const SourceRemoved = "removed"
|
||||
|
||||
// RollbackPlan is what a rollback to a run's record would change.
|
||||
type RollbackPlan struct {
|
||||
Node string `json:"node"`
|
||||
Of string `json:"of"`
|
||||
// RecordAt is when the record was taken; the dated repository is that day's.
|
||||
RecordAt time.Time `json:"record_at"`
|
||||
// Items brings each package back, or removes one installed since when RemoveAdded.
|
||||
Items []RollbackItem `json:"items"`
|
||||
// Added is the packages installed since the record, not foreign.
|
||||
Added []PackageVersion `json:"added"`
|
||||
RemoveAdded bool `json:"remove_added"`
|
||||
// Foreign is how many foreign packages the plan leaves untouched.
|
||||
Foreign int `json:"foreign_untouched"`
|
||||
// CacheRead says whether the node's package cache was read: only on the node itself.
|
||||
CacheRead bool `json:"cache_read"`
|
||||
}
|
||||
|
||||
// cacheFile is `<name>-<version>-<arch>.pkg.tar.<ext>`; a name may hold dashes, a version and an arch none.
|
||||
var cacheFile = regexp.MustCompile(`^(.+)-([^-]+-[^-]+)-([^-]+)\.pkg\.tar\.(zst|xz)$`)
|
||||
|
||||
// packageCache is the node's cache: by package name, by version, the file's path; and each name's arch.
|
||||
type packageCache struct {
|
||||
files map[string]map[string]string
|
||||
arch map[string]string
|
||||
}
|
||||
|
||||
// readCache lists the cache's package files.
|
||||
func (m *Machine) readCache() (packageCache, error) {
|
||||
c := packageCache{files: map[string]map[string]string{}, arch: map[string]string{}}
|
||||
out, err := m.Out("find", CacheDir, "-mindepth", "1", "-maxdepth", "1", "-name", "*.pkg.tar.*", "-printf", "%f\n")
|
||||
if err != nil {
|
||||
return c, err
|
||||
}
|
||||
for _, l := range lines(out) {
|
||||
f := cacheFile.FindStringSubmatch(strings.TrimSpace(l))
|
||||
if f == nil || !packageName.MatchString(f[1]) || !packageVersion.MatchString(f[2]) {
|
||||
continue
|
||||
}
|
||||
if c.files[f[1]] == nil {
|
||||
c.files[f[1]] = map[string]string{}
|
||||
}
|
||||
if _, ok := c.files[f[1]][f[2]]; !ok || f[4] == "zst" {
|
||||
c.files[f[1]][f[2]] = path.Join(CacheDir, f[0])
|
||||
}
|
||||
c.arch[f[1]] = f[3]
|
||||
}
|
||||
return c, nil
|
||||
}
|
||||
|
||||
// installedArch is an installed package's architecture from the local database, or "".
|
||||
func (m *Machine) installedArch(name, version string) string {
|
||||
desc, err := m.ReadFile(path.Join(PacmanDBDir, "local", name+"-"+version, "desc"))
|
||||
if err != nil {
|
||||
return ""
|
||||
}
|
||||
ls := strings.Split(string(desc), "\n")
|
||||
for i, l := range ls {
|
||||
if strings.TrimSpace(l) == "%ARCH%" && i+1 < len(ls) {
|
||||
if a := strings.TrimSpace(ls[i+1]); repoName.MatchString(a) {
|
||||
return a
|
||||
}
|
||||
}
|
||||
}
|
||||
return ""
|
||||
}
|
||||
|
||||
// source is where a package's record version comes from.
|
||||
func source(name, version, repo string, arch string, cache *packageCache, recordAt time.Time) (string, string) {
|
||||
if cache != nil {
|
||||
if p, ok := cache.files[name][version]; ok {
|
||||
return SourceCache, p
|
||||
}
|
||||
}
|
||||
if arch != "" {
|
||||
file := name + "-" + version + "-" + arch + ".pkg.tar.zst"
|
||||
return SourceArchive, ArchiveURL + "/packages/" + name[:1] + "/" + name + "/" + file
|
||||
}
|
||||
day := recordAt.UTC().Format("2006/01/02")
|
||||
return SourceRepository, ArchiveURL + "/repos/" + day + "/" + repo + "/os/$arch"
|
||||
}
|
||||
|
||||
// keyGetter reads one key: a Bucket the keeper writes, or a StateReader a verb reads.
|
||||
type keyGetter interface {
|
||||
Get(key string) (*stdio.StateEntry, error)
|
||||
}
|
||||
|
||||
// PlanRollback is the plan for returning node to the record of run, from the states: `records.<node>.<run>` and
|
||||
// `installed.<node>`, which must be current. local says this is the node itself, whose cache and local database
|
||||
// are read.
|
||||
func PlanRollback(m *Machine, installed, records keyGetter, node, run string, removeAdded, local bool) (RollbackPlan, error) {
|
||||
p := RollbackPlan{Node: node, Of: run, Items: []RollbackItem{}, Added: []PackageVersion{}, RemoveAdded: removeAdded}
|
||||
if e, err := installed.Get(NotCurrentKey(node)); err != nil {
|
||||
return p, err
|
||||
} else if e != nil {
|
||||
return p, fmt.Errorf("the packages of %s on the bus are not current: a plan from them would not be what is installed", node)
|
||||
}
|
||||
var rec Record
|
||||
if e, err := records.Get(NodeKey(node, run)); err != nil {
|
||||
return p, err
|
||||
} else if e == nil {
|
||||
return p, fmt.Errorf("%s has no record of run %s: it was never taken, or is older than record-days", node, run)
|
||||
} else if err := json.Unmarshal(e.Value, &rec); err != nil {
|
||||
return p, fmt.Errorf("the record of run %s is not readable: %w", run, err)
|
||||
}
|
||||
var now InstalledValue
|
||||
if e, err := installed.Get(node); err != nil {
|
||||
return p, err
|
||||
} else if e == nil {
|
||||
return p, fmt.Errorf("no packages of %s are on the bus", node)
|
||||
} else if err := json.Unmarshal(e.Value, &now); err != nil {
|
||||
return p, fmt.Errorf("installed.%s is not readable: %w", node, err)
|
||||
}
|
||||
p.RecordAt = rec.At
|
||||
var cache *packageCache
|
||||
if local {
|
||||
c, err := m.readCache()
|
||||
if err != nil {
|
||||
return p, fmt.Errorf("reading the package cache: %w", err)
|
||||
}
|
||||
cache, p.CacheRead = &c, true
|
||||
}
|
||||
foreign := map[string]bool{}
|
||||
nowBy := map[string]InstalledPackage{}
|
||||
for _, ip := range now.Packages {
|
||||
nowBy[ip.Name] = ip
|
||||
if ip.Repo == RepoForeign {
|
||||
foreign[ip.Name] = true
|
||||
}
|
||||
}
|
||||
recBy := map[string]InstalledPackage{}
|
||||
for _, ip := range rec.Installed.Packages {
|
||||
recBy[ip.Name] = ip
|
||||
if ip.Repo == RepoForeign {
|
||||
foreign[ip.Name] = true
|
||||
}
|
||||
}
|
||||
p.Foreign = len(foreign)
|
||||
for _, want := range rec.Installed.Packages {
|
||||
if foreign[want.Name] {
|
||||
continue
|
||||
}
|
||||
have, installedNow := nowBy[want.Name]
|
||||
if installedNow && have.Version == want.Version {
|
||||
continue
|
||||
}
|
||||
arch := ""
|
||||
if cache != nil {
|
||||
arch = cache.arch[want.Name]
|
||||
}
|
||||
if arch == "" && local && installedNow {
|
||||
arch = m.installedArch(have.Name, have.Version)
|
||||
}
|
||||
src, loc := source(want.Name, want.Version, want.Repo, arch, cache, rec.At)
|
||||
p.Items = append(p.Items, RollbackItem{Name: want.Name, From: have.Version, To: want.Version, Source: src, Location: loc})
|
||||
}
|
||||
for _, have := range now.Packages {
|
||||
if _, inRecord := recBy[have.Name]; inRecord || foreign[have.Name] {
|
||||
continue
|
||||
}
|
||||
p.Added = append(p.Added, PackageVersion{have.Name, have.Version})
|
||||
if removeAdded {
|
||||
p.Items = append(p.Items, RollbackItem{Name: have.Name, From: have.Version, Source: SourceRemoved})
|
||||
}
|
||||
}
|
||||
sort.Slice(p.Items, func(i, j int) bool { return p.Items[i].Name < p.Items[j].Name })
|
||||
return p, nil
|
||||
}
|
||||
|
||||
// AskRollback writes an ask for a rollback of node to the record of run, for that node's keeper to take.
|
||||
func AskRollback(asks Bucket, node, run string, removeAdded bool, now time.Time) (Ask, error) {
|
||||
a := Ask{ID: AskID(now), Node: node, Act: ActRollback, At: now.UTC(), Of: run, RemoveAdded: removeAdded}
|
||||
if _, err := asks.Create(NodeKey(node, a.ID), a); err != nil {
|
||||
return Ask{}, fmt.Errorf("writing the ask: %w", err)
|
||||
}
|
||||
return a, nil
|
||||
}
|
||||
|
||||
// StartRollback runs a rollback's steps up to its transaction's start: the plan made again on the node, and the
|
||||
// transaction started by the root helper with it. The end and the check follow as for an upgrade (Follow).
|
||||
func (u *Upgrades) StartRollback(of string, removeAdded bool, ask string) (Run, error) {
|
||||
now := u.Machine.Now().UTC()
|
||||
r := &openRun{Run: Run{Node: u.Node, Run: RunID(now), Kind: RunRollback, From: FromAsk, Ask: ask, Of: of}}
|
||||
runs, err := u.ownRuns()
|
||||
if err != nil {
|
||||
return r.Run, err
|
||||
}
|
||||
for _, o := range runs {
|
||||
if o.Run.Run == r.Run.Run {
|
||||
return r.Run, fmt.Errorf("run %s of %s is there already: one run a second", o.Run.Run, u.Node)
|
||||
}
|
||||
if !ended(o.Run) && o.From != FromLog {
|
||||
return r.Run, u.refuse(r, fmt.Sprintf("run %s is still going (its last step is %s): one run at a time", o.Run.Run, lastStep(o.Run)))
|
||||
}
|
||||
}
|
||||
if u.Machine.locked() {
|
||||
return r.Run, u.refuse(r, "another pacman holds "+DBLock+": a transaction is running, or one was killed and left its lock")
|
||||
}
|
||||
plan, err := PlanRollback(u.Machine, u.Installed, u.Records, u.Node, of, removeAdded, true)
|
||||
if err != nil {
|
||||
return r.Run, u.refuse(r, err.Error())
|
||||
}
|
||||
if len(plan.Items) == 0 {
|
||||
return r.Run, u.refuse(r, "the node's packages are those of the record already: there is nothing to bring back")
|
||||
}
|
||||
unit := TransactionUnit(RunRollback, r.Run.Run)
|
||||
if err := u.Root.Rollback(r.Run.Run, plan.Items); err != nil {
|
||||
return r.Run, u.end(r, Step{Step: StepFailed, At: u.Machine.Now().UTC(), Error: "the transaction could not be started: " + err.Error()})
|
||||
}
|
||||
if err := u.step(r, Step{Step: StepStarted, At: now, Unit: unit, Plan: plan.Items}); err != nil {
|
||||
return r.Run, err
|
||||
}
|
||||
return r.Run, u.emit(EventRollbackStarted, RollbackStarted{Node: u.Node, Run: r.Run.Run, Of: of, Unit: unit})
|
||||
}
|
||||
|
||||
// takeRollback carries out an ask for a rollback; the ask is deleted once a run names it.
|
||||
func (u *Upgrades) takeRollback(key string, a Ask) error {
|
||||
if a.Of == "" {
|
||||
return errors.Join(fmt.Errorf("ask %s names no run to return to", key), u.Asks.Delete(key))
|
||||
}
|
||||
r, err := u.StartRollback(a.Of, a.RemoveAdded, a.ID)
|
||||
if len(r.Steps) == 0 {
|
||||
return err
|
||||
}
|
||||
if derr := u.Asks.Delete(key); derr != nil {
|
||||
err = errors.Join(err, fmt.Errorf("deleting the ask %s: %w", key, derr))
|
||||
}
|
||||
return err
|
||||
}
|
||||
|
||||
// Rollback is pacman_rollback: the plan, read from the bus (and on the node itself its cache), and with apply
|
||||
// the ask that has the node's keeper carry it out.
|
||||
func (m *Machine) Rollback(node, run string, removeAdded, apply bool) (map[string]any, error) {
|
||||
if node == "" {
|
||||
node = m.Node
|
||||
}
|
||||
if node == "" {
|
||||
return nil, errors.New("node is required: this bundle does not know its own")
|
||||
}
|
||||
if err := m.states(); err != nil {
|
||||
return nil, err
|
||||
}
|
||||
plan, err := PlanRollback(m, m.States(StateInstalled), m.States(StateRecords), node, run, removeAdded, node == m.Node)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
out := map[string]any{"plan": plan}
|
||||
if !apply {
|
||||
return out, nil
|
||||
}
|
||||
if len(plan.Items) == 0 {
|
||||
return nil, fmt.Errorf("the packages of %s are those of the record already: there is nothing to bring back", node)
|
||||
}
|
||||
a, err := AskRollback(upgradeState(StateAsks), node, run, removeAdded, m.Now())
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
out["asked"] = a
|
||||
out["follow"] = "pacman_upgrades with node " + node + " lists the rollback run that names this ask; its keeper makes the plan again from its own cache"
|
||||
return out, nil
|
||||
}
|
||||
@@ -0,0 +1,195 @@
|
||||
package main
|
||||
|
||||
// A rollback to a record (novox/hq ADR 0508 §7; acceptance 1841/AC8): the plan says each package now -> record with
|
||||
// its source, the packages added and no foreign package; apply is an ask the node's keeper runs as one
|
||||
// transaction, emitting rollback.started and rollback.finished and ending with the after-upgrade check.
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
)
|
||||
|
||||
var recordAt = time.Date(2026, 10, 3, 4, 0, 12, 0, time.UTC)
|
||||
|
||||
// rollbackRecord is node-a's record of run 20261003T040012Z; rollbackNow its packages since.
|
||||
var (
|
||||
rollbackRecord = Record{Node: "node-a", Run: "20261003T040012Z", At: recordAt, Installed: InstalledValue{At: recordAt, Packages: []InstalledPackage{
|
||||
{"bash", "5.3.15-1", true, "core"}, {"glibc", "2.42-1", false, "core"}, {"linux", "7.2.8.arch1-2", true, "core"},
|
||||
{"made-up-old-library", "1.0-1", false, "extra"}, {"made-up-foreign", "1-1", true, RepoForeign},
|
||||
}}}
|
||||
rollbackNow = InstalledValue{At: recordAt.Add(time.Hour), Packages: []InstalledPackage{
|
||||
{"bash", "5.3.20-1", true, "core"}, {"glibc", "2.42-1", false, "core"}, {"linux", "7.3.1.arch1-1", true, "core"},
|
||||
{"made-up-foreign", "2-1", true, RepoForeign}, {"made-up-library", "1.0-1", false, "extra"},
|
||||
}}
|
||||
)
|
||||
|
||||
// rollbackCache is the node's package cache: bash's recorded version, and linux at its new version only.
|
||||
const rollbackCache = "bash-5.3.15-1-x86_64.pkg.tar.zst\nbash-5.3.20-1-x86_64.pkg.tar.zst\nlinux-7.3.1.arch1-1-x86_64.pkg.tar.zst\n" +
|
||||
"not-a-package-file\n"
|
||||
|
||||
func rollbackStates(t *testing.T) (*fakeStates, *Machine, *[]call) {
|
||||
t.Helper()
|
||||
states := &fakeStates{}
|
||||
states.store(StateRecords).Put(NodeKey("node-a", rollbackRecord.Run), rollbackRecord)
|
||||
states.store(StateInstalled).Put("node-a", rollbackNow)
|
||||
var calls []call
|
||||
m := machine(byLine(map[string]Ran{
|
||||
"find " + CacheDir + " -mindepth 1 -maxdepth 1 -name *.pkg.tar.* -printf %f\n": {Stdout: rollbackCache},
|
||||
}, &calls), 1000)
|
||||
m.Node, m.States = "node-a", readerOf(states)
|
||||
return states, m, &calls
|
||||
}
|
||||
|
||||
func TestThePlanSaysEachPackageNowToTheRecordWithItsSourceTheAddedAndNoForeign(t *testing.T) {
|
||||
_, m, calls := rollbackStates(t)
|
||||
p, err := PlanRollback(m, m.States(StateInstalled), m.States(StateRecords), "node-a", rollbackRecord.Run, false, true)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
want := []RollbackItem{
|
||||
{"bash", "5.3.20-1", "5.3.15-1", SourceCache, CacheDir + "/bash-5.3.15-1-x86_64.pkg.tar.zst"},
|
||||
{"linux", "7.3.1.arch1-1", "7.2.8.arch1-2", SourceArchive, "https://archive.archlinux.org/packages/l/linux/linux-7.2.8.arch1-2-x86_64.pkg.tar.zst"},
|
||||
{"made-up-old-library", "", "1.0-1", SourceRepository, "https://archive.archlinux.org/repos/2026/10/03/extra/os/$arch"},
|
||||
}
|
||||
if len(p.Items) != len(want) {
|
||||
t.Fatalf("items: %+v", p.Items)
|
||||
}
|
||||
for i := range want {
|
||||
if p.Items[i] != want[i] {
|
||||
t.Errorf("item %d: %+v, want %+v", i, p.Items[i], want[i])
|
||||
}
|
||||
}
|
||||
if len(p.Added) != 1 || p.Added[0] != (PackageVersion{"made-up-library", "1.0-1"}) || p.Foreign != 1 || !p.CacheRead {
|
||||
t.Errorf("added %+v, foreign %d, cache %v", p.Added, p.Foreign, p.CacheRead)
|
||||
}
|
||||
for _, c := range *calls {
|
||||
if strings.Contains(c.String(), "made-up-foreign") {
|
||||
t.Errorf("a command named the foreign package: %s", c)
|
||||
}
|
||||
}
|
||||
// asked to remove what was added: the added package is removed too.
|
||||
p, _ = PlanRollback(m, m.States(StateInstalled), m.States(StateRecords), "node-a", rollbackRecord.Run, true, true)
|
||||
if len(p.Items) != 4 || p.Items[2] != (RollbackItem{Name: "made-up-library", From: "1.0-1", Source: SourceRemoved}) {
|
||||
t.Errorf("with remove_added: %+v", p.Items)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAnotherNodesPlanReadsNoCacheAndTakesUnknownArchitecturesFromTheDatedRepository(t *testing.T) {
|
||||
_, m, calls := rollbackStates(t)
|
||||
m.Node = "node-b"
|
||||
r, err := m.Rollback("node-a", rollbackRecord.Run, false, false)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
p := r["plan"].(RollbackPlan)
|
||||
if p.CacheRead || len(*calls) != 0 {
|
||||
t.Errorf("another node's cache was read: %v", *calls)
|
||||
}
|
||||
for _, i := range p.Items {
|
||||
if i.Source != SourceRepository || !strings.HasPrefix(i.Location, "https://archive.archlinux.org/repos/2026/10/03/") {
|
||||
t.Errorf("%+v", i)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestNoPlanFromPackagesNotCurrentOrWithoutARecord(t *testing.T) {
|
||||
states, m, _ := rollbackStates(t)
|
||||
if _, err := m.Rollback("node-a", "20250101T000000Z", false, false); err == nil || !strings.Contains(err.Error(), "no record") {
|
||||
t.Errorf("no record: %v", err)
|
||||
}
|
||||
states.store(StateInstalled).Put(NotCurrentKey("node-a"), NotCurrent{Bytes: 300000, Since: recordAt})
|
||||
if _, err := m.Rollback("node-a", rollbackRecord.Run, false, false); err == nil || !strings.Contains(err.Error(), "not current") {
|
||||
t.Errorf("not current: %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
func TestApplyWritesAnAskForTheNodesKeeper(t *testing.T) {
|
||||
_, m, _ := rollbackStates(t)
|
||||
asks := newBucket()
|
||||
saved := upgradeState
|
||||
upgradeState = func(name string) Bucket { return asks }
|
||||
t.Cleanup(func() { upgradeState = saved })
|
||||
r, err := m.Rollback("", rollbackRecord.Run, true, true)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
a := r["asked"].(Ask)
|
||||
var got Ask
|
||||
if !asks.value(NodeKey("node-a", a.ID), &got) || got.Act != ActRollback || got.Of != rollbackRecord.Run || !got.RemoveAdded {
|
||||
t.Errorf("the ask: %+v", got)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAnAskedRollbackIsOneTransactionWithItsEventsAndTheCheck(t *testing.T) {
|
||||
k := newKeeperBed(t)
|
||||
k.records.Put(NodeKey("node-a", rollbackRecord.Run), rollbackRecord)
|
||||
k.installed.Put("node-a", rollbackNow)
|
||||
inner := k.u.Machine.Run
|
||||
k.u.Machine.Run = func(ctx context.Context, name string, args ...string) Ran {
|
||||
if name == "find" {
|
||||
return Ran{Stdout: rollbackCache}
|
||||
}
|
||||
return inner(ctx, name, args...)
|
||||
}
|
||||
a := Ask{ID: "20261004T120000Z-abc123", Node: "node-a", Act: ActRollback, At: k.now, Of: rollbackRecord.Run}
|
||||
if err := k.takeAsk(t, a); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
var r Run
|
||||
keys, _ := k.runs.Keys()
|
||||
for _, key := range keys {
|
||||
var x Run
|
||||
k.runs.value(key, &x)
|
||||
if x.Kind == RunRollback {
|
||||
r = x
|
||||
}
|
||||
}
|
||||
if r.Of != rollbackRecord.Run || r.Ask != a.ID || steps(r) != "started" || len(r.Steps[0].Plan) != 3 {
|
||||
t.Fatalf("the rollback run: %+v", r)
|
||||
}
|
||||
if strings.Join(k.root.started, ";") != "rollback "+r.Run+" plan=bash:5.3.15-1:cache,linux:7.2.8.arch1-2:archive,made-up-old-library:1.0-1:repository" {
|
||||
t.Errorf("the root helper was asked: %v", k.root.started)
|
||||
}
|
||||
if _, there := k.asks.values[NodeKey("node-a", a.ID)]; there {
|
||||
t.Error("the ask is deleted once the run names it")
|
||||
}
|
||||
k.unitState, k.now = "LoadState=not-found\n", k.now.Add(time.Minute)
|
||||
if err := k.u.Follow(); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
k.runs.value(NodeKey("node-a", r.Run), &r)
|
||||
if steps(r) != "started finished checked-after" || r.Steps[2].Check == nil {
|
||||
t.Errorf("steps: %s", steps(r))
|
||||
}
|
||||
if got := strings.Join(k.events.names(), " "); got != "rollback.started rollback.finished upgrade.checked" {
|
||||
t.Errorf("events: %s", got)
|
||||
}
|
||||
// the same record again: nothing to bring back once the packages are the record's.
|
||||
k.installed.Put("node-a", rollbackRecord.Installed)
|
||||
k.now = k.now.Add(time.Minute)
|
||||
if err := k.takeAsk(t, Ask{ID: "20261004T120200Z-def456", Node: "node-a", Act: ActRollback, At: k.now, Of: rollbackRecord.Run}); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
var again Run
|
||||
k.runs.value(NodeKey("node-a", "20261004T120200Z"), &again)
|
||||
if steps(again) != "refused" || !strings.Contains(again.Steps[0].Why, "nothing to bring back") {
|
||||
t.Errorf("a rollback with nothing to do: %+v", again)
|
||||
}
|
||||
}
|
||||
|
||||
// takeAsk writes an ask and hands its change to the keeper, as the watch of `asks` would.
|
||||
func (k *keeperBed) takeAsk(t *testing.T, a Ask) error {
|
||||
t.Helper()
|
||||
key := NodeKey(a.Node, a.ID)
|
||||
rev, err := k.asks.Create(key, a)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
raw, _ := json.Marshal(a)
|
||||
return k.u.TakeAsk(stdio.StateChange{State: StateAsks, Key: key, Op: "put", Value: raw, Revision: rev})
|
||||
}
|
||||
@@ -0,0 +1,506 @@
|
||||
package main
|
||||
|
||||
// The root helper (novox/hq ADR 0510): the module's acts that need root go through a second process of the same
|
||||
// build, `pacman-root`, running as root with no bus account, never through a sudo rule. It serves a closed list
|
||||
// of verbs on a unix socket only the keeper's account opens, the peer's account checked on every connection:
|
||||
//
|
||||
// - transaction: start a run's full upgrade as a unit of the service manager, named by the run, leaving out
|
||||
// the packages the run holds back;
|
||||
// - rollback: start a rollback's transaction as such a unit, carrying out the plan sent with the verb;
|
||||
// - maps: the processes still mapping a file an upgrade replaced, by unit or container and executable;
|
||||
// - journal: the last lines a transaction unit of the module logged.
|
||||
//
|
||||
// Every other verb is refused, and every argument that is not a unit name of the module's own transactions, a
|
||||
// package name, a version, a run's id, or a plan item's location in the package cache or the Arch Linux
|
||||
// Archive. The helper's program is root's (ADR 0510 §3): the node-engine places a process's program in a
|
||||
// directory its own account owns, and that account is root.
|
||||
//
|
||||
// A rollback's unit runs this same program (`pacman-tools root-rollback <plan>`) on a plan file the helper
|
||||
// wrote into a directory only root reads: it finds each package of the dated repository by its file's name,
|
||||
// installs the plan's packages in one pacman transaction, then removes the ones the rollback was asked to
|
||||
// remove. Nothing is changed when a package cannot be found.
|
||||
|
||||
import (
|
||||
"bufio"
|
||||
"context"
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"fmt"
|
||||
"io"
|
||||
"net"
|
||||
"net/http"
|
||||
"os"
|
||||
"os/exec"
|
||||
"os/signal"
|
||||
"os/user"
|
||||
"path"
|
||||
"path/filepath"
|
||||
"regexp"
|
||||
"runtime"
|
||||
"strconv"
|
||||
"strings"
|
||||
"syscall"
|
||||
"time"
|
||||
|
||||
"golang.org/x/sys/unix"
|
||||
)
|
||||
|
||||
// The helper's verbs.
|
||||
const (
|
||||
VerbTransaction = "transaction"
|
||||
VerbRollback = "rollback"
|
||||
VerbMaps = "maps"
|
||||
VerbJournal = "journal"
|
||||
)
|
||||
|
||||
// The helper's environment, and the keeper's word for where the helper listens.
|
||||
const (
|
||||
EnvRootSocket = "PACMAN_ROOT_SOCKET" // the helper's socket, for the helper and the keeper
|
||||
EnvRootPeer = "PACMAN_ROOT_PEER" // the only account the helper answers: the keeper's
|
||||
// RootSocketPath is the socket as the manifest names it: in a directory of the module's state the helper
|
||||
// makes, root's.
|
||||
RootSocketPath = "${dir:state}/root/helper.sock"
|
||||
)
|
||||
|
||||
// rootTimeout bounds one exchange with the helper: every verb answers at once (a transaction is started, not
|
||||
// waited for).
|
||||
const rootTimeout = 30 * time.Second
|
||||
|
||||
// mostJournalLines is the most lines the journal verb reads.
|
||||
const mostJournalLines = 400
|
||||
|
||||
// RootRequest is one verb sent to the helper, one per connection.
|
||||
type RootRequest struct {
|
||||
Verb string `json:"verb"`
|
||||
Kind string `json:"kind,omitempty"`
|
||||
Run string `json:"run,omitempty"`
|
||||
Ignore []string `json:"ignore,omitempty"`
|
||||
Plan []RollbackItem `json:"plan,omitempty"`
|
||||
Unit string `json:"unit,omitempty"`
|
||||
Lines int `json:"lines,omitempty"`
|
||||
}
|
||||
|
||||
// RootAnswer is the helper's answer: an error, or what the verb read.
|
||||
type RootAnswer struct {
|
||||
Error string `json:"error,omitempty"`
|
||||
Lines []string `json:"lines,omitempty"`
|
||||
Mapped []Mapped `json:"mapped,omitempty"`
|
||||
}
|
||||
|
||||
// RootHelper is the helper: what it runs, the process table it reads, where it keeps plans, its own program,
|
||||
// and the one account it answers.
|
||||
type RootHelper struct {
|
||||
Run Runner
|
||||
Proc ProcFS
|
||||
ProcRoot string
|
||||
PlanDir string
|
||||
Self string
|
||||
Peer int
|
||||
// PeerOf is the account at the other end of a connection; nil reads it from the socket (SO_PEERCRED).
|
||||
PeerOf func(net.Conn) (int, error)
|
||||
Logf func(format string, a ...any)
|
||||
}
|
||||
|
||||
var (
|
||||
// runIDShape is a run's id (RunID): its start in UTC to the second.
|
||||
runIDShape = regexp.MustCompile(`^[0-9]{8}T[0-9]{6}Z$`)
|
||||
// ownUnit is a transaction unit of the module's own runs (TransactionUnit).
|
||||
ownUnit = regexp.MustCompile(`^` + unitPrefix + `(` + RunUpgrade + `|` + RunRollback + `)-[0-9]{8}t[0-9]{6}z$`)
|
||||
// archArg is a package's architecture in a file's name.
|
||||
archArg = regexp.MustCompile(`^[a-z0-9_]+$`)
|
||||
// datedRepository is the archive's repository as it stood on one day, its architecture left to pacman.
|
||||
datedRepository = regexp.MustCompile(`^` + regexp.QuoteMeta(ArchiveURL) + `/repos/[0-9]{4}/[0-9]{2}/[0-9]{2}/[A-Za-z0-9_-]+/os/\$arch$`)
|
||||
)
|
||||
|
||||
func refused(format string, a ...any) RootAnswer {
|
||||
return RootAnswer{Error: "refused: " + fmt.Sprintf(format, a...)}
|
||||
}
|
||||
|
||||
func checkRun(run string) error {
|
||||
if !runIDShape.MatchString(run) {
|
||||
return fmt.Errorf("%q is not a run's id", run)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// checkItem refuses a plan item that is not a package brought back from the cache or the archive, or removed.
|
||||
func checkItem(i RollbackItem) error {
|
||||
if !packageName.MatchString(i.Name) {
|
||||
return fmt.Errorf("%q is not a package's name", i.Name)
|
||||
}
|
||||
if i.Source == SourceRemoved {
|
||||
if i.To != "" || i.Location != "" {
|
||||
return fmt.Errorf("%s is removed, and names a version or a place to take it from", i.Name)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
if !packageVersion.MatchString(i.To) {
|
||||
return fmt.Errorf("%q is not a version of %s", i.To, i.Name)
|
||||
}
|
||||
switch i.Source {
|
||||
case SourceCache:
|
||||
f := cacheFile.FindStringSubmatch(path.Base(i.Location))
|
||||
if path.Clean(i.Location) != i.Location || path.Dir(i.Location) != CacheDir || f == nil || f[1] != i.Name || f[2] != i.To {
|
||||
return fmt.Errorf("%q is not %s %s's file in the package cache", i.Location, i.Name, i.To)
|
||||
}
|
||||
case SourceArchive:
|
||||
prefix := ArchiveURL + "/packages/" + i.Name[:1] + "/" + i.Name + "/" + i.Name + "-" + i.To + "-"
|
||||
arch, ok := strings.CutPrefix(i.Location, prefix)
|
||||
if !ok || !strings.HasSuffix(arch, ".pkg.tar.zst") || !archArg.MatchString(strings.TrimSuffix(arch, ".pkg.tar.zst")) {
|
||||
return fmt.Errorf("%q is not %s %s's file in the archive", i.Location, i.Name, i.To)
|
||||
}
|
||||
case SourceRepository:
|
||||
if !datedRepository.MatchString(i.Location) {
|
||||
return fmt.Errorf("%q is not a dated repository of the archive", i.Location)
|
||||
}
|
||||
default:
|
||||
return fmt.Errorf("%q is not a source a package is brought back from", i.Source)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// Handle carries out one verb, or refuses it naming why; nothing runs for a refused verb.
|
||||
func (h *RootHelper) Handle(req RootRequest) RootAnswer {
|
||||
switch req.Verb {
|
||||
case VerbTransaction:
|
||||
if req.Kind != RunUpgrade {
|
||||
return refused("a transaction is a full upgrade, not %q", req.Kind)
|
||||
}
|
||||
if err := checkRun(req.Run); err != nil {
|
||||
return refused("%v", err)
|
||||
}
|
||||
for _, n := range req.Ignore {
|
||||
if !packageName.MatchString(n) {
|
||||
return refused("%q is not a package's name to leave out", n)
|
||||
}
|
||||
}
|
||||
args := append([]string{"pacman"}, UpgradeArgs...)
|
||||
if len(req.Ignore) > 0 {
|
||||
args = append(args, "--ignore", strings.Join(req.Ignore, ","))
|
||||
}
|
||||
return h.startUnit(TransactionUnit(RunUpgrade, req.Run), "pacman full upgrade of run "+req.Run, args)
|
||||
case VerbRollback:
|
||||
if err := checkRun(req.Run); err != nil {
|
||||
return refused("%v", err)
|
||||
}
|
||||
if len(req.Plan) == 0 {
|
||||
return refused("a rollback with no plan")
|
||||
}
|
||||
for _, i := range req.Plan {
|
||||
if err := checkItem(i); err != nil {
|
||||
return refused("%v", err)
|
||||
}
|
||||
}
|
||||
raw, err := json.Marshal(req.Plan)
|
||||
if err != nil {
|
||||
return RootAnswer{Error: err.Error()}
|
||||
}
|
||||
if err := os.MkdirAll(h.PlanDir, 0o700); err != nil {
|
||||
return RootAnswer{Error: "the plan's directory: " + err.Error()}
|
||||
}
|
||||
file := filepath.Join(h.PlanDir, req.Run+".json")
|
||||
if err := os.WriteFile(file, raw, 0o600); err != nil {
|
||||
return RootAnswer{Error: "writing the plan: " + err.Error()}
|
||||
}
|
||||
return h.startUnit(TransactionUnit(RunRollback, req.Run), "pacman rollback of run "+req.Run, []string{h.Self, "root-rollback", file})
|
||||
case VerbMaps:
|
||||
mapped, err := MappedDeleted(h.Proc, h.ProcRoot)
|
||||
if err != nil {
|
||||
return RootAnswer{Error: "reading the process table: " + err.Error()}
|
||||
}
|
||||
return RootAnswer{Mapped: mapped}
|
||||
case VerbJournal:
|
||||
if !ownUnit.MatchString(req.Unit) {
|
||||
return refused("%q is not a transaction unit of the module", req.Unit)
|
||||
}
|
||||
if req.Lines < 1 || req.Lines > mostJournalLines {
|
||||
return refused("a journal is read 1 to %d lines at a time, not %d", mostJournalLines, req.Lines)
|
||||
}
|
||||
r := h.Run(context.Background(), "journalctl", "--no-pager", "-o", "cat", "-n", strconv.Itoa(req.Lines), "-u", req.Unit)
|
||||
if r.Status != 0 || r.Err != "" {
|
||||
return RootAnswer{Error: failure("journalctl", "journalctl", r).Error()}
|
||||
}
|
||||
return RootAnswer{Lines: lines(r.Stdout)}
|
||||
}
|
||||
return refused("%q is not a verb of the root helper (%s, %s, %s, %s)", req.Verb, VerbTransaction, VerbRollback, VerbMaps, VerbJournal)
|
||||
}
|
||||
|
||||
// startUnit starts command as a transient unit of the service manager and answers once it started.
|
||||
func (h *RootHelper) startUnit(unit, description string, command []string) RootAnswer {
|
||||
args := append([]string{"--unit=" + unit, "--description=" + description + ", started by the mesh's pacman keeper", "--quiet"}, command...)
|
||||
r := h.Run(context.Background(), "systemd-run", args...)
|
||||
if r.Status != 0 || r.Err != "" {
|
||||
return RootAnswer{Error: failure("systemd-run", "systemd-run", r).Error()}
|
||||
}
|
||||
return RootAnswer{}
|
||||
}
|
||||
|
||||
// peerCred is the account at the other end of a unix socket.
|
||||
func peerCred(c net.Conn) (int, error) {
|
||||
uc, ok := c.(*net.UnixConn)
|
||||
if !ok {
|
||||
return -1, errors.New("not a unix socket")
|
||||
}
|
||||
raw, err := uc.SyscallConn()
|
||||
if err != nil {
|
||||
return -1, err
|
||||
}
|
||||
uid := -1
|
||||
var gerr error
|
||||
if err := raw.Control(func(fd uintptr) {
|
||||
cred, err := unix.GetsockoptUcred(int(fd), unix.SOL_SOCKET, unix.SO_PEERCRED)
|
||||
if err != nil {
|
||||
gerr = err
|
||||
return
|
||||
}
|
||||
uid = int(cred.Uid)
|
||||
}); err != nil {
|
||||
return -1, err
|
||||
}
|
||||
return uid, gerr
|
||||
}
|
||||
|
||||
// Serve answers each connection on l, one request each, until ctx ends: a peer that is not the keeper's
|
||||
// account is told so and nothing it sent is read.
|
||||
func (h *RootHelper) Serve(ctx context.Context, l net.Listener) error {
|
||||
go func() { <-ctx.Done(); l.Close() }()
|
||||
peerOf := h.PeerOf
|
||||
if peerOf == nil {
|
||||
peerOf = peerCred
|
||||
}
|
||||
for {
|
||||
c, err := l.Accept()
|
||||
if err != nil {
|
||||
if ctx.Err() != nil {
|
||||
return nil
|
||||
}
|
||||
return err
|
||||
}
|
||||
go h.serveOne(c, peerOf)
|
||||
}
|
||||
}
|
||||
|
||||
func (h *RootHelper) serveOne(c net.Conn, peerOf func(net.Conn) (int, error)) {
|
||||
defer c.Close()
|
||||
_ = c.SetDeadline(time.Now().Add(rootTimeout))
|
||||
answer := func(a RootAnswer) { _ = json.NewEncoder(c).Encode(a) }
|
||||
uid, err := peerOf(c)
|
||||
if err != nil || uid != h.Peer {
|
||||
if h.Logf != nil {
|
||||
h.Logf("[pacman] root helper: a connection from account %d refused: only the keeper's (%d) is answered", uid, h.Peer)
|
||||
}
|
||||
answer(refused("only the keeper's account is answered"))
|
||||
return
|
||||
}
|
||||
var req RootRequest
|
||||
if err := json.NewDecoder(io.LimitReader(c, 4<<20)).Decode(&req); err != nil {
|
||||
answer(refused("no request: %v", err))
|
||||
return
|
||||
}
|
||||
a := h.Handle(req)
|
||||
if h.Logf != nil {
|
||||
said := "carried out"
|
||||
if a.Error != "" {
|
||||
said = a.Error
|
||||
}
|
||||
h.Logf("[pacman] root helper: %s %s %s: %s", req.Verb, req.Kind, req.Run+req.Unit, said)
|
||||
}
|
||||
answer(a)
|
||||
}
|
||||
|
||||
// rootClient is the keeper's side: the RootActs over the helper's socket.
|
||||
type rootClient struct{ Socket string }
|
||||
|
||||
func (c rootClient) ask(req RootRequest) (RootAnswer, error) {
|
||||
conn, err := net.DialTimeout("unix", c.Socket, rootTimeout)
|
||||
if err != nil {
|
||||
return RootAnswer{}, fmt.Errorf("the root helper cannot be reached at %s: %w", c.Socket, err)
|
||||
}
|
||||
defer conn.Close()
|
||||
_ = conn.SetDeadline(time.Now().Add(rootTimeout))
|
||||
if err := json.NewEncoder(conn).Encode(req); err != nil {
|
||||
return RootAnswer{}, fmt.Errorf("asking the root helper: %w", err)
|
||||
}
|
||||
var a RootAnswer
|
||||
if err := json.NewDecoder(bufio.NewReader(conn)).Decode(&a); err != nil {
|
||||
return RootAnswer{}, fmt.Errorf("the root helper's answer: %w", err)
|
||||
}
|
||||
if a.Error != "" {
|
||||
return a, fmt.Errorf("the root helper: %s", a.Error)
|
||||
}
|
||||
return a, nil
|
||||
}
|
||||
|
||||
func (c rootClient) Transaction(kind, run string, ignore []string) error {
|
||||
_, err := c.ask(RootRequest{Verb: VerbTransaction, Kind: kind, Run: run, Ignore: ignore})
|
||||
return err
|
||||
}
|
||||
|
||||
func (c rootClient) Rollback(run string, plan []RollbackItem) error {
|
||||
_, err := c.ask(RootRequest{Verb: VerbRollback, Run: run, Plan: plan})
|
||||
return err
|
||||
}
|
||||
|
||||
func (c rootClient) Journal(unit string, n int) ([]string, error) {
|
||||
a, err := c.ask(RootRequest{Verb: VerbJournal, Unit: unit, Lines: n})
|
||||
return a.Lines, err
|
||||
}
|
||||
|
||||
func (c rootClient) Maps() ([]Mapped, error) {
|
||||
a, err := c.ask(RootRequest{Verb: VerbMaps})
|
||||
if a.Mapped == nil && err == nil {
|
||||
a.Mapped = []Mapped{}
|
||||
}
|
||||
return a.Mapped, err
|
||||
}
|
||||
|
||||
// rootHelper is `pacman-tools root`: it listens on its socket, made the keeper's alone, until stopped.
|
||||
func rootHelper() int {
|
||||
logf := func(f string, a ...any) { fmt.Fprintf(os.Stderr, f+"\n", a...) }
|
||||
if os.Getuid() != 0 {
|
||||
logf("[pacman] root helper: it runs as root, and runs as account %d", os.Getuid())
|
||||
return 2
|
||||
}
|
||||
sock, peerName := os.Getenv(EnvRootSocket), os.Getenv(EnvRootPeer)
|
||||
if sock == "" || peerName == "" {
|
||||
logf("[pacman] root helper was started without %s or %s", EnvRootSocket, EnvRootPeer)
|
||||
return 2
|
||||
}
|
||||
peer, err := user.Lookup(peerName)
|
||||
if err != nil {
|
||||
logf("[pacman] root helper: the keeper's account %s: %v", peerName, err)
|
||||
return 1
|
||||
}
|
||||
uid, err := strconv.Atoi(peer.Uid)
|
||||
if err != nil {
|
||||
logf("[pacman] root helper: account %s has no number: %v", peerName, err)
|
||||
return 1
|
||||
}
|
||||
self, err := os.Executable()
|
||||
if err != nil {
|
||||
logf("[pacman] root helper: its own program: %v", err)
|
||||
return 1
|
||||
}
|
||||
dir := filepath.Dir(sock)
|
||||
if err := os.MkdirAll(dir, 0o711); err != nil || os.Chown(dir, 0, 0) != nil || os.Chmod(dir, 0o711) != nil {
|
||||
logf("[pacman] root helper: its directory %s cannot be made root's: %v", dir, err)
|
||||
return 1
|
||||
}
|
||||
_ = os.Remove(sock)
|
||||
l, err := net.Listen("unix", sock)
|
||||
if err != nil {
|
||||
logf("[pacman] root helper: listening on %s: %v", sock, err)
|
||||
return 1
|
||||
}
|
||||
if err := os.Chown(sock, uid, 0); err != nil || os.Chmod(sock, 0o600) != nil {
|
||||
logf("[pacman] root helper: the socket cannot be made the keeper's alone: %v", err)
|
||||
return 1
|
||||
}
|
||||
ctx, stop := signal.NotifyContext(context.Background(), os.Interrupt, syscall.SIGTERM)
|
||||
defer stop()
|
||||
h := &RootHelper{Run: ExecRunner, Proc: osProc{}, ProcRoot: "/proc", PlanDir: filepath.Join(dir, "plans"), Self: self, Peer: uid, Logf: logf}
|
||||
logf("[pacman] root helper listening on %s for account %s (%d)", sock, peerName, uid)
|
||||
if err := h.Serve(ctx, l); err != nil {
|
||||
logf("[pacman] root helper stopped: %v", err)
|
||||
return 1
|
||||
}
|
||||
return 0
|
||||
}
|
||||
|
||||
// machineArch is this machine's architecture as Arch Linux names it.
|
||||
func machineArch() string {
|
||||
switch runtime.GOARCH {
|
||||
case "amd64":
|
||||
return "x86_64"
|
||||
case "arm64":
|
||||
return "aarch64"
|
||||
}
|
||||
return runtime.GOARCH
|
||||
}
|
||||
|
||||
// archiveHas says whether the archive holds a file, by asking for its head.
|
||||
func archiveHas(url string) bool {
|
||||
c := &http.Client{Timeout: 30 * time.Second}
|
||||
resp, err := c.Head(url)
|
||||
if err != nil {
|
||||
return false
|
||||
}
|
||||
resp.Body.Close()
|
||||
return resp.StatusCode == http.StatusOK
|
||||
}
|
||||
|
||||
// rollbackUnit carries out a rollback plan inside its unit: every package of a dated repository found by its
|
||||
// file's name (this machine's architecture, then any), the plan's packages installed in one transaction, then
|
||||
// the packages to remove removed. Nothing is changed when a package is not found.
|
||||
func rollbackUnit(file, arch string, run func(string, ...string) int, probe func(string) bool, log io.Writer) int {
|
||||
raw, err := os.ReadFile(file)
|
||||
if err != nil {
|
||||
fmt.Fprintf(log, "error: reading the plan: %v\n", err)
|
||||
return 1
|
||||
}
|
||||
var plan []RollbackItem
|
||||
if err := json.Unmarshal(raw, &plan); err != nil {
|
||||
fmt.Fprintf(log, "error: the plan is not one: %v\n", err)
|
||||
return 1
|
||||
}
|
||||
var targets, remove []string
|
||||
for _, i := range plan {
|
||||
if err := checkItem(i); err != nil {
|
||||
fmt.Fprintf(log, "error: %v\n", err)
|
||||
return 1
|
||||
}
|
||||
switch i.Source {
|
||||
case SourceRemoved:
|
||||
remove = append(remove, i.Name)
|
||||
case SourceRepository:
|
||||
base := strings.Replace(i.Location, "$arch", arch, 1)
|
||||
found := ""
|
||||
for _, a := range []string{arch, "any"} {
|
||||
if u := base + "/" + i.Name + "-" + i.To + "-" + a + ".pkg.tar.zst"; probe(u) {
|
||||
found = u
|
||||
break
|
||||
}
|
||||
}
|
||||
if found == "" {
|
||||
fmt.Fprintf(log, "error: %s %s is not in %s: nothing was changed\n", i.Name, i.To, base)
|
||||
return 1
|
||||
}
|
||||
targets = append(targets, found)
|
||||
default:
|
||||
targets = append(targets, i.Location)
|
||||
}
|
||||
}
|
||||
if len(targets) > 0 {
|
||||
if code := run("pacman", append([]string{"-U", "--noconfirm", "--"}, targets...)...); code != 0 {
|
||||
return code
|
||||
}
|
||||
}
|
||||
if len(remove) > 0 {
|
||||
return run("pacman", append([]string{"-R", "--noconfirm", "--"}, remove...)...)
|
||||
}
|
||||
return 0
|
||||
}
|
||||
|
||||
// rootRollback is `pacman-tools root-rollback <plan>`, the rollback unit's command: pacman's output goes to the
|
||||
// unit's journal.
|
||||
func rootRollback(args []string) int {
|
||||
if len(args) != 1 || os.Getuid() != 0 {
|
||||
fmt.Fprintln(os.Stderr, "error: root-rollback takes one plan file, and runs as root in its unit")
|
||||
return 2
|
||||
}
|
||||
run := func(name string, a ...string) int {
|
||||
cmd := exec.Command(name, a...)
|
||||
cmd.Stdout, cmd.Stderr = os.Stdout, os.Stderr
|
||||
cmd.Env = append(os.Environ(), "LC_ALL=C")
|
||||
if err := cmd.Run(); err != nil {
|
||||
var exit *exec.ExitError
|
||||
if errors.As(err, &exit) {
|
||||
return exit.ExitCode()
|
||||
}
|
||||
fmt.Fprintf(os.Stderr, "error: %s: %v\n", name, err)
|
||||
return 1
|
||||
}
|
||||
return 0
|
||||
}
|
||||
return rollbackUnit(args[0], machineArch(), run, archiveHas, os.Stderr)
|
||||
}
|
||||
@@ -0,0 +1,312 @@
|
||||
package main
|
||||
|
||||
// The root helper (novox/hq ADR 0510; acceptance 1841/AC10 as reworded on the issue): each listed verb with valid
|
||||
// arguments is carried out; every other verb, a unit name outside the module's own, a package name with other
|
||||
// characters and a run of no run's shape are refused; a connection from an account other than the keeper's is
|
||||
// refused; the manifest declares `pacman-root` as root with no bus and places nothing under the sudo
|
||||
// configuration.
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"net"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"reflect"
|
||||
"strings"
|
||||
"testing"
|
||||
)
|
||||
|
||||
// rootBed is a helper over a recording runner, a stand-in process table and a plan directory of its own.
|
||||
type rootBed struct {
|
||||
h *RootHelper
|
||||
ran [][]string
|
||||
}
|
||||
|
||||
func newRootBed(t *testing.T) *rootBed {
|
||||
b := &rootBed{}
|
||||
b.h = &RootHelper{
|
||||
Run: func(_ context.Context, name string, args ...string) Ran {
|
||||
b.ran = append(b.ran, append([]string{name}, args...))
|
||||
if name == "journalctl" {
|
||||
return Ran{Stdout: "error: failed to commit transaction\n"}
|
||||
}
|
||||
return Ran{}
|
||||
},
|
||||
Proc: stubProc{"/proc": {"1"}, "/proc/1/maps": {"7f00 r-xp 0 0:1 1 /usr/lib/libc.so.6 (deleted)"}, "/proc/1/cgroup": {"0::/system.slice/sshd.service"}, "/proc/1/exe": {"/usr/bin/sshd"}},
|
||||
ProcRoot: "/proc",
|
||||
PlanDir: t.TempDir(),
|
||||
Self: "/var/lib/mesh/daemons/pacman-root/pacman-tools",
|
||||
}
|
||||
return b
|
||||
}
|
||||
|
||||
// stubProc is a process table in memory: a directory lists its names, a file holds its one line, a link its target.
|
||||
type stubProc map[string][]string
|
||||
|
||||
func (s stubProc) ReadDir(dir string) ([]string, error) { return s[dir], nil }
|
||||
func (s stubProc) ReadFile(name string) ([]byte, error) {
|
||||
v, ok := s[name]
|
||||
if !ok {
|
||||
return nil, os.ErrNotExist
|
||||
}
|
||||
return []byte(strings.Join(v, "\n")), nil
|
||||
}
|
||||
func (s stubProc) Readlink(name string) (string, error) {
|
||||
v, ok := s[name]
|
||||
if !ok {
|
||||
return "", os.ErrNotExist
|
||||
}
|
||||
return v[0], nil
|
||||
}
|
||||
|
||||
const aRun = "20261024T040000Z"
|
||||
|
||||
// testSocket is a socket in the abstract namespace, named for the test: a path under a test's temporary
|
||||
// directory can pass the 108 bytes a socket's path may hold.
|
||||
func testSocket(t *testing.T) string {
|
||||
return fmt.Sprintf("@pacman-root-test-%d-%s", os.Getpid(), t.Name())
|
||||
}
|
||||
|
||||
func TestTheRootHelperStartsAnUpgradeAsTheRunsUnit(t *testing.T) {
|
||||
b := newRootBed(t)
|
||||
a := b.h.Handle(RootRequest{Verb: VerbTransaction, Kind: RunUpgrade, Run: aRun, Ignore: []string{"linux", "linux-headers"}})
|
||||
if a.Error != "" {
|
||||
t.Fatalf("refused: %s", a.Error)
|
||||
}
|
||||
want := []string{"systemd-run", "--unit=" + TransactionUnit(RunUpgrade, aRun), "--description=pacman full upgrade of run " + aRun + ", started by the mesh's pacman keeper",
|
||||
"--quiet", "pacman", "-Syu", "--noconfirm", "--ignore", "linux,linux-headers"}
|
||||
if len(b.ran) != 1 || !reflect.DeepEqual(b.ran[0], want) {
|
||||
t.Fatalf("ran %q, want %q", b.ran, want)
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheRootHelperStartsARollbackWithThePlanInTheRequest(t *testing.T) {
|
||||
b := newRootBed(t)
|
||||
plan := []RollbackItem{
|
||||
{Name: "bash", From: "5.3.20-1", To: "5.3.15-1", Source: SourceCache, Location: CacheDir + "/bash-5.3.15-1-x86_64.pkg.tar.zst"},
|
||||
{Name: "glibc", From: "2.43-2", To: "2.43-1", Source: SourceArchive, Location: ArchiveURL + "/packages/g/glibc/glibc-2.43-1-x86_64.pkg.tar.zst"},
|
||||
{Name: "zlib", From: "1.3.2-1", To: "1.3.1-2", Source: SourceRepository, Location: ArchiveURL + "/repos/2026/10/04/core/os/$arch"},
|
||||
{Name: "made-up-library", From: "1.0-1", Source: SourceRemoved},
|
||||
}
|
||||
a := b.h.Handle(RootRequest{Verb: VerbRollback, Run: aRun, Plan: plan})
|
||||
if a.Error != "" {
|
||||
t.Fatalf("refused: %s", a.Error)
|
||||
}
|
||||
file := filepath.Join(b.h.PlanDir, aRun+".json")
|
||||
want := []string{"systemd-run", "--unit=" + TransactionUnit(RunRollback, aRun), "--description=pacman rollback of run " + aRun + ", started by the mesh's pacman keeper",
|
||||
"--quiet", b.h.Self, "root-rollback", file}
|
||||
if len(b.ran) != 1 || !reflect.DeepEqual(b.ran[0], want) {
|
||||
t.Fatalf("ran %q, want %q", b.ran, want)
|
||||
}
|
||||
raw, err := os.ReadFile(file)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
var got []RollbackItem
|
||||
if err := json.Unmarshal(raw, &got); err != nil || !reflect.DeepEqual(got, plan) {
|
||||
t.Fatalf("the plan file holds %s (%v)", raw, err)
|
||||
}
|
||||
if fi, _ := os.Stat(file); fi.Mode().Perm() != 0o600 {
|
||||
t.Errorf("the plan file is mode %v, want 0600: only root reads what root carries out", fi.Mode().Perm())
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheRootHelperReadsTheReplacedFilesMappings(t *testing.T) {
|
||||
b := newRootBed(t)
|
||||
a := b.h.Handle(RootRequest{Verb: VerbMaps})
|
||||
want := []Mapped{{Unit: "sshd.service", Executable: "/usr/bin/sshd"}}
|
||||
if a.Error != "" || !reflect.DeepEqual(a.Mapped, want) {
|
||||
t.Fatalf("maps answered %+v, want %+v", a, want)
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheRootHelperReadsATransactionUnitsJournal(t *testing.T) {
|
||||
b := newRootBed(t)
|
||||
unit := TransactionUnit(RunUpgrade, aRun)
|
||||
a := b.h.Handle(RootRequest{Verb: VerbJournal, Unit: unit, Lines: 40})
|
||||
if a.Error != "" || !reflect.DeepEqual(a.Lines, []string{"error: failed to commit transaction"}) {
|
||||
t.Fatalf("journal answered %+v", a)
|
||||
}
|
||||
want := []string{"journalctl", "--no-pager", "-o", "cat", "-n", "40", "-u", unit}
|
||||
if !reflect.DeepEqual(b.ran[0], want) {
|
||||
t.Fatalf("ran %q, want %q", b.ran[0], want)
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheRootHelperRefusesEverythingOffItsList(t *testing.T) {
|
||||
good := RollbackItem{Name: "bash", From: "5.3.20-1", To: "5.3.15-1", Source: SourceCache, Location: CacheDir + "/bash-5.3.15-1-x86_64.pkg.tar.zst"}
|
||||
with := func(f func(*RollbackItem)) []RollbackItem { i := good; f(&i); return []RollbackItem{i} }
|
||||
cases := map[string]RootRequest{
|
||||
"another verb": {Verb: "shell", Run: aRun},
|
||||
"an empty verb": {},
|
||||
"another kind of transaction": {Verb: VerbTransaction, Kind: "remove", Run: aRun},
|
||||
"a rollback as a transaction": {Verb: VerbTransaction, Kind: RunRollback, Run: aRun},
|
||||
"a run of no run's shape": {Verb: VerbTransaction, Kind: RunUpgrade, Run: "../../etc"},
|
||||
"a run with an option in it": {Verb: VerbTransaction, Kind: RunUpgrade, Run: "--now"},
|
||||
"an ignored name with a space": {Verb: VerbTransaction, Kind: RunUpgrade, Run: aRun, Ignore: []string{"linux headers"}},
|
||||
"an ignored name as an option": {Verb: VerbTransaction, Kind: RunUpgrade, Run: aRun, Ignore: []string{"--root=/tmp"}},
|
||||
"an ignored name with a comma": {Verb: VerbTransaction, Kind: RunUpgrade, Run: aRun, Ignore: []string{"linux,glibc"}},
|
||||
"an ignored name in capitals": {Verb: VerbTransaction, Kind: RunUpgrade, Run: aRun, Ignore: []string{"Linux"}},
|
||||
"a journal of another unit": {Verb: VerbJournal, Unit: "sshd.service", Lines: 10},
|
||||
"a journal of a look-alike unit": {Verb: VerbJournal, Unit: "mesh-pacman-upgrade-x;reboot", Lines: 10},
|
||||
"a journal of too many lines": {Verb: VerbJournal, Unit: TransactionUnit(RunUpgrade, aRun), Lines: 100000},
|
||||
"a rollback of no plan": {Verb: VerbRollback, Run: aRun},
|
||||
"a rollback of a bad run": {Verb: VerbRollback, Run: "x", Plan: []RollbackItem{good}},
|
||||
"a plan item with a bad name": {Verb: VerbRollback, Run: aRun, Plan: with(func(i *RollbackItem) { i.Name = "bash;rm" })},
|
||||
"a plan item with a bad version": {Verb: VerbRollback, Run: aRun, Plan: with(func(i *RollbackItem) { i.To = "1 2" })},
|
||||
"a cache file outside the cache": {Verb: VerbRollback, Run: aRun, Plan: with(func(i *RollbackItem) { i.Location = "/tmp/bash-5.3.15-1-x86_64.pkg.tar.zst" })},
|
||||
"a cache file climbing out": {Verb: VerbRollback, Run: aRun, Plan: with(func(i *RollbackItem) { i.Location = CacheDir + "/../../../tmp/bash-5.3.15-1-x86_64.pkg.tar.zst" })},
|
||||
"a cache file of another package": {Verb: VerbRollback, Run: aRun, Plan: with(func(i *RollbackItem) { i.Location = CacheDir + "/evil-1-1-x86_64.pkg.tar.zst" })},
|
||||
"an archive URL elsewhere": {Verb: VerbRollback, Run: aRun, Plan: with(func(i *RollbackItem) {
|
||||
i.Source, i.Location = SourceArchive, "https://example.org/packages/b/bash/bash-5.3.15-1-x86_64.pkg.tar.zst"
|
||||
})},
|
||||
"a repository URL elsewhere": {Verb: VerbRollback, Run: aRun, Plan: with(func(i *RollbackItem) {
|
||||
i.Source, i.Location = SourceRepository, ArchiveURL+"/repos/2026/10/04/../../../../x"
|
||||
})},
|
||||
"another source": {Verb: VerbRollback, Run: aRun, Plan: with(func(i *RollbackItem) { i.Source = "file" })},
|
||||
"a removal with a location": {Verb: VerbRollback, Run: aRun, Plan: with(func(i *RollbackItem) {
|
||||
i.Source, i.To = SourceRemoved, ""
|
||||
})},
|
||||
}
|
||||
for name, req := range cases {
|
||||
b := newRootBed(t)
|
||||
if a := b.h.Handle(req); a.Error == "" {
|
||||
t.Errorf("%s: carried out (%+v)", name, a)
|
||||
}
|
||||
if len(b.ran) != 0 {
|
||||
t.Errorf("%s: ran %q although refused", name, b.ran)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// serveOnce starts the helper on a socket of a test directory, allowing peer, and answers one request.
|
||||
func serveOnce(t *testing.T, peer int, req RootRequest) (RootAnswer, error) {
|
||||
t.Helper()
|
||||
b := newRootBed(t)
|
||||
b.h.Peer = peer
|
||||
sock := testSocket(t)
|
||||
l, err := net.Listen("unix", sock)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
ctx, cancel := context.WithCancel(context.Background())
|
||||
t.Cleanup(cancel)
|
||||
go func() { _ = b.h.Serve(ctx, l) }()
|
||||
return rootClient{Socket: sock}.ask(req)
|
||||
}
|
||||
|
||||
func TestTheRootHelperAnswersTheKeepersAccountOverItsSocket(t *testing.T) {
|
||||
a, err := serveOnce(t, os.Getuid(), RootRequest{Verb: VerbMaps})
|
||||
if err != nil || a.Error != "" || len(a.Mapped) != 1 {
|
||||
t.Fatalf("the keeper's own account was answered %+v, %v", a, err)
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheRootHelperRefusesAnyOtherAccount(t *testing.T) {
|
||||
a, err := serveOnce(t, os.Getuid()+1, RootRequest{Verb: VerbMaps})
|
||||
if err == nil && a.Error == "" {
|
||||
t.Fatalf("another account was answered: %+v", a)
|
||||
}
|
||||
if a.Mapped != nil {
|
||||
t.Fatalf("another account was handed the mappings: %+v", a)
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheClientCarriesTheRootActsAndSaysARefusal(t *testing.T) {
|
||||
b := newRootBed(t)
|
||||
b.h.Peer = os.Getuid()
|
||||
sock := testSocket(t)
|
||||
l, err := net.Listen("unix", sock)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
ctx, cancel := context.WithCancel(context.Background())
|
||||
defer cancel()
|
||||
go func() { _ = b.h.Serve(ctx, l) }()
|
||||
var acts RootActs = rootClient{Socket: sock}
|
||||
if err := acts.Transaction(RunUpgrade, aRun, nil); err != nil {
|
||||
t.Fatalf("transaction: %v", err)
|
||||
}
|
||||
if err := acts.Transaction(RunUpgrade, "bad", nil); err == nil || !strings.Contains(err.Error(), "run") {
|
||||
t.Fatalf("a bad run was not refused by name: %v", err)
|
||||
}
|
||||
if lines, err := acts.Journal(TransactionUnit(RunUpgrade, aRun), 5); err != nil || len(lines) != 1 {
|
||||
t.Fatalf("journal: %q, %v", lines, err)
|
||||
}
|
||||
if _, err := (rootClient{Socket: testSocket(t) + "-none"}).Maps(); err == nil || !strings.Contains(err.Error(), "root helper") {
|
||||
t.Fatalf("a helper that is not there is not said: %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheRootHelperIsRootWithNoBusAndNoSudoRule(t *testing.T) {
|
||||
m := manifest(t)
|
||||
r := m.resource(t, "root-helper")
|
||||
if r["type"] != "process" || r["name"] != "pacman-root" || r["artifact"] != "tools" ||
|
||||
strings.Join(strs(r["run"]), " ") != "./"+binaryName+" root" {
|
||||
t.Fatalf("the root helper: %v", r)
|
||||
}
|
||||
if _, ok := r["user"]; ok {
|
||||
t.Errorf("the root helper runs as root (ADR 0510 §1, §3), and names a user: %v", r["user"])
|
||||
}
|
||||
if _, ok := r["bus"]; ok {
|
||||
t.Errorf("the root helper has no bus account (ADR 0510 §1): %v", r["bus"])
|
||||
}
|
||||
env, _ := r["env"].(map[string]any)
|
||||
if env[EnvRootSocket] != RootSocketPath || env[EnvRootPeer] != "pacman-keeper" {
|
||||
t.Errorf("the root helper's environment names its socket and the keeper's account: %v", env)
|
||||
}
|
||||
k := m.resource(t, "keeper")
|
||||
kenv, _ := k["env"].(map[string]any)
|
||||
if kenv[EnvRootSocket] != RootSocketPath {
|
||||
t.Errorf("the keeper is told the helper's socket: %v", kenv)
|
||||
}
|
||||
for _, res := range m.Resources {
|
||||
if p, _ := res["path"].(string); strings.Contains(p, "sudoers") {
|
||||
t.Errorf("a file under the sudo configuration is placed: %s (ADR 0510 §1)", p)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheRollbackUnitInstallsThePlanThenRemovesWhatWasAskedInOrder(t *testing.T) {
|
||||
dir := t.TempDir()
|
||||
plan := []RollbackItem{
|
||||
{Name: "bash", To: "5.3.15-1", Source: SourceCache, Location: CacheDir + "/bash-5.3.15-1-x86_64.pkg.tar.zst"},
|
||||
{Name: "zlib", To: "1.3.1-2", Source: SourceRepository, Location: ArchiveURL + "/repos/2026/10/04/core/os/$arch"},
|
||||
{Name: "made-up-library", From: "1.0-1", Source: SourceRemoved},
|
||||
}
|
||||
raw, _ := json.Marshal(plan)
|
||||
file := filepath.Join(dir, aRun+".json")
|
||||
if err := os.WriteFile(file, raw, 0o600); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
var ran [][]string
|
||||
run := func(name string, args ...string) int { ran = append(ran, append([]string{name}, args...)); return 0 }
|
||||
probed := []string{}
|
||||
probe := func(url string) bool { probed = append(probed, url); return strings.HasSuffix(url, "-any.pkg.tar.zst") }
|
||||
if code := rollbackUnit(file, "x86_64", run, probe, os.Stderr); code != 0 {
|
||||
t.Fatalf("exit %d", code)
|
||||
}
|
||||
want := [][]string{
|
||||
{"pacman", "-U", "--noconfirm", "--", CacheDir + "/bash-5.3.15-1-x86_64.pkg.tar.zst", ArchiveURL + "/repos/2026/10/04/core/os/x86_64/zlib-1.3.1-2-any.pkg.tar.zst"},
|
||||
{"pacman", "-R", "--noconfirm", "--", "made-up-library"},
|
||||
}
|
||||
if !reflect.DeepEqual(ran, want) {
|
||||
t.Fatalf("ran %q, want %q", ran, want)
|
||||
}
|
||||
if len(probed) != 2 || !strings.HasSuffix(probed[0], "-x86_64.pkg.tar.zst") {
|
||||
t.Errorf("the machine's architecture is tried first, then any: %q", probed)
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheRollbackUnitChangesNothingWhenAPackageIsNotFound(t *testing.T) {
|
||||
file := filepath.Join(t.TempDir(), aRun+".json")
|
||||
raw, _ := json.Marshal([]RollbackItem{{Name: "zlib", To: "1.3.1-2", Source: SourceRepository, Location: ArchiveURL + "/repos/2026/10/04/core/os/$arch"}})
|
||||
_ = os.WriteFile(file, raw, 0o600)
|
||||
var ran [][]string
|
||||
run := func(name string, args ...string) int { ran = append(ran, append([]string{name}, args...)); return 0 }
|
||||
if code := rollbackUnit(file, "x86_64", run, func(string) bool { return false }, &strings.Builder{}); code == 0 || len(ran) != 0 {
|
||||
t.Fatalf("exit %d, ran %q: a package the archive does not hold fails the unit before any change", code, ran)
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,180 @@
|
||||
//go:build integration
|
||||
|
||||
package main
|
||||
|
||||
// The round over a real bus (novox/hq ADR 0508 §5, §6; acceptance 1841/AC15): the keepers of two made-up nodes,
|
||||
// each whole (its own bus account, its states, its watches) against a JetStream server of the test's own, with
|
||||
// fake machines and a fake root helper each. The first node of the order opens the week's round and runs; the
|
||||
// second starts only once the first's after-upgrade check passed; a failed check stops the round there.
|
||||
// Compiled by the merge gate's `go vet -tags integration`; run by the merge check, never by hand.
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"strings"
|
||||
"sync"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
"git.novox.be/novox/mesh-sdk/go/process"
|
||||
"github.com/nats-io/nats.go"
|
||||
"github.com/nats-io/nats.go/jetstream"
|
||||
)
|
||||
|
||||
// countingRoot is a node's root helper: it counts the transactions it was asked to start.
|
||||
type countingRoot struct {
|
||||
mu sync.Mutex
|
||||
started []string
|
||||
}
|
||||
|
||||
func (r *countingRoot) Transaction(kind, run string, ignore []string) error {
|
||||
r.mu.Lock()
|
||||
defer r.mu.Unlock()
|
||||
r.started = append(r.started, kind+" "+run)
|
||||
return nil
|
||||
}
|
||||
func (r *countingRoot) Rollback(string, []RollbackItem) error { return nil }
|
||||
func (r *countingRoot) Journal(string, int) ([]string, error) { return nil, nil }
|
||||
func (r *countingRoot) Maps() ([]Mapped, error) { return []Mapped{}, nil }
|
||||
func (r *countingRoot) count() int {
|
||||
r.mu.Lock()
|
||||
defer r.mu.Unlock()
|
||||
return len(r.started)
|
||||
}
|
||||
|
||||
// roundNode is one made-up node's keeper on the test's bus.
|
||||
type roundNode struct {
|
||||
root *countingRoot
|
||||
signals chan Signal
|
||||
}
|
||||
|
||||
// startRoundNode starts node's keeper; its machine's clock begins at the Saturday's opening, its transaction
|
||||
// unit is gone (ended well) whenever asked, and its after-upgrade check passes unless passes says otherwise.
|
||||
func startRoundNode(t *testing.T, ctx context.Context, url, node string, order []string, passes bool) *roundNode {
|
||||
t.Helper()
|
||||
db := &pacmanDB{sl: syncDBs, checkup: Ran{Status: 2}}
|
||||
db.set("glibc 2.42-1\nzsh 5.9-1\n", "zsh\n")
|
||||
base := db.runner()
|
||||
run := func(c context.Context, name string, args ...string) Ran {
|
||||
line := strings.TrimSpace(name + " " + strings.Join(args, " "))
|
||||
switch {
|
||||
case strings.HasPrefix(line, "systemctl list-units --state=failed"), line == "dkms status":
|
||||
return Ran{}
|
||||
case strings.HasPrefix(line, "systemctl show "+unitPrefix):
|
||||
return Ran{Stdout: "LoadState=not-found\n"}
|
||||
}
|
||||
return base(c, name, args...)
|
||||
}
|
||||
m := machine(run, 1000)
|
||||
began := time.Now()
|
||||
m.Now = func() time.Time { return t0.Add(time.Since(began)) }
|
||||
m.ReadFile = func(path string) ([]byte, error) {
|
||||
if path == OSRelease {
|
||||
return []byte("7.2.8-arch1-2\n"), nil
|
||||
}
|
||||
return nil, errNoFile
|
||||
}
|
||||
n := &roundNode{root: &countingRoot{}, signals: make(chan Signal)}
|
||||
k := &Keeper{Config: KeeperConfig{UpgradeOrder: order, RecordDays: 180, RoundStartMinutes: 120}, Machine: m, Clock: realClock{},
|
||||
Watcher: fakeWatcher{n.signals}, Logf: func(f string, a ...any) { t.Logf(node+": "+f, a...) }, MakeHooks: keeperHooks,
|
||||
Root: n.root, Check: func(Run, *Record) Check { return Check{Passed: passes} }}
|
||||
account := process.Account{URL: url, Node: node, Module: "pacman", User: node + ".pacman.keeper", Password: "pw"}
|
||||
go func() { _ = runKeeper(ctx, account, k) }()
|
||||
return n
|
||||
}
|
||||
|
||||
func roundBus(t *testing.T) (string, map[string]jetstream.KeyValue) {
|
||||
t.Helper()
|
||||
url := newTestBus(t)
|
||||
nc, err := nats.Connect(url)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
t.Cleanup(nc.Close)
|
||||
js, err := jetstream.New(nc)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
ctx := context.Background()
|
||||
buckets := map[string]jetstream.KeyValue{}
|
||||
for _, name := range States {
|
||||
kv, err := js.CreateKeyValue(ctx, jetstream.KeyValueConfig{Bucket: "pacman_" + name, History: 8})
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
buckets[name] = kv
|
||||
}
|
||||
if _, err := js.CreateStream(ctx, jetstream.StreamConfig{Name: "pacman-events", Subjects: []string{"mesh.mod.pacman.event.>"}}); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
return url, buckets
|
||||
}
|
||||
|
||||
func eventually(t *testing.T, what string, within time.Duration, ok func() bool) {
|
||||
t.Helper()
|
||||
for deadline := time.Now().Add(within); time.Now().Before(deadline); time.Sleep(50 * time.Millisecond) {
|
||||
if ok() {
|
||||
return
|
||||
}
|
||||
}
|
||||
t.Fatalf("never: %s", what)
|
||||
}
|
||||
|
||||
func roundNodeOnBus(kv jetstream.KeyValue, round, node string) RoundNode {
|
||||
var rn RoundNode
|
||||
if e, err := kv.Get(context.Background(), RoundNodeKey(round, node)); err == nil {
|
||||
_ = json.Unmarshal(e.Value(), &rn)
|
||||
}
|
||||
return rn
|
||||
}
|
||||
|
||||
func TestTheRoundGoesNodeByNodeOnARealBus(t *testing.T) {
|
||||
url, buckets := roundBus(t)
|
||||
ctx, cancel := context.WithCancel(context.Background())
|
||||
defer cancel()
|
||||
order := []string{"node-a", "node-b"}
|
||||
a := startRoundNode(t, ctx, url, "node-a", order, true)
|
||||
b := startRoundNode(t, ctx, url, "node-b", order, true)
|
||||
round := RoundID(t0)
|
||||
|
||||
eventually(t, "node-a opened the round and started its run", 20*time.Second, func() bool {
|
||||
return a.root.count() == 1 && roundNodeOnBus(buckets[StateRounds], round, "node-a").Run != ""
|
||||
})
|
||||
time.Sleep(time.Second)
|
||||
if b.root.count() != 0 {
|
||||
t.Fatalf("node-b started before node-a passed: %v", b.root.started)
|
||||
}
|
||||
// node-a's transaction ends: its lock is released, its run is finished and checked.
|
||||
a.signals <- LockReleased
|
||||
eventually(t, "node-b started once node-a passed", 20*time.Second, func() bool {
|
||||
return b.root.count() == 1 && roundNodeOnBus(buckets[StateRounds], round, "node-b").Run != ""
|
||||
})
|
||||
if a.root.count() != 1 {
|
||||
t.Errorf("node-a ran again: %v", a.root.started)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAFailedCheckStopsTheRoundOnARealBus(t *testing.T) {
|
||||
url, buckets := roundBus(t)
|
||||
ctx, cancel := context.WithCancel(context.Background())
|
||||
defer cancel()
|
||||
order := []string{"node-a", "node-b"}
|
||||
a := startRoundNode(t, ctx, url, "node-a", order, false)
|
||||
b := startRoundNode(t, ctx, url, "node-b", order, true)
|
||||
round := RoundID(t0)
|
||||
eventually(t, "node-a started its run", 20*time.Second, func() bool { return a.root.count() == 1 })
|
||||
a.signals <- LockReleased
|
||||
runA := ""
|
||||
eventually(t, "node-a's run checked", 20*time.Second, func() bool {
|
||||
runA = roundNodeOnBus(buckets[StateRounds], round, "node-a").Run
|
||||
if runA == "" {
|
||||
return false
|
||||
}
|
||||
e, err := buckets[StateRuns].Get(context.Background(), NodeKey("node-a", runA))
|
||||
return err == nil && strings.Contains(string(e.Value()), `"checked-after"`)
|
||||
})
|
||||
time.Sleep(2 * time.Second)
|
||||
if b.root.count() != 0 || roundNodeOnBus(buckets[StateRounds], round, "node-b").Run != "" {
|
||||
t.Fatalf("node-b started after node-a's check failed: %v", b.root.started)
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,565 @@
|
||||
package main
|
||||
|
||||
// The week's round (novox/hq ADR 0508 §5, §6): the nodes upgrade one at a time in the order of the mesh-level key
|
||||
// `upgrade-order`, each after the node before it passed its after-upgrade check. On Saturday at 04:00 local time
|
||||
// the first node of the order opens the week's round in `rounds` (a key created once, so one round a week); every
|
||||
// node's keeper watches `rounds` and `runs` and starts its own run when its turn came. A failed run or check
|
||||
// stops the round at that node. Only the operator passes a node: `pacman_round` with pass and a reason asks the
|
||||
// operator through the operator's channel (ADR 0259), and the keeper of the node waiting on it lets the round go
|
||||
// on only on the operator's Approve, read from the channel's own record. With no order, no round opens.
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"fmt"
|
||||
"os"
|
||||
"sort"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
"git.novox.be/novox/mesh-sdk/go/asks"
|
||||
)
|
||||
|
||||
// Module is this module's name: the asker of its asks on the operator's channel.
|
||||
const Module = "pacman"
|
||||
|
||||
// MessengerAsksBucket is the operator channel's records bucket, where the keeper reads the answer to a pass.
|
||||
const MessengerAsksBucket = "messenger_asks"
|
||||
|
||||
// The operator's answers to a pass, besides PassApproved and PassDeclined: an ask nobody answered in time.
|
||||
const PassExpired = "expired"
|
||||
|
||||
// passLasts is how long a pass waits for the operator's answer.
|
||||
const passLasts = 12 * time.Hour
|
||||
|
||||
// refusedRetry is how long after a run of the round was refused (it started nothing) the node tries again.
|
||||
const refusedRetry = time.Hour
|
||||
|
||||
// passCheckEvery is how often a keeper reads the channel's record of a pass that waits for its answer.
|
||||
const passCheckEvery = time.Minute
|
||||
|
||||
// defaultRoundStartMinutes is round-start-minutes' default in the manifest.
|
||||
const defaultRoundStartMinutes = 120
|
||||
|
||||
// RoundOpening is the Saturday at 04:00 of the ISO week that holds at, in at's own time zone.
|
||||
func RoundOpening(at time.Time) time.Time {
|
||||
day := int(at.Weekday()+6) % 7 // Monday 0 … Sunday 6
|
||||
monday := time.Date(at.Year(), at.Month(), at.Day()-day, 0, 0, 0, 0, at.Location())
|
||||
return monday.AddDate(0, 0, 5).Add(4 * time.Hour)
|
||||
}
|
||||
|
||||
// PassOperatorAsk is the ask on the operator's channel for a pass: composed from the pass alone, so the tool that
|
||||
// publishes it and the keeper that reads its answer compose the same ask, with the same digest.
|
||||
func PassOperatorAsk(round, node, next, reason, id string, at time.Time) asks.Ask {
|
||||
headline := "Upgrade round: pass " + node
|
||||
if len([]rune(headline)) > asks.HeadlineLength {
|
||||
headline = "Upgrade round: pass a node"
|
||||
}
|
||||
binds, _ := asks.ActDigest(passAct(round, node, reason, id))
|
||||
return asks.Ask{
|
||||
ID: id,
|
||||
Headline: headline,
|
||||
Explanation: fmt.Sprintf("The week's upgrade round %s stands at %s. Passing it lets %s and the nodes after it upgrade; "+
|
||||
"%s is not upgraded in this round.\nThe reason given: %s", round, node, next, node, reason),
|
||||
Options: []asks.Option{
|
||||
{ID: "pass", Label: "Pass it", Does: "the round goes on past " + node, Level: asks.Approve, Binds: binds},
|
||||
{ID: "keep", Label: "Keep it waiting", Does: "the round stays at " + node, Level: asks.Acknowledge},
|
||||
},
|
||||
Who: asks.Operator,
|
||||
Expires: at.UTC().Add(passLasts),
|
||||
OnExpiry: "the round stays at " + node,
|
||||
}
|
||||
}
|
||||
|
||||
func passAct(round, node, reason, id string) asks.Act {
|
||||
return asks.Act{"verb": Module + ".pass", "round": round, "node": node, "reason": reason, "ask": id}
|
||||
}
|
||||
|
||||
// roundView is one round as the bus holds it: the round, each node's place, and each node's run in it.
|
||||
type roundView struct {
|
||||
Round Round
|
||||
Nodes map[string]RoundNode
|
||||
revs map[string]uint64
|
||||
Runs map[string]Run
|
||||
}
|
||||
|
||||
// latestRound is the newest round's id, or none.
|
||||
func latestRound(rounds Bucket) (string, error) {
|
||||
keys, err := rounds.Keys()
|
||||
if err != nil {
|
||||
return "", fmt.Errorf("listing rounds: %w", err)
|
||||
}
|
||||
latest := ""
|
||||
for _, k := range keys {
|
||||
if !strings.Contains(k, ".") && k > latest {
|
||||
latest = k
|
||||
}
|
||||
}
|
||||
return latest, nil
|
||||
}
|
||||
|
||||
func readRound(rounds, runs Bucket, id string) (*roundView, error) {
|
||||
e, err := rounds.Get(id)
|
||||
if err != nil || e == nil {
|
||||
return nil, err
|
||||
}
|
||||
v := &roundView{Nodes: map[string]RoundNode{}, revs: map[string]uint64{}, Runs: map[string]Run{}}
|
||||
if err := json.Unmarshal(e.Value, &v.Round); err != nil {
|
||||
return nil, fmt.Errorf("round %s: %w", id, err)
|
||||
}
|
||||
for _, node := range v.Round.Order {
|
||||
e, err := rounds.Get(RoundNodeKey(id, node))
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
if e == nil {
|
||||
continue
|
||||
}
|
||||
var rn RoundNode
|
||||
if err := json.Unmarshal(e.Value, &rn); err != nil {
|
||||
return nil, fmt.Errorf("round %s, node %s: %w", id, node, err)
|
||||
}
|
||||
v.Nodes[node], v.revs[node] = rn, e.Revision
|
||||
if rn.Run == "" {
|
||||
continue
|
||||
}
|
||||
re, err := runs.Get(NodeKey(node, rn.Run))
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
if re != nil {
|
||||
var r Run
|
||||
if err := json.Unmarshal(re.Value, &r); err != nil {
|
||||
return nil, fmt.Errorf("run %s of %s: %w", rn.Run, node, err)
|
||||
}
|
||||
v.Runs[node] = r
|
||||
}
|
||||
}
|
||||
return v, nil
|
||||
}
|
||||
|
||||
// checkedAt is when a run's after-upgrade check passed, or zero.
|
||||
func checkedAt(r Run) time.Time {
|
||||
for _, s := range r.Steps {
|
||||
if s.Step == StepCheckedAfter && s.Check != nil && s.Check.Passed {
|
||||
return s.At
|
||||
}
|
||||
}
|
||||
return time.Time{}
|
||||
}
|
||||
|
||||
// passedAt is when a node let the round go on: its check passed, or the operator approved its pass; zero when
|
||||
// it has not.
|
||||
func (v *roundView) passedAt(node string) time.Time {
|
||||
if rn := v.Nodes[node]; rn.Pass != nil && rn.Pass.Answer == PassApproved && rn.Pass.Answered != nil {
|
||||
return *rn.Pass.Answered
|
||||
}
|
||||
if r, ok := v.Runs[node]; ok {
|
||||
return checkedAt(r)
|
||||
}
|
||||
return time.Time{}
|
||||
}
|
||||
|
||||
// turnSince is when a node's turn came: the round's opening for the first, the passing of the node before it
|
||||
// for the rest; zero when it has not come.
|
||||
func (v *roundView) turnSince(node string) time.Time {
|
||||
for i, n := range v.Round.Order {
|
||||
if n != node {
|
||||
continue
|
||||
}
|
||||
if i == 0 {
|
||||
return v.Round.Opened
|
||||
}
|
||||
return v.passedAt(v.Round.Order[i-1])
|
||||
}
|
||||
return time.Time{}
|
||||
}
|
||||
|
||||
// next is the node after node in the order, or none.
|
||||
func (v *roundView) next(node string) string {
|
||||
for i, n := range v.Round.Order {
|
||||
if n == node && i+1 < len(v.Round.Order) {
|
||||
return v.Round.Order[i+1]
|
||||
}
|
||||
}
|
||||
return ""
|
||||
}
|
||||
|
||||
// putRoundNode writes a node's place in a round from the revision read, so a pass and a run written by two
|
||||
// keepers never overwrite each other: a lost race is read again and changed again.
|
||||
func putRoundNode(rounds Bucket, round, node string, change func(*RoundNode)) error {
|
||||
key := RoundNodeKey(round, node)
|
||||
for attempt := 0; attempt < 5; attempt++ {
|
||||
e, err := rounds.Get(key)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
rn := RoundNode{Round: round, Node: node}
|
||||
if e != nil {
|
||||
if err := json.Unmarshal(e.Value, &rn); err != nil {
|
||||
return fmt.Errorf("%s: %w", key, err)
|
||||
}
|
||||
change(&rn)
|
||||
_, err = rounds.Update(key, rn, e.Revision)
|
||||
} else {
|
||||
change(&rn)
|
||||
_, err = rounds.Create(key, rn)
|
||||
}
|
||||
if !errors.Is(err, stdio.ErrChanged) {
|
||||
return err
|
||||
}
|
||||
}
|
||||
return fmt.Errorf("%s was written by another keeper five times in a row", key)
|
||||
}
|
||||
|
||||
// RoundStep does what the node's keeper has to do for the rounds now, and answers when it should be called
|
||||
// again (zero: only when `rounds` or `runs` change). The keeper calls it at its start, at the time it answered,
|
||||
// and whenever `rounds` or `runs` change. The first node of the order opens the week's round once it is
|
||||
// Saturday 04:00; a node whose turn came starts its run; a node waiting on a pass reads the operator's answer.
|
||||
func (u *Upgrades) RoundStep() (time.Time, error) {
|
||||
now := u.Machine.Now()
|
||||
order := u.Config.UpgradeOrder
|
||||
var next time.Time
|
||||
soonest := func(t time.Time) {
|
||||
if !t.IsZero() && (next.IsZero() || t.Before(next)) {
|
||||
next = t
|
||||
}
|
||||
}
|
||||
if len(order) == 0 {
|
||||
return time.Time{}, nil // no order: no round (the health says so)
|
||||
}
|
||||
var errs []error
|
||||
if order[0] == u.Node {
|
||||
opening := RoundOpening(now)
|
||||
if now.Before(opening) {
|
||||
soonest(opening)
|
||||
} else {
|
||||
soonest(opening.AddDate(0, 0, 7))
|
||||
id := RoundID(now)
|
||||
round := Round{Round: id, Opened: now.UTC(), OpenedBy: u.Node, Order: append([]string(nil), order...)}
|
||||
if _, err := u.Rounds.Create(id, round); err != nil && !errors.Is(err, stdio.ErrChanged) {
|
||||
errs = append(errs, fmt.Errorf("opening round %s: %w", id, err))
|
||||
}
|
||||
}
|
||||
}
|
||||
id, err := latestRound(u.Rounds)
|
||||
if err != nil || id == "" {
|
||||
return next, errors.Join(append(errs, err)...)
|
||||
}
|
||||
v, err := readRound(u.Rounds, u.Runs, id)
|
||||
if err != nil || v == nil {
|
||||
return next, errors.Join(append(errs, err)...)
|
||||
}
|
||||
// A pass of the node before this one that waits for the operator's answer.
|
||||
for i, node := range v.Round.Order {
|
||||
if i+1 >= len(v.Round.Order) || v.Round.Order[i+1] != u.Node {
|
||||
continue
|
||||
}
|
||||
if rn := v.Nodes[node]; rn.Pass != nil && rn.Pass.Answer == "" {
|
||||
answered, err := u.readPass(v, node)
|
||||
if err != nil {
|
||||
errs = append(errs, err)
|
||||
}
|
||||
if !answered {
|
||||
soonest(now.Add(passCheckEvery))
|
||||
} else if v, err = readRound(u.Rounds, u.Runs, id); err != nil {
|
||||
return next, errors.Join(append(errs, err)...)
|
||||
}
|
||||
}
|
||||
}
|
||||
t, err := u.takeTurn(v, now)
|
||||
soonest(t)
|
||||
return next, errors.Join(append(errs, err)...)
|
||||
}
|
||||
|
||||
// takeTurn starts this node's run in the round when its turn came and it has none, and again an hour after a
|
||||
// run of it was refused (it started nothing); it answers when to look again.
|
||||
func (u *Upgrades) takeTurn(v *roundView, now time.Time) (time.Time, error) {
|
||||
if !contains(v.Round.Order, u.Node) || v.turnSince(u.Node).IsZero() || !v.passedAt(u.Node).IsZero() {
|
||||
return time.Time{}, nil
|
||||
}
|
||||
if rn := v.Nodes[u.Node]; rn.Pass != nil && rn.Pass.Answer == "" {
|
||||
return time.Time{}, nil // the operator is asked to pass this node: it waits for the answer
|
||||
}
|
||||
own, err := u.ownRuns()
|
||||
if err != nil {
|
||||
return time.Time{}, err
|
||||
}
|
||||
var latest *Run
|
||||
for i := range own {
|
||||
if own[i].Round == v.Round.Round && (latest == nil || own[i].Run.Run > latest.Run) {
|
||||
latest = &own[i].Run
|
||||
}
|
||||
}
|
||||
if latest != nil {
|
||||
if v.Nodes[u.Node].Run != latest.Run {
|
||||
// a keeper that went away between starting its run and naming it in the round
|
||||
if err := putRoundNode(u.Rounds, v.Round.Round, u.Node, func(rn *RoundNode) { rn.Run = latest.Run }); err != nil {
|
||||
return time.Time{}, err
|
||||
}
|
||||
}
|
||||
if lastStep(*latest) != StepRefused {
|
||||
return time.Time{}, nil // running, or ended: a failure stops the round here
|
||||
}
|
||||
retry := latest.Steps[len(latest.Steps)-1].At.Add(refusedRetry)
|
||||
if now.Before(retry) {
|
||||
return retry, nil
|
||||
}
|
||||
}
|
||||
r, err := u.StartUpgrade(FromRound, v.Round.Round, "")
|
||||
if len(r.Steps) == 0 {
|
||||
return time.Time{}, err
|
||||
}
|
||||
if perr := putRoundNode(u.Rounds, v.Round.Round, u.Node, func(rn *RoundNode) { rn.Run = r.Run }); perr != nil {
|
||||
err = errors.Join(err, perr)
|
||||
}
|
||||
if lastStep(r) == StepRefused {
|
||||
return r.Steps[len(r.Steps)-1].At.Add(refusedRetry), err
|
||||
}
|
||||
return time.Time{}, err
|
||||
}
|
||||
|
||||
// readPass reads the operator's answer to a pass of node from the channel's record, and writes it into the
|
||||
// round once there is one. It answers whether the pass is answered.
|
||||
func (u *Upgrades) readPass(v *roundView, node string) (bool, error) {
|
||||
rn := v.Nodes[node]
|
||||
p := rn.Pass
|
||||
if u.AskRecord == nil {
|
||||
return false, errors.New("no reader of the operator channel's records: a pass cannot be answered")
|
||||
}
|
||||
e, err := u.AskRecord(p.Ask)
|
||||
if err != nil {
|
||||
return false, fmt.Errorf("reading the operator's answer to the pass of %s: %w", node, err)
|
||||
}
|
||||
if e == nil {
|
||||
return false, nil
|
||||
}
|
||||
var rec struct {
|
||||
State string `json:"state"`
|
||||
Warrant *asks.Warrant `json:"warrant"`
|
||||
}
|
||||
if err := json.Unmarshal(e.Value, &rec); err != nil {
|
||||
return false, fmt.Errorf("the channel's record of the pass of %s: %w", node, err)
|
||||
}
|
||||
if rec.State == "" || rec.State == "open" || rec.Warrant == nil {
|
||||
return false, nil
|
||||
}
|
||||
w := *rec.Warrant
|
||||
answer := PassDeclined
|
||||
switch {
|
||||
case w.Outcome == asks.OutcomeExpired:
|
||||
answer = PassExpired
|
||||
case w.Outcome == asks.OutcomeChosen:
|
||||
ask := PassOperatorAsk(v.Round.Round, node, v.next(node), p.Reason, p.Ask, p.Asked)
|
||||
option, err := w.For(Module, ask)
|
||||
if err == nil {
|
||||
err = option.Performs(passAct(v.Round.Round, node, p.Reason, p.Ask))
|
||||
}
|
||||
if err != nil {
|
||||
return false, fmt.Errorf("the answer to the pass of %s is not the operator's to this pass: %w", node, err)
|
||||
}
|
||||
if option.ID == "pass" {
|
||||
answer = PassApproved
|
||||
}
|
||||
}
|
||||
at := u.Machine.Now().UTC()
|
||||
return true, putRoundNode(u.Rounds, v.Round.Round, node, func(rn *RoundNode) {
|
||||
if rn.Pass != nil && rn.Pass.Ask == p.Ask && rn.Pass.Answer == "" {
|
||||
rn.Pass.Answer, rn.Pass.Answered = answer, &at
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
// takePass is a pass ask taken by the keeper of the node waiting on the passed node: the pass is written into
|
||||
// the round, waiting for the operator's answer (the tool asked the operator).
|
||||
func (u *Upgrades) takePass(a Ask) error {
|
||||
v, err := readRound(u.Rounds, u.Runs, a.Round)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
switch {
|
||||
case v == nil:
|
||||
return fmt.Errorf("ask %s passes %s in round %s, which is not open", a.ID, a.Pass, a.Round)
|
||||
case v.next(a.Pass) != u.Node:
|
||||
return fmt.Errorf("ask %s passes %s, on which %s does not wait", a.ID, a.Pass, u.Node)
|
||||
}
|
||||
return putRoundNode(u.Rounds, a.Round, a.Pass, func(rn *RoundNode) {
|
||||
if rn.Pass == nil || rn.Pass.Ask != a.ID { // taken again after a restart: the answer already read stays
|
||||
rn.Pass = &Pass{Reason: a.Reason, Ask: a.ID, Asked: a.At.UTC()}
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
// NodeInRound is where one node stands in a round, as pacman_round and the health say it.
|
||||
type NodeInRound struct {
|
||||
Node string `json:"node"`
|
||||
State string `json:"state"`
|
||||
Run string `json:"run,omitempty"`
|
||||
Since *time.Time `json:"since,omitempty"`
|
||||
WaitsFor string `json:"waits_for,omitempty"`
|
||||
Pass *Pass `json:"pass,omitempty"`
|
||||
}
|
||||
|
||||
// The states of a node in a round.
|
||||
const (
|
||||
InRoundWaiting = "waiting" // the node before it has not passed
|
||||
InRoundTurn = "turn" // its turn came, its run has not started
|
||||
InRoundStalled = "stalled" // its turn came longer than round-start-minutes ago, and it has not started
|
||||
InRoundRunning = "running" // its run goes on
|
||||
InRoundPassed = "passed" // its check passed, or the operator passed it
|
||||
InRoundFailed = "failed" // its run failed or its check did not pass: the round stops here
|
||||
InRoundRefused = "refused" // its run did not start (it is tried again)
|
||||
InRoundOff = "off-order" // not in the round's order
|
||||
)
|
||||
|
||||
// RoundState is where every node of the order stands in a round at now.
|
||||
func RoundState(v *roundView, now time.Time, startMinutes int) []NodeInRound {
|
||||
if startMinutes < 1 {
|
||||
startMinutes = defaultRoundStartMinutes
|
||||
}
|
||||
out := []NodeInRound{}
|
||||
for i, node := range v.Round.Order {
|
||||
s := NodeInRound{Node: node, Run: v.Nodes[node].Run, Pass: v.Nodes[node].Pass}
|
||||
at := func(t time.Time) *time.Time {
|
||||
if t.IsZero() {
|
||||
return nil
|
||||
}
|
||||
t = t.UTC()
|
||||
return &t
|
||||
}
|
||||
r, hasRun := v.Runs[node]
|
||||
switch {
|
||||
case !v.passedAt(node).IsZero():
|
||||
s.State, s.Since = InRoundPassed, at(v.passedAt(node))
|
||||
case hasRun && lastStep(r) == StepRefused:
|
||||
s.State, s.Since = InRoundRefused, at(r.Steps[len(r.Steps)-1].At)
|
||||
case hasRun && lastStep(r) == StepCheckedAfter:
|
||||
s.State, s.Since = InRoundFailed, at(r.Steps[len(r.Steps)-1].At)
|
||||
case hasRun:
|
||||
s.State = InRoundRunning
|
||||
if len(r.Steps) > 0 {
|
||||
s.Since = at(r.Steps[0].At)
|
||||
}
|
||||
case !v.turnSince(node).IsZero():
|
||||
since := v.turnSince(node)
|
||||
s.State, s.Since = InRoundTurn, at(since)
|
||||
if now.Sub(since) > time.Duration(startMinutes)*time.Minute {
|
||||
s.State = InRoundStalled
|
||||
}
|
||||
default:
|
||||
s.State = InRoundWaiting
|
||||
for j := i - 1; j >= 0; j-- {
|
||||
if v.passedAt(v.Round.Order[j]).IsZero() {
|
||||
s.WaitsFor = v.Round.Order[j]
|
||||
}
|
||||
}
|
||||
if s.WaitsFor != "" {
|
||||
// since the node it waits for stopped or its turn came
|
||||
prev := out[indexOf(v.Round.Order, s.WaitsFor)]
|
||||
s.Since = prev.Since
|
||||
}
|
||||
}
|
||||
out = append(out, s)
|
||||
}
|
||||
return out
|
||||
}
|
||||
|
||||
func indexOf(list []string, s string) int {
|
||||
for i, x := range list {
|
||||
if x == s {
|
||||
return i
|
||||
}
|
||||
}
|
||||
return -1
|
||||
}
|
||||
|
||||
// seatPublish publishes on a seat for the tools bundle; a test replaces it.
|
||||
var seatPublish = stdio.SeatPublish
|
||||
|
||||
// ListRounds answers pacman_round's read: the newest rounds, or one, each with every node's state.
|
||||
func ListRounds(rounds, runs Bucket, round string, now time.Time, startMinutes, limit int) (map[string]any, error) {
|
||||
keys, err := rounds.Keys()
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
var ids []string
|
||||
for _, k := range keys {
|
||||
if !strings.Contains(k, ".") && (round == "" || k == round) {
|
||||
ids = append(ids, k)
|
||||
}
|
||||
}
|
||||
if round != "" && len(ids) == 0 {
|
||||
return nil, fmt.Errorf("no round %s", round)
|
||||
}
|
||||
sort.Sort(sort.Reverse(sort.StringSlice(ids)))
|
||||
if len(ids) > limit {
|
||||
ids = ids[:limit]
|
||||
}
|
||||
listed := []map[string]any{}
|
||||
for _, id := range ids {
|
||||
v, err := readRound(rounds, runs, id)
|
||||
if err != nil || v == nil {
|
||||
return nil, err
|
||||
}
|
||||
listed = append(listed, map[string]any{"round": id, "opened": v.Round.Opened, "opened_by": v.Round.OpenedBy,
|
||||
"order": v.Round.Order, "nodes": RoundState(v, now, startMinutes)})
|
||||
}
|
||||
return map[string]any{"rounds": listed}, nil
|
||||
}
|
||||
|
||||
// AskPass asks the operator to let a round go on past a node: the ask on the operator's channel, and the ask in
|
||||
// `asks` for the keeper of the node waiting on it, which reads the answer. A reason is required.
|
||||
func AskPass(rounds, runs, asksState Bucket, round, node, reason string, now time.Time) (map[string]any, error) {
|
||||
if strings.TrimSpace(reason) == "" {
|
||||
return nil, errors.New("a pass needs a reason: the operator is asked to let the round go on past a node, and is told why")
|
||||
}
|
||||
if round == "" {
|
||||
id, err := latestRound(rounds)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
round = id
|
||||
}
|
||||
v, err := readRound(rounds, runs, round)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
if v == nil {
|
||||
return nil, fmt.Errorf("no round %s is open", round)
|
||||
}
|
||||
if !contains(v.Round.Order, node) {
|
||||
return nil, fmt.Errorf("%s is not in round %s's order (%s)", node, round, strings.Join(v.Round.Order, ", "))
|
||||
}
|
||||
if !v.passedAt(node).IsZero() {
|
||||
return nil, fmt.Errorf("%s has passed already in round %s", node, round)
|
||||
}
|
||||
next := v.next(node)
|
||||
if next == "" {
|
||||
return nil, fmt.Errorf("%s is the last of the order: no node waits on it", node)
|
||||
}
|
||||
a := Ask{ID: AskID(now), Node: next, Act: ActPass, At: now.UTC(), Round: round, Pass: node, Reason: reason}
|
||||
opAsk := PassOperatorAsk(round, node, next, reason, a.ID, a.At)
|
||||
if err := opAsk.Check(now); err != nil {
|
||||
return nil, err
|
||||
}
|
||||
if err := seatPublish(asks.AskSubject(Module), opAsk, opAsk.ID); err != nil {
|
||||
return nil, fmt.Errorf("the operator could not be asked: %w; nothing was passed", err)
|
||||
}
|
||||
if _, err := asksState.Create(NodeKey(next, a.ID), a); err != nil {
|
||||
return nil, fmt.Errorf("the operator was asked, and the ask for %s's keeper could not be written: %w; an Approve passes nothing", next, err)
|
||||
}
|
||||
return map[string]any{"asked": opAsk.ID, "round": round, "pass": node, "waiting": next,
|
||||
"next": "the round goes on past " + node + " only when the operator approves on their channel; pacman_round says where it stands"}, nil
|
||||
}
|
||||
|
||||
// roundStartMinutes is round-start-minutes as the tools bundle reads it: from the keeper's config file, which
|
||||
// the manifest names to the bundle too (EnvConfig); its default when the file cannot be read.
|
||||
var roundStartMinutes = func() int {
|
||||
raw, err := os.ReadFile(os.Getenv(EnvConfig))
|
||||
var c KeeperConfig
|
||||
if err != nil || json.Unmarshal(raw, &c) != nil || c.RoundStartMinutes < 1 {
|
||||
return defaultRoundStartMinutes
|
||||
}
|
||||
return c.RoundStartMinutes
|
||||
}
|
||||
@@ -0,0 +1,305 @@
|
||||
package main
|
||||
|
||||
// The week's round (novox/hq ADR 0508 §5, §6; acceptance 1841/AC6 and 1841/AC7), for three made-up nodes over
|
||||
// states in memory.
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
"git.novox.be/novox/mesh-sdk/go/asks"
|
||||
)
|
||||
|
||||
// roundBed is three keepers over one bed's states, each with the order node-a, node-b, node-c.
|
||||
type roundBed struct {
|
||||
*keeperBed
|
||||
nodes map[string]*Upgrades
|
||||
}
|
||||
|
||||
func newRoundBed(t *testing.T, order ...string) *roundBed {
|
||||
k := newKeeperBed(t)
|
||||
k.now = time.Date(2026, 10, 24, 3, 0, 0, 0, time.UTC) // a Saturday
|
||||
b := &roundBed{keeperBed: k, nodes: map[string]*Upgrades{}}
|
||||
var updates UpdatesValue
|
||||
k.updates.value("node-a", &updates)
|
||||
for _, n := range []string{"node-a", "node-b", "node-c"} {
|
||||
u := k.keeper(k.u.Machine)
|
||||
u.Node, u.Config.UpgradeOrder, u.Config.RoundStartMinutes = n, order, 120
|
||||
if _, err := k.installed.Put(n, nodeAPackages); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if _, err := k.updates.Put(n, updates); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
b.nodes[n] = u
|
||||
}
|
||||
return b
|
||||
}
|
||||
|
||||
func (b *roundBed) step(t *testing.T, node string) time.Time {
|
||||
t.Helper()
|
||||
next, err := b.nodes[node].RoundStep()
|
||||
if err != nil {
|
||||
t.Fatalf("%s: %v", node, err)
|
||||
}
|
||||
return next
|
||||
}
|
||||
|
||||
// end ends node's running transaction as unitState says and follows it.
|
||||
func (b *roundBed) end(t *testing.T, node, unitState string) {
|
||||
t.Helper()
|
||||
b.unitState = unitState
|
||||
if err := b.nodes[node].Follow(); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
b.now = b.now.Add(5 * time.Minute)
|
||||
}
|
||||
|
||||
func (b *roundBed) runsOf(node string) []Run {
|
||||
keys, _ := b.runs.Keys()
|
||||
var out []Run
|
||||
for _, k := range keys {
|
||||
if strings.HasPrefix(k, node+".") {
|
||||
var r Run
|
||||
b.runs.value(k, &r)
|
||||
out = append(out, r)
|
||||
}
|
||||
}
|
||||
return out
|
||||
}
|
||||
|
||||
func (b *roundBed) state(t *testing.T) map[string]NodeInRound {
|
||||
t.Helper()
|
||||
v, err := readRound(b.rounds, b.runs, "2026-W43")
|
||||
if err != nil || v == nil {
|
||||
t.Fatalf("round: %v", err)
|
||||
}
|
||||
out := map[string]NodeInRound{}
|
||||
for _, s := range RoundState(v, b.now, 120) {
|
||||
out[s.Node] = s
|
||||
}
|
||||
return out
|
||||
}
|
||||
|
||||
func TestTheRoundOpensSaturdayAtFourAndThePassedCheckGivesTheNextItsTurn(t *testing.T) {
|
||||
if o := RoundOpening(time.Date(2026, 10, 22, 15, 0, 0, 0, time.UTC)); !o.Equal(time.Date(2026, 10, 24, 4, 0, 0, 0, time.UTC)) {
|
||||
t.Fatalf("the opening of the week of Thursday 22 October: %s", o)
|
||||
}
|
||||
b := newRoundBed(t, "node-a", "node-b", "node-c")
|
||||
if next := b.step(t, "node-a"); !next.Equal(time.Date(2026, 10, 24, 4, 0, 0, 0, time.UTC)) {
|
||||
t.Errorf("before 04:00 the first node waits for 04:00: %s", next)
|
||||
}
|
||||
if keys, _ := b.rounds.Keys(); len(keys) != 0 {
|
||||
t.Fatalf("a round before 04:00: %v", keys)
|
||||
}
|
||||
b.step(t, "node-b")
|
||||
if keys, _ := b.rounds.Keys(); len(keys) != 0 {
|
||||
t.Fatalf("only the first of the order opens the round: %v", keys)
|
||||
}
|
||||
|
||||
b.now = time.Date(2026, 10, 24, 4, 0, 0, 0, time.UTC)
|
||||
if next := b.step(t, "node-a"); !next.Equal(time.Date(2026, 10, 31, 4, 0, 0, 0, time.UTC)) {
|
||||
t.Errorf("after opening, next Saturday: %s", next)
|
||||
}
|
||||
var round Round
|
||||
if !b.rounds.value("2026-W43", &round) || round.OpenedBy != "node-a" || strings.Join(round.Order, " ") != "node-a node-b node-c" {
|
||||
t.Fatalf("the round: %+v", round)
|
||||
}
|
||||
if r := b.runsOf("node-a"); len(r) != 1 || r[0].Round != "2026-W43" || r[0].From != FromRound || lastStep(r[0]) != StepStarted {
|
||||
t.Fatalf("the first node starts its run in the round: %+v", r)
|
||||
}
|
||||
b.now = b.now.Add(time.Minute)
|
||||
b.step(t, "node-a")
|
||||
if w := b.rounds.writesTo("2026-W43"); len(w) != 2 || len(b.runsOf("node-a")) != 1 {
|
||||
t.Errorf("the round is created once and the run started once: %v", w)
|
||||
}
|
||||
b.step(t, "node-b")
|
||||
if len(b.runsOf("node-b")) != 0 || b.state(t)["node-b"].State != InRoundWaiting {
|
||||
t.Fatalf("node-b waits while node-a runs: %+v", b.state(t)["node-b"])
|
||||
}
|
||||
|
||||
b.end(t, "node-a", "LoadState=not-found\n")
|
||||
if s := b.state(t)["node-a"]; s.State != InRoundPassed {
|
||||
t.Fatalf("node-a: %+v", s)
|
||||
}
|
||||
b.step(t, "node-c")
|
||||
if len(b.runsOf("node-c")) != 0 {
|
||||
t.Fatal("node-c started before node-b")
|
||||
}
|
||||
b.step(t, "node-b")
|
||||
if r := b.runsOf("node-b"); len(r) != 1 || r[0].Round != "2026-W43" {
|
||||
t.Fatalf("node-b's turn after node-a's check passed: %+v", r)
|
||||
}
|
||||
|
||||
// node-b's transaction fails: the round stops at node-b, and node-c waits.
|
||||
b.end(t, "node-b", "LoadState=loaded\nActiveState=failed\nResult=exit-code\nExecMainStatus=1\n")
|
||||
b.step(t, "node-c")
|
||||
st := b.state(t)
|
||||
if len(b.runsOf("node-c")) != 0 || st["node-b"].State != InRoundFailed || st["node-c"].State != InRoundWaiting || st["node-c"].WaitsFor != "node-b" {
|
||||
t.Errorf("a failed check stops the round: %+v %+v", st["node-b"], st["node-c"])
|
||||
}
|
||||
}
|
||||
|
||||
func TestNoOrderNoRound(t *testing.T) {
|
||||
b := newRoundBed(t)
|
||||
b.now = time.Date(2026, 10, 24, 5, 0, 0, 0, time.UTC)
|
||||
if next := b.step(t, "node-a"); !next.IsZero() {
|
||||
t.Errorf("no wake without an order: %s", next)
|
||||
}
|
||||
if keys, _ := b.rounds.Keys(); len(keys) != 0 || len(b.runsOf("node-a")) != 0 {
|
||||
t.Errorf("a round without an order: %v", keys)
|
||||
}
|
||||
}
|
||||
|
||||
func TestANodeOffAtItsTurnRunsAtItsNextStartAndAStalledTurnIsSaid(t *testing.T) {
|
||||
b := newRoundBed(t, "node-a", "node-b", "node-c")
|
||||
b.now = time.Date(2026, 10, 24, 4, 0, 0, 0, time.UTC)
|
||||
b.step(t, "node-a")
|
||||
b.end(t, "node-a", "LoadState=not-found\n")
|
||||
passed := b.now.Add(-5 * time.Minute)
|
||||
b.now = passed.Add(121 * time.Minute) // node-b is off
|
||||
st := b.state(t)
|
||||
if st["node-b"].State != InRoundStalled || st["node-b"].Since == nil || !st["node-b"].Since.Equal(passed) {
|
||||
t.Errorf("node-b's turn came at %s and it has not started: %+v", passed, st["node-b"])
|
||||
}
|
||||
if st["node-c"].WaitsFor != "node-b" || st["node-c"].State != InRoundWaiting {
|
||||
t.Errorf("node-c names the node it waits for: %+v", st["node-c"])
|
||||
}
|
||||
b.step(t, "node-b") // node-b starts again
|
||||
if len(b.runsOf("node-b")) != 1 {
|
||||
t.Error("a node off at its turn runs at its next start")
|
||||
}
|
||||
}
|
||||
|
||||
// warrant is the operator channel's record of an ask answered with option.
|
||||
func warrant(a asks.Ask, option string, level asks.Level, at time.Time) json.RawMessage {
|
||||
w := asks.Warrant{Ask: a.ID, Asker: Module, Outcome: asks.OutcomeChosen, Option: option, Level: level,
|
||||
By: &asks.Person{Who: asks.Operator, Kind: "made-up-channel", Identity: "1", Verified: "user id"}, At: at, AskDigest: a.Digest()}
|
||||
raw, _ := json.Marshal(map[string]any{"state": "chosen", "warrant": w})
|
||||
return raw
|
||||
}
|
||||
|
||||
func TestOnlyTheOperatorsApprovePassesANodeAndAPassNeedsAReason(t *testing.T) {
|
||||
b := newRoundBed(t, "node-a", "node-b", "node-c")
|
||||
b.now = time.Date(2026, 10, 24, 4, 0, 0, 0, time.UTC)
|
||||
b.step(t, "node-a")
|
||||
b.end(t, "node-a", "LoadState=not-found\n")
|
||||
b.step(t, "node-b")
|
||||
b.end(t, "node-b", "LoadState=loaded\nActiveState=failed\nResult=exit-code\nExecMainStatus=1\n")
|
||||
|
||||
var published []asks.Ask
|
||||
var subjects []string
|
||||
saved := seatPublish
|
||||
seatPublish = func(subject string, body any, id string) error {
|
||||
subjects, published = append(subjects, subject), append(published, body.(asks.Ask))
|
||||
return nil
|
||||
}
|
||||
defer func() { seatPublish = saved }()
|
||||
if _, err := AskPass(b.rounds, b.runs, b.asks, "", "node-b", " ", b.now); err == nil || len(published) != 0 {
|
||||
t.Fatalf("a pass without a reason: %v", err)
|
||||
}
|
||||
out, err := AskPass(b.rounds, b.runs, b.asks, "", "node-b", "its disk is being replaced", b.now)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if len(published) != 1 || subjects[0] != "mesh.seat.operator-channel.accept.ask.pacman" || out["waiting"] != "node-c" {
|
||||
t.Fatalf("the operator is asked on their channel: %v %v", subjects, out)
|
||||
}
|
||||
opAsk := published[0]
|
||||
if err := opAsk.Check(b.now); err != nil || !strings.Contains(opAsk.Explanation, "its disk is being replaced") {
|
||||
t.Errorf("the ask: %v %+v", err, opAsk)
|
||||
}
|
||||
keys, _ := b.asks.Keys()
|
||||
if len(keys) != 1 || keys[0] != NodeKey("node-c", opAsk.ID) {
|
||||
t.Fatalf("the ask for node-c's keeper: %v", keys)
|
||||
}
|
||||
var a Ask
|
||||
b.asks.value(keys[0], &a)
|
||||
raw, _ := json.Marshal(a)
|
||||
if err := b.nodes["node-c"].TakeAsk(stdio.StateChange{State: StateAsks, Key: keys[0], Op: "put", Value: raw}); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if st := b.state(t)["node-b"]; st.Pass == nil || st.Pass.Reason != "its disk is being replaced" || st.Pass.Answer != "" {
|
||||
t.Fatalf("the pass waits in the round: %+v", st.Pass)
|
||||
}
|
||||
|
||||
record := map[string]json.RawMessage{}
|
||||
b.nodes["node-c"].AskRecord = func(id string) (*stdio.StateEntry, error) {
|
||||
if v, ok := record[id]; ok {
|
||||
return &stdio.StateEntry{Key: Module + "." + id, Value: v}, nil
|
||||
}
|
||||
return nil, nil
|
||||
}
|
||||
if next := b.step(t, "node-c"); len(b.runsOf("node-c")) != 0 || !next.Equal(b.now.Add(passCheckEvery)) {
|
||||
t.Fatalf("no answer yet: node-c waits and looks again in a minute (%s)", next)
|
||||
}
|
||||
// An answer to another ask, or to this one as it was not shown, passes nothing.
|
||||
other := opAsk
|
||||
other.Explanation = "something else"
|
||||
record[opAsk.ID] = warrant(other, "pass", asks.Approve, b.now)
|
||||
if _, err := b.nodes["node-c"].RoundStep(); err == nil || len(b.runsOf("node-c")) != 0 {
|
||||
t.Fatalf("a warrant for another ask: %v", err)
|
||||
}
|
||||
record[opAsk.ID] = warrant(opAsk, "pass", asks.Approve, b.now)
|
||||
b.step(t, "node-c")
|
||||
if st := b.state(t)["node-b"]; st.State != InRoundPassed || st.Pass.Answer != PassApproved {
|
||||
t.Errorf("the operator's Approve passes node-b: %+v", st)
|
||||
}
|
||||
if r := b.runsOf("node-c"); len(r) != 1 || r[0].Round != "2026-W43" {
|
||||
t.Errorf("node-c's turn after the pass: %+v", r)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAPassTheOperatorKeepsWaitingPassesNothing(t *testing.T) {
|
||||
b := newRoundBed(t, "node-a", "node-b")
|
||||
b.now = time.Date(2026, 10, 24, 4, 0, 0, 0, time.UTC)
|
||||
b.step(t, "node-a") // node-a runs and never ends
|
||||
saved := seatPublish
|
||||
var opAsk asks.Ask
|
||||
seatPublish = func(_ string, body any, _ string) error { opAsk = body.(asks.Ask); return nil }
|
||||
defer func() { seatPublish = saved }()
|
||||
if _, err := AskPass(b.rounds, b.runs, b.asks, "2026-W43", "node-a", "it hangs", b.now); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
keys, _ := b.asks.Keys()
|
||||
var a Ask
|
||||
b.asks.value(keys[0], &a)
|
||||
raw, _ := json.Marshal(a)
|
||||
if err := b.nodes["node-b"].TakeAsk(stdio.StateChange{State: StateAsks, Key: keys[0], Op: "put", Value: raw}); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
b.nodes["node-b"].AskRecord = func(string) (*stdio.StateEntry, error) {
|
||||
return &stdio.StateEntry{Value: warrant(opAsk, "keep", asks.Acknowledge, b.now)}, nil
|
||||
}
|
||||
b.step(t, "node-b")
|
||||
if st := b.state(t); st["node-a"].Pass.Answer != PassDeclined || len(b.runsOf("node-b")) != 0 {
|
||||
t.Errorf("kept waiting: %+v", st["node-a"])
|
||||
}
|
||||
if _, err := AskPass(b.rounds, b.runs, b.asks, "2026-W43", "node-b", "x", b.now); err == nil {
|
||||
t.Error("the last of the order has no node waiting on it")
|
||||
}
|
||||
if _, err := AskPass(b.rounds, b.runs, b.asks, "2026-W43", "node-z", "x", b.now); err == nil {
|
||||
t.Error("a node not in the order")
|
||||
}
|
||||
}
|
||||
|
||||
func TestPacmanRoundReadsTheRounds(t *testing.T) {
|
||||
b := newRoundBed(t, "node-a", "node-b")
|
||||
b.now = time.Date(2026, 10, 24, 4, 0, 0, 0, time.UTC)
|
||||
b.step(t, "node-a")
|
||||
out, err := ListRounds(b.rounds, b.runs, "", b.now, 120, 4)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
listed := out["rounds"].([]map[string]any)
|
||||
nodes := listed[0]["nodes"].([]NodeInRound)
|
||||
if len(listed) != 1 || listed[0]["round"] != "2026-W43" || nodes[0].State != InRoundRunning || nodes[1].WaitsFor != "node-a" {
|
||||
t.Errorf("%+v", listed)
|
||||
}
|
||||
if _, err := ListRounds(b.rounds, b.runs, "2026-W40", b.now, 120, 4); err == nil {
|
||||
t.Error("a round that is not there")
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,217 @@
|
||||
package main
|
||||
|
||||
// Runs on the bus and the log read once (novox/hq ADR 0508 §1; acceptance 1841/AC1 for the writing of steps,
|
||||
// 1841/AC2 for the log and pacman_upgrades).
|
||||
|
||||
import (
|
||||
"errors"
|
||||
"os"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
)
|
||||
|
||||
func capturedLog(t *testing.T) string {
|
||||
t.Helper()
|
||||
raw, err := os.ReadFile("testdata/pacman-log-three-full-upgrades/pacman.log")
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
return string(raw)
|
||||
}
|
||||
|
||||
// The captured log holds three full upgrades among other commands: one that ended before any transaction, one
|
||||
// whose transaction completed, and a later one that upgraded one package.
|
||||
func TestEveryFullUpgradeOfTheLogIsARunReadFromIt(t *testing.T) {
|
||||
runs := PastUpgrades("node-a", capturedLog(t))
|
||||
if len(runs) != 3 {
|
||||
t.Fatalf("%d runs, want the three full upgrades and none of the other commands: %+v", len(runs), runs)
|
||||
}
|
||||
for _, r := range runs {
|
||||
if r.Node != "node-a" || r.Kind != RunUpgrade || r.From != FromLog || len(r.Steps) != 2 || r.Steps[0].Step != StepStarted {
|
||||
t.Errorf("a run read from the log: %+v", r)
|
||||
}
|
||||
}
|
||||
first, second, third := runs[0], runs[1], runs[2]
|
||||
if first.Run != "20261004T130648Z" || first.Steps[1].Step != StepFinished || first.Steps[1].Why != noTransaction ||
|
||||
len(first.Steps[1].Changes.Changed) != 0 {
|
||||
t.Errorf("an upgrade with no transaction says so: %+v", first)
|
||||
}
|
||||
end := second.Steps[1]
|
||||
if second.Run != "20261004T130724Z" || end.Step != StepFinished || !end.At.Equal(time.Date(2026, 10, 4, 13, 22, 50, 0, time.UTC)) {
|
||||
t.Errorf("the second upgrade finished when its transaction completed: %+v", second)
|
||||
}
|
||||
var names []string
|
||||
for _, c := range end.Changes.Changed {
|
||||
names = append(names, c.Name)
|
||||
}
|
||||
if strings.Join(names, " ") != "bash glibc iana-etc readline tzdata wireshark-qt xfsprogs xpad xterm zathura-pdf-mupdf" {
|
||||
t.Errorf("what it changed, sorted: %v", names)
|
||||
}
|
||||
if c := end.Changes.Changed[0]; c.From != "5.3.15-1" || c.To != "5.3.20-1" {
|
||||
t.Errorf("bash: %+v", c)
|
||||
}
|
||||
if len(end.Changes.Removed) != 0 {
|
||||
t.Errorf("the removal run between them is no part of an upgrade: %+v", end.Changes.Removed)
|
||||
}
|
||||
if c := third.Steps[1].Changes.Changed; third.Run != "20261004T132817Z" || len(c) != 1 || c[0].Name != "dunst" {
|
||||
t.Errorf("the third upgrade: %+v", third)
|
||||
}
|
||||
}
|
||||
|
||||
func TestATransactionWithoutAnEndOrThatFailedIsAFailedRun(t *testing.T) {
|
||||
log := "[2026-10-10T04:00:00+0000] [PACMAN] Running 'pacman -Syu --noconfirm'\n" +
|
||||
"[2026-10-10T04:00:01+0000] [PACMAN] starting full system upgrade\n" +
|
||||
"[2026-10-10T04:00:05+0000] [ALPM] transaction started\n" +
|
||||
"[2026-10-10T04:00:06+0000] [ALPM] upgraded linux (7.2.8.arch1-2 -> 7.2.9.arch1-1)\n" +
|
||||
"[2026-10-10T04:00:09+0000] [ALPM] transaction failed\n" +
|
||||
"[2026-10-11T04:00:00+0000] [PACMAN] Running 'pacman -Syu --noconfirm'\n" +
|
||||
"[2026-10-11T04:00:01+0000] [PACMAN] starting full system upgrade\n" +
|
||||
"[2026-10-11T04:00:05+0000] [ALPM] transaction started\n"
|
||||
runs := PastUpgrades("node-b", log)
|
||||
if len(runs) != 2 {
|
||||
t.Fatalf("%+v", runs)
|
||||
}
|
||||
if s := runs[0].Steps[1]; s.Step != StepFailed || s.Error != "transaction failed" || len(s.Changes.Changed) != 1 {
|
||||
t.Errorf("a failed transaction: %+v", s)
|
||||
}
|
||||
if s := runs[1].Steps[1]; s.Step != StepFailed || !strings.Contains(s.Error, "no end") {
|
||||
t.Errorf("a transaction the log holds no end of: %+v", s)
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheLogIsReadIntoRunsOnceAndASecondStartReadsNothing(t *testing.T) {
|
||||
m := machine(nil, 1000)
|
||||
reads := 0
|
||||
log := capturedLog(t)
|
||||
m.ReadFile = func(path string) ([]byte, error) {
|
||||
if path != PacmanLog {
|
||||
t.Fatalf("read %s", path)
|
||||
}
|
||||
reads++
|
||||
return []byte(log), nil
|
||||
}
|
||||
runs := newBucket()
|
||||
other := Run{Node: "node-b", Run: "20261001T040000Z", Kind: RunUpgrade, From: FromRound}
|
||||
if _, err := runs.Create(NodeKey("node-b", other.Run), other); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
u := &Upgrades{Node: "node-a", Machine: m, Runs: runs}
|
||||
n, err := u.ImportLog()
|
||||
if err != nil || n != 3 || reads != 1 {
|
||||
t.Fatalf("first start: %d runs, %d reads, %v", n, reads, err)
|
||||
}
|
||||
var r Run
|
||||
if !runs.value("node-a.20261004T132817Z", &r) || r.From != FromLog {
|
||||
t.Errorf("the run is kept at <node>.<run>, marked as read from the log: %+v", r)
|
||||
}
|
||||
before := len(runs.writes)
|
||||
n, err = u.ImportLog()
|
||||
if err != nil || n != 0 || reads != 1 || len(runs.writes) != before {
|
||||
t.Errorf("a second start: %d runs, %d reads, %d writes, %v", n, reads, len(runs.writes)-before, err)
|
||||
}
|
||||
}
|
||||
|
||||
// A run's key is created at its first step and then only updated from the revision its writer holds: a step
|
||||
// is written once, and a run another writer changed meanwhile is refused, never overwritten.
|
||||
func TestAStepIsWrittenOnceAndAnotherWritersChangeIsNeverOverwritten(t *testing.T) {
|
||||
runs := newBucket()
|
||||
u := &Upgrades{Node: "node-a", Runs: runs}
|
||||
r := &openRun{Run: Run{Node: "node-a", Run: "20261024T040000Z", Kind: RunUpgrade, From: FromAsk}}
|
||||
at := time.Date(2026, 10, 24, 4, 0, 0, 0, time.UTC)
|
||||
for _, s := range []string{StepRecorded, StepCheckedBefore} {
|
||||
if err := u.step(r, Step{Step: s, At: at}); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
}
|
||||
if w := strings.Join(runs.writes, ", "); w != "create node-a.20261024T040000Z, update node-a.20261024T040000Z" {
|
||||
t.Errorf("writes: %s", w)
|
||||
}
|
||||
var kept Run
|
||||
runs.value("node-a.20261024T040000Z", &kept)
|
||||
if len(kept.Steps) != 2 || kept.Steps[1].Step != StepCheckedBefore {
|
||||
t.Errorf("kept: %+v", kept)
|
||||
}
|
||||
if _, err := runs.Put("node-a.20261024T040000Z", kept); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if err := u.step(r, Step{Step: StepStarted, At: at}); !errors.Is(err, stdio.ErrChanged) || len(r.Steps) != 2 {
|
||||
t.Errorf("a step over another writer's change: %v, %d steps held", err, len(r.Steps))
|
||||
}
|
||||
}
|
||||
|
||||
func TestPacmanUpgradesAnswersAllOneNodeDaysAndOneRun(t *testing.T) {
|
||||
runs := newBucket()
|
||||
put := func(node, id string, steps ...Step) {
|
||||
r := Run{Node: node, Run: id, Kind: RunUpgrade, From: FromRound, Round: "2026-W43", Steps: steps}
|
||||
if _, err := runs.Create(NodeKey(node, id), r); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
}
|
||||
at := time.Date(2026, 10, 24, 4, 30, 0, 0, time.UTC)
|
||||
passed := &Check{Passed: true}
|
||||
put("node-a", "20261024T040000Z", Step{Step: StepRecorded, At: at}, Step{Step: StepFinished, At: at}, Step{Step: StepCheckedAfter, At: at, Check: passed})
|
||||
put("node-b", "20261024T050000Z", Step{Step: StepRecorded, At: at}, Step{Step: StepStarted, At: at})
|
||||
put("node-b", "20261017T050000Z", Step{Step: StepRefused, At: at, Why: "installed is not current"})
|
||||
all, err := ListUpgrades(runs, "", "", "", "", 50)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
listed := all["runs"].([]RunSummary)
|
||||
var order []string
|
||||
for _, s := range listed {
|
||||
order = append(order, s.Node+"/"+s.Run+"/"+s.Outcome+"/"+s.Checked)
|
||||
}
|
||||
if strings.Join(order, " ") != "node-b/20261024T050000Z/running/ node-a/20261024T040000Z/finished/passed node-b/20261017T050000Z/refused/" {
|
||||
t.Fatalf("all runs, newest first: %v", order)
|
||||
}
|
||||
if listed[2].Error != "installed is not current" {
|
||||
t.Errorf("a refused run says why: %+v", listed[2])
|
||||
}
|
||||
one, _ := ListUpgrades(runs, "node-b", "", "", "", 50)
|
||||
if one["count"] != 2 {
|
||||
t.Errorf("one node: %v", one)
|
||||
}
|
||||
days, _ := ListUpgrades(runs, "", "2026-10-18", "2026-10-24", "", 50)
|
||||
if days["count"] != 2 {
|
||||
t.Errorf("between two days, the last one included: %v", days)
|
||||
}
|
||||
limited, _ := ListUpgrades(runs, "", "", "", "", 1)
|
||||
if limited["count"] != 1 || limited["truncated"] != true {
|
||||
t.Errorf("limited: %v", limited)
|
||||
}
|
||||
whole, err := ListUpgrades(runs, "node-a", "", "", "20261024T040000Z", 50)
|
||||
if err != nil || len(whole["run"].(Run).Steps) != 3 {
|
||||
t.Errorf("one run whole: %v %v", whole, err)
|
||||
}
|
||||
if _, err := ListUpgrades(runs, "node-a", "", "", "20261024T050000Z", 50); err == nil {
|
||||
t.Error("a run the node does not have was answered")
|
||||
}
|
||||
if _, err := ListUpgrades(runs, "", "", "", "20261024T040000Z", 50); err == nil {
|
||||
t.Error("a run without its node was answered")
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheUpgradesToolReadsTheRunsState(t *testing.T) {
|
||||
runs := newBucket()
|
||||
saved := upgradeState
|
||||
upgradeState = func(name string) Bucket {
|
||||
if name != StateRuns {
|
||||
t.Fatalf("read %s", name)
|
||||
}
|
||||
return runs
|
||||
}
|
||||
defer func() { upgradeState = saved }()
|
||||
for _, tool := range tools(machine(nil, 1000)) {
|
||||
if tool.Name == "pacman_upgrades" {
|
||||
out, err := tool.Run(map[string]any{"node": "node-a"})
|
||||
if err != nil || out.(map[string]any)["count"] != 0 {
|
||||
t.Errorf("%v %v", out, err)
|
||||
}
|
||||
return
|
||||
}
|
||||
}
|
||||
t.Fatal("no pacman_upgrades")
|
||||
}
|
||||
@@ -0,0 +1,474 @@
|
||||
package main
|
||||
|
||||
// The keeper's bus shapes (novox/hq ADR 0507, ADR 0508): the value of every key of the module's states and the
|
||||
// body of every event its process emits, in one place, so the process that writes them and the verbs that read
|
||||
// them agree by type. Each is a struct of named fields only: no map and no free-form value, so nothing but what
|
||||
// is named here goes on the bus (ADR 0507 §1: package names, versions, unit and container names, file paths and
|
||||
// exit statuses, never a process's command line or environment). A time in a value is the time of the thing it
|
||||
// describes (a transaction, a step, an opening), never the time it was read: a key is written only when what it
|
||||
// describes changed (ADR 0289 §1), and no field changes on a read alone (ADR 0289 §4). When a key last changed
|
||||
// is the bus's to say (its revision), not a field's.
|
||||
//
|
||||
// A key is `<node>` or `<node>.<id>`: a node name and an id are each one key token ([A-Za-z0-9_-]), so the
|
||||
// dot only ever separates them.
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"time"
|
||||
)
|
||||
|
||||
// The module's states (manifest `state` and the keeper's `bus` `state`).
|
||||
const (
|
||||
// StateInstalled holds each node's installed packages: key `<node>` an InstalledValue, and key
|
||||
// `<node>.not-current` a NotCurrent while the node's packages are over the bus's cap (ADR 0507 §2).
|
||||
StateInstalled = "installed"
|
||||
// StateUpdates holds what a full upgrade would change: key `<node>` an UpdatesValue (ADR 0507 §4).
|
||||
StateUpdates = "updates"
|
||||
// StateRuns holds every upgrade and rollback: key `<node>.<run>` a Run, kept (ADR 0508 §1).
|
||||
StateRuns = "runs"
|
||||
// StateRecords holds the package record taken before a run: key `<node>.<run>` a Record, deleted after
|
||||
// the config value `record-days` (ADR 0508 §2).
|
||||
StateRecords = "records"
|
||||
// StateRounds holds the weekly rounds: key `<round>` a Round, created once by the first node of the order,
|
||||
// and key `<round>.<node>` that node's RoundNode (ADR 0508 §5, §6).
|
||||
StateRounds = "rounds"
|
||||
// StateAsks holds the acts asked of a node: key `<node>.<id>` an Ask, taken by that node's keeper alone
|
||||
// (ADR 0508 §8).
|
||||
StateAsks = "asks"
|
||||
)
|
||||
|
||||
// States are the module's states, in the manifest's order.
|
||||
var States = []string{StateInstalled, StateUpdates, StateRuns, StateRecords, StateRounds, StateAsks}
|
||||
|
||||
// The events the keeper emits (the keeper's `bus` `emits`, and the manifest's `emits`).
|
||||
const (
|
||||
EventPackageChanged = "package.changed" // PackageChanged (ADR 0507 §3)
|
||||
EventRecorded = "upgrade.recorded" // UpgradeRecorded (ADR 0508 §1, §2)
|
||||
EventKernelHeld = "upgrade.kernel-held" // KernelHeld (ADR 0508 §3)
|
||||
EventStarted = "upgrade.started" // UpgradeStarted
|
||||
EventFinished = "upgrade.finished" // UpgradeFinished
|
||||
EventFailed = "upgrade.failed" // UpgradeFailed
|
||||
EventChecked = "upgrade.checked" // UpgradeChecked (ADR 0508 §4)
|
||||
EventRollbackStarted = "rollback.started" // RollbackStarted (ADR 0508 §7)
|
||||
EventRollbackFinished = "rollback.finished" // RollbackFinished
|
||||
)
|
||||
|
||||
// Events are the events the keeper emits, each with the type of its body.
|
||||
var Events = []struct {
|
||||
Name string
|
||||
Body any
|
||||
}{
|
||||
{EventPackageChanged, PackageChanged{}},
|
||||
{EventRecorded, UpgradeRecorded{}},
|
||||
{EventKernelHeld, KernelHeld{}},
|
||||
{EventStarted, UpgradeStarted{}},
|
||||
{EventFinished, UpgradeFinished{}},
|
||||
{EventFailed, UpgradeFailed{}},
|
||||
{EventChecked, UpgradeChecked{}},
|
||||
{EventRollbackStarted, RollbackStarted{}},
|
||||
{EventRollbackFinished, RollbackFinished{}},
|
||||
}
|
||||
|
||||
// Key names.
|
||||
|
||||
// NotCurrentKey is the key in `installed` that marks a node's packages as not current.
|
||||
func NotCurrentKey(node string) string { return node + ".not-current" }
|
||||
|
||||
// NodeKey is a key `<node>.<id>`: a run in `runs` or `records`, an ask in `asks`.
|
||||
func NodeKey(node, id string) string { return node + "." + id }
|
||||
|
||||
// RoundNodeKey is a node's key in a round: `<round>.<node>` in `rounds`.
|
||||
func RoundNodeKey(round, node string) string { return round + "." + node }
|
||||
|
||||
// RunID names a run by the moment it began, in UTC: `20261024T040012Z`. A run read from the package manager's
|
||||
// log is named by the log's time of its start the same way.
|
||||
func RunID(began time.Time) string { return began.UTC().Format("20060102T150405Z") }
|
||||
|
||||
// RoundID names the round of the ISO week that holds the given moment, in its own time zone: `2026-W43`. The
|
||||
// round opens on Saturday 04:00 local time, so the local date decides the week.
|
||||
func RoundID(at time.Time) string {
|
||||
year, week := at.ISOWeek()
|
||||
return fmt.Sprintf("%04d-W%02d", year, week)
|
||||
}
|
||||
|
||||
// installed
|
||||
|
||||
// InstalledValue is `installed.<node>`: every installed package, sorted by name, and the time of the transaction
|
||||
// that left the database so (ADR 0507 §2). Written compactly: one short entry per package.
|
||||
type InstalledValue struct {
|
||||
At time.Time `json:"at"`
|
||||
Packages []InstalledPackage `json:"packages"`
|
||||
}
|
||||
|
||||
// InstalledPackage is one installed package: its version, whether it was installed explicitly (else as a
|
||||
// dependency), and its repository, or RepoForeign when no repository holds it (ADR 0507 §7).
|
||||
type InstalledPackage struct {
|
||||
Name string `json:"name"`
|
||||
Version string `json:"version"`
|
||||
Explicit bool `json:"explicit"`
|
||||
Repo string `json:"repo"`
|
||||
}
|
||||
|
||||
// RepoForeign marks a package from outside the repositories.
|
||||
const RepoForeign = "foreign"
|
||||
|
||||
// NotCurrent is `installed.<node>.not-current`: the node's packages came to more than the bus's cap on a value,
|
||||
// so `installed.<node>` keeps the last value written. Written once, when it first happens, and deleted at the
|
||||
// next successful write of `installed.<node>`.
|
||||
type NotCurrent struct {
|
||||
Bytes int `json:"bytes"`
|
||||
Since time.Time `json:"since"`
|
||||
}
|
||||
|
||||
// updates
|
||||
|
||||
// UpdatesValue is `updates.<node>`: every package a full upgrade would change, sorted by name, and whether the
|
||||
// node reached its mirrors (ADR 0507 §4). A failed read writes Reachable false and keeps the last list, never an
|
||||
// empty list in its place.
|
||||
type UpdatesValue struct {
|
||||
Reachable bool `json:"reachable"`
|
||||
Packages []VersionChange `json:"packages"`
|
||||
}
|
||||
|
||||
// VersionChange is one package going from one version to another.
|
||||
type VersionChange struct {
|
||||
Name string `json:"name"`
|
||||
From string `json:"from"`
|
||||
To string `json:"to"`
|
||||
}
|
||||
|
||||
// PackageVersion is one package at one version.
|
||||
type PackageVersion struct {
|
||||
Name string `json:"name"`
|
||||
Version string `json:"version"`
|
||||
}
|
||||
|
||||
// runs
|
||||
|
||||
// The kinds of run.
|
||||
const (
|
||||
RunUpgrade = "upgrade"
|
||||
RunRollback = "rollback"
|
||||
)
|
||||
|
||||
// What started a run.
|
||||
const (
|
||||
FromRound = "round" // its turn in the week's round
|
||||
FromAsk = "ask" // an ask: pacman_upgrade, or pacman_rollback with apply
|
||||
FromLog = "log" // read from the package manager's log at the keeper's first start (ADR 0508 §1)
|
||||
)
|
||||
|
||||
// The steps of a run, in the order they are written. A run that does not start ends at StepRefused; a
|
||||
// transaction ends at StepFinished or StepFailed, and the after-upgrade check follows either.
|
||||
const (
|
||||
StepRecorded = "recorded" // the record is written and read back (upgrade only)
|
||||
StepCheckedBefore = "checked-before" // the DKMS gate ran; Held names what it held
|
||||
StepRefused = "refused" // the run did not start; Why says why
|
||||
StepStarted = "started" // the transaction's unit started
|
||||
StepFinished = "finished" // the transaction exited 0
|
||||
StepFailed = "failed" // the transaction failed
|
||||
StepCheckedAfter = "checked-after" // the after-upgrade check
|
||||
)
|
||||
|
||||
// Run is `runs.<node>.<run>`: one upgrade or rollback of a node, written at each step and at no other time
|
||||
// (ADR 0508 §1). Round names the round it ran in; Ask the ask that started it (the keeper deletes the ask once
|
||||
// the run names it); Of, for a rollback, the run whose record it returns to.
|
||||
type Run struct {
|
||||
Node string `json:"node"`
|
||||
Run string `json:"run"`
|
||||
Kind string `json:"kind"`
|
||||
From string `json:"from"`
|
||||
Round string `json:"round,omitempty"`
|
||||
Ask string `json:"ask,omitempty"`
|
||||
Of string `json:"of,omitempty"`
|
||||
Steps []Step `json:"steps"`
|
||||
}
|
||||
|
||||
// Step is one step of a run with its time and outcome; only the fields of its kind are set.
|
||||
type Step struct {
|
||||
Step string `json:"step"`
|
||||
At time.Time `json:"at"`
|
||||
// StepRefused: why the run did not start.
|
||||
Why string `json:"why,omitempty"`
|
||||
// StepCheckedBefore: the kernels held from this run.
|
||||
Held []HeldKernel `json:"held,omitempty"`
|
||||
// StepStarted: the transaction's unit, the packages left out of it (held kernels and their headers), and
|
||||
// for a rollback its plan.
|
||||
Unit string `json:"unit,omitempty"`
|
||||
Ignored []string `json:"ignored,omitempty"`
|
||||
Plan []RollbackItem `json:"plan,omitempty"`
|
||||
// StepFinished and StepFailed: the exit status, the last error line, and what the transaction changed,
|
||||
// read from its lines in the package manager's log.
|
||||
Exit *int `json:"exit,omitempty"`
|
||||
Error string `json:"error,omitempty"`
|
||||
Changes *Changes `json:"changes,omitempty"`
|
||||
// StepCheckedAfter: the after-upgrade check.
|
||||
Check *Check `json:"check,omitempty"`
|
||||
}
|
||||
|
||||
// HeldKernel is a kernel package left out of a run because an installed DKMS module's own bound excludes its
|
||||
// new version (ADR 0508 §3): `Linux-Maximum` in the module's META, or `BUILD_EXCLUSIVE_KERNEL` in its dkms.conf.
|
||||
type HeldKernel struct {
|
||||
Package string `json:"package"`
|
||||
Version string `json:"version"`
|
||||
Module string `json:"module"`
|
||||
Bound string `json:"bound"`
|
||||
}
|
||||
|
||||
// Changes is what a transaction changed, each list sorted by name.
|
||||
type Changes struct {
|
||||
Added []PackageVersion `json:"added"`
|
||||
Removed []PackageVersion `json:"removed"`
|
||||
Changed []VersionChange `json:"changed"`
|
||||
}
|
||||
|
||||
// Check is the after-upgrade check (ADR 0508 §4). Passed when the transaction succeeded, no unit newly failed
|
||||
// and every DKMS module is built for every installed kernel; the rest is said and does not stop the round.
|
||||
type Check struct {
|
||||
Passed bool `json:"passed"`
|
||||
// The .pacnew and .pacsave files the transaction left, by path.
|
||||
Pacnew []string `json:"pacnew"`
|
||||
Pacsave []string `json:"pacsave"`
|
||||
// The distribution's news since the node's previous upgrade.
|
||||
News []NewsTitle `json:"news"`
|
||||
// Why the news could not be read, when it could not: no news is then not said to be none.
|
||||
NewsError string `json:"news_error,omitempty"`
|
||||
// The units and containers whose processes still map a file the upgrade replaced.
|
||||
Mapped []Mapped `json:"mapped"`
|
||||
// Why the mapped files could not be read, when they could not.
|
||||
MappedError string `json:"mapped_error,omitempty"`
|
||||
// The units failed now that were not failed in the record.
|
||||
NewlyFailed []string `json:"newly_failed"`
|
||||
// Whether a restart is needed: the running kernel is not the installed version of its package.
|
||||
RestartNeeded bool `json:"restart_needed"`
|
||||
RunningKernel string `json:"running_kernel"`
|
||||
InstalledKernel string `json:"installed_kernel"`
|
||||
// The DKMS modules not built for an installed kernel.
|
||||
Unbuilt []DKMSBuild `json:"unbuilt"`
|
||||
// Why the DKMS builds could not be read, when they could not; the check then does not pass.
|
||||
DKMSError string `json:"dkms_error,omitempty"`
|
||||
}
|
||||
|
||||
// NewsTitle is one item of the distribution's news, without its text.
|
||||
type NewsTitle struct {
|
||||
Title string `json:"title"`
|
||||
Link string `json:"link"`
|
||||
Published string `json:"published"`
|
||||
}
|
||||
|
||||
// Mapped is a process still mapping a replaced file, named by its unit or its container and its executable's
|
||||
// path; never its command line or environment.
|
||||
type Mapped struct {
|
||||
Unit string `json:"unit,omitempty"`
|
||||
Container string `json:"container,omitempty"`
|
||||
Executable string `json:"executable"`
|
||||
}
|
||||
|
||||
// DKMSBuild is one DKMS module at one version for one kernel.
|
||||
type DKMSBuild struct {
|
||||
Module string `json:"module"`
|
||||
Version string `json:"version"`
|
||||
Kernel string `json:"kernel"`
|
||||
}
|
||||
|
||||
// RollbackItem is one package a rollback brings back (ADR 0508 §7): from its version now to the record's, and
|
||||
// where that version comes from.
|
||||
type RollbackItem struct {
|
||||
Name string `json:"name"`
|
||||
From string `json:"from"`
|
||||
To string `json:"to"`
|
||||
Source string `json:"source"`
|
||||
Location string `json:"location"`
|
||||
}
|
||||
|
||||
// Where a rollback takes a package's version from, in the order it is looked for.
|
||||
const (
|
||||
SourceCache = "cache" // the node's package cache: a file path
|
||||
SourceArchive = "archive" // the Arch Linux Archive's copy of that exact version: a URL
|
||||
SourceRepository = "repository" // the archive's repository as it stood on the record's date: a URL
|
||||
)
|
||||
|
||||
// records
|
||||
|
||||
// Record is `records.<node>.<run>`: the node's packages taken before a run and read back (ADR 0508 §2): a copy of
|
||||
// `installed.<node>`, the running kernel (/proc/sys/kernel/osrelease), the units failed at that moment, and the
|
||||
// installed DKMS modules.
|
||||
type Record struct {
|
||||
Node string `json:"node"`
|
||||
Run string `json:"run"`
|
||||
At time.Time `json:"at"`
|
||||
Installed InstalledValue `json:"installed"`
|
||||
Kernel string `json:"kernel"`
|
||||
FailedUnits []string `json:"failed_units"`
|
||||
DKMS []DKMSModule `json:"dkms"`
|
||||
}
|
||||
|
||||
// DKMSModule is one installed DKMS module and the kernels it is built for.
|
||||
type DKMSModule struct {
|
||||
Module string `json:"module"`
|
||||
Version string `json:"version"`
|
||||
Kernels []string `json:"kernels"`
|
||||
}
|
||||
|
||||
// rounds
|
||||
|
||||
// Round is `rounds.<round>`: the week's round, created once (a create, so one round per week) by the first node
|
||||
// of the order when it opens it, with the order as it stood then (ADR 0508 §5).
|
||||
type Round struct {
|
||||
Round string `json:"round"`
|
||||
Opened time.Time `json:"opened"`
|
||||
OpenedBy string `json:"opened_by"`
|
||||
Order []string `json:"order"`
|
||||
}
|
||||
|
||||
// RoundNode is `rounds.<round>.<node>`: the node's place in the round. Run is the run it started in its turn,
|
||||
// written by the node itself; Pass, the operator's pass of a node that stopped the round, written by the keeper
|
||||
// of the node waiting on it (ADR 0508 §6). Both write through a compare-and-set.
|
||||
type RoundNode struct {
|
||||
Round string `json:"round"`
|
||||
Node string `json:"node"`
|
||||
Run string `json:"run,omitempty"`
|
||||
Pass *Pass `json:"pass,omitempty"`
|
||||
}
|
||||
|
||||
// Pass is a pass of a node asked of the operator through the operator's channel (ADR 0259): the reason, the
|
||||
// channel's ask, and its answer once given (Answer empty while it waits).
|
||||
type Pass struct {
|
||||
Reason string `json:"reason"`
|
||||
Ask string `json:"ask"`
|
||||
Asked time.Time `json:"asked"`
|
||||
Answer string `json:"answer,omitempty"`
|
||||
Answered *time.Time `json:"answered,omitempty"`
|
||||
}
|
||||
|
||||
// The operator's answers to a pass.
|
||||
const (
|
||||
PassApproved = "approved"
|
||||
PassDeclined = "declined"
|
||||
)
|
||||
|
||||
// asks
|
||||
|
||||
// The acts an ask asks.
|
||||
const (
|
||||
ActUpgrade = "upgrade" // pacman_upgrade
|
||||
ActRollback = "rollback" // pacman_rollback with apply
|
||||
ActPass = "pass" // pacman_round with pass
|
||||
)
|
||||
|
||||
// Ask is `asks.<node>.<id>`: an act asked of a node, taken by that node's keeper alone, which answers in `runs`
|
||||
// (an upgrade, a rollback) or in `rounds` (a pass) and then deletes the ask (ADR 0508 §8). A rollback names the
|
||||
// run whose record it returns to and whether the packages the upgrade added are removed; a pass names the
|
||||
// round, the node passed and the reason, and is asked of the node waiting on it.
|
||||
type Ask struct {
|
||||
ID string `json:"id"`
|
||||
Node string `json:"node"`
|
||||
Act string `json:"act"`
|
||||
At time.Time `json:"at"`
|
||||
Of string `json:"of,omitempty"`
|
||||
RemoveAdded bool `json:"remove_added,omitempty"`
|
||||
Round string `json:"round,omitempty"`
|
||||
Pass string `json:"pass,omitempty"`
|
||||
Reason string `json:"reason,omitempty"`
|
||||
}
|
||||
|
||||
// events
|
||||
|
||||
// PackageChanged is `package.changed`: a write of `installed.<node>` that changed it (ADR 0507 §3).
|
||||
type PackageChanged struct {
|
||||
Node string `json:"node"`
|
||||
At time.Time `json:"at"`
|
||||
Added []PackageVersion `json:"added"`
|
||||
Removed []PackageVersion `json:"removed"`
|
||||
Changed []VersionChange `json:"changed"`
|
||||
}
|
||||
|
||||
// UpgradeRecorded is `upgrade.recorded`: the run's record is written and read back.
|
||||
type UpgradeRecorded struct {
|
||||
Node string `json:"node"`
|
||||
Run string `json:"run"`
|
||||
Packages int `json:"packages"`
|
||||
Kernel string `json:"kernel"`
|
||||
}
|
||||
|
||||
// KernelHeld is `upgrade.kernel-held`: kernels left out of a run, each with the module and bound that hold it.
|
||||
type KernelHeld struct {
|
||||
Node string `json:"node"`
|
||||
Run string `json:"run"`
|
||||
Held []HeldKernel `json:"held"`
|
||||
}
|
||||
|
||||
// UpgradeStarted is `upgrade.started`: a run's transaction started.
|
||||
type UpgradeStarted struct {
|
||||
Node string `json:"node"`
|
||||
Run string `json:"run"`
|
||||
From string `json:"from"`
|
||||
Round string `json:"round,omitempty"`
|
||||
Unit string `json:"unit"`
|
||||
Ignored []string `json:"ignored"`
|
||||
}
|
||||
|
||||
// UpgradeFinished is `upgrade.finished`: a run's transaction exited 0, with what it changed.
|
||||
type UpgradeFinished struct {
|
||||
Node string `json:"node"`
|
||||
Run string `json:"run"`
|
||||
Changes Changes `json:"changes"`
|
||||
}
|
||||
|
||||
// UpgradeFailed is `upgrade.failed`: a run that did not start (Step StepRefused, with why in Error) or whose
|
||||
// transaction failed (Step StepFailed, with its exit status and last error line).
|
||||
type UpgradeFailed struct {
|
||||
Node string `json:"node"`
|
||||
Run string `json:"run"`
|
||||
Step string `json:"step"`
|
||||
Exit *int `json:"exit,omitempty"`
|
||||
Error string `json:"error"`
|
||||
}
|
||||
|
||||
// UpgradeChecked is `upgrade.checked`: the after-upgrade check of a run, an upgrade's or a rollback's; the
|
||||
// check whole is in the run.
|
||||
type UpgradeChecked struct {
|
||||
Node string `json:"node"`
|
||||
Run string `json:"run"`
|
||||
Kind string `json:"kind"`
|
||||
Passed bool `json:"passed"`
|
||||
}
|
||||
|
||||
// RollbackStarted is `rollback.started`: a rollback's transaction started, returning to the record of run Of.
|
||||
type RollbackStarted struct {
|
||||
Node string `json:"node"`
|
||||
Run string `json:"run"`
|
||||
Of string `json:"of"`
|
||||
Unit string `json:"unit"`
|
||||
}
|
||||
|
||||
// RollbackFinished is `rollback.finished`: a rollback's transaction ended, with its exit status and, when it
|
||||
// failed, its last error line.
|
||||
type RollbackFinished struct {
|
||||
Node string `json:"node"`
|
||||
Run string `json:"run"`
|
||||
Of string `json:"of"`
|
||||
Exit int `json:"exit"`
|
||||
Error string `json:"error,omitempty"`
|
||||
}
|
||||
|
||||
// The keeper's config file
|
||||
|
||||
// The keeper's environment, named by the manifest's process resource.
|
||||
const (
|
||||
EnvBroker = "PACMAN_KEEPER_BROKER" // the keeper's bus account file
|
||||
EnvConfig = "PACMAN_KEEPER_CONFIG" // the keeper's config file, a KeeperConfig
|
||||
)
|
||||
|
||||
// KeeperConfig is the keeper's config file: the module's config values the mesh merges into it (ADR 0348 §5).
|
||||
// The process reads it at its start; the mesh restarts the process when it changes (the manifest's
|
||||
// `restart-on`).
|
||||
type KeeperConfig struct {
|
||||
// UpgradeOrder is the nodes in the order they upgrade; the first opens each week's round. Empty: no round.
|
||||
UpgradeOrder []string `json:"upgrade-order"`
|
||||
// RecordDays is how long a record is kept, in days.
|
||||
RecordDays int `json:"record-days"`
|
||||
// RoundStartMinutes is how long a node whose turn came may wait before starting, in minutes, before the
|
||||
// round is said to stand still.
|
||||
RoundStartMinutes int `json:"round-start-minutes"`
|
||||
}
|
||||
+4
@@ -0,0 +1,4 @@
|
||||
openrazer-driver/3.12.4, 7.2.8-arch1-2, x86_64: installed
|
||||
vboxhost/7.2.20_OSE, 7.2.8-arch1-2, x86_64: installed
|
||||
xone/0.5.8, 7.2.8-arch1-2, x86_64: installed
|
||||
xpad/0.4: added
|
||||
@@ -0,0 +1,10 @@
|
||||
Meta: 1
|
||||
Name: zfs
|
||||
Branch: 1.0
|
||||
Version: 2.4.4
|
||||
Release: 1
|
||||
Release-Tags: relext
|
||||
License: CDDL
|
||||
Author: OpenZFS
|
||||
Linux-Maximum: 7.2
|
||||
Linux-Minimum: 4.18
|
||||
+56
@@ -0,0 +1,56 @@
|
||||
PACKAGE_NAME="zfs"
|
||||
PACKAGE_VERSION="2.4.4"
|
||||
PACKAGE_CONFIG="/etc/sysconfig/zfs"
|
||||
NO_WEAK_MODULES="yes"
|
||||
PRE_BUILD="configure
|
||||
--disable-dependency-tracking
|
||||
--prefix=/usr
|
||||
--with-config=kernel
|
||||
--with-linux="${kernel_source_dir}"
|
||||
--with-linux-obj="${kernel_source_dir}"
|
||||
$(
|
||||
[[ -n \"${ICP_ROOT}\" ]] && \
|
||||
{
|
||||
echo --with-qat=\"${ICP_ROOT}\"
|
||||
}
|
||||
)
|
||||
$(
|
||||
[[ -r ${PACKAGE_CONFIG} ]] \
|
||||
&& source ${PACKAGE_CONFIG} \
|
||||
&& shopt -q -s extglob \
|
||||
&& \
|
||||
{
|
||||
if [[ ${ZFS_DKMS_ENABLE_DEBUG,,} == @(y|yes) ]]
|
||||
then
|
||||
echo --enable-debug
|
||||
fi
|
||||
if [[ ${ZFS_DKMS_ENABLE_DEBUGINFO,,} == @(y|yes) ]]
|
||||
then
|
||||
echo --enable-debuginfo
|
||||
fi
|
||||
}
|
||||
)
|
||||
"
|
||||
POST_BUILD="scripts/dkms.postbuild
|
||||
-n ${PACKAGE_NAME}
|
||||
-v ${PACKAGE_VERSION}
|
||||
-a ${arch}
|
||||
-k ${kernelver}
|
||||
-t ${dkms_tree}
|
||||
"
|
||||
AUTOINSTALL="yes"
|
||||
MAKE[0]="make -C module/"
|
||||
STRIP[0]="$(
|
||||
[[ -r ${PACKAGE_CONFIG} ]] \
|
||||
&& source ${PACKAGE_CONFIG} \
|
||||
&& shopt -q -s extglob \
|
||||
&& [[ ${ZFS_DKMS_DISABLE_STRIP,,} == @(y|yes) ]] \
|
||||
&& echo -n no
|
||||
)"
|
||||
STRIP[1]="${STRIP[0]}"
|
||||
BUILT_MODULE_NAME[0]="zfs"
|
||||
BUILT_MODULE_LOCATION[0]="module/"
|
||||
DEST_MODULE_LOCATION[0]="/extra"
|
||||
BUILT_MODULE_NAME[1]="spl"
|
||||
BUILT_MODULE_LOCATION[1]="module/"
|
||||
DEST_MODULE_LOCATION[1]="/extra"
|
||||
+63
@@ -0,0 +1,63 @@
|
||||
[2026-10-04T14:52:51+0200] [ALPM] installed logrotate (3.22.0-1)
|
||||
[2026-10-04T14:52:51+0200] [ALPM] transaction completed
|
||||
[2026-10-04T14:52:51+0200] [ALPM] running '30-systemd-daemon-reload-system.hook'...
|
||||
[2026-10-04T14:54:22+0200] [ALPM] running '35-systemd-update.hook'...
|
||||
[2026-10-04T15:06:13+0200] [PACMAN] Running 'pacman -Syu --noconfirm'
|
||||
[2026-10-04T15:06:13+0200] [PACMAN] synchronizing package lists
|
||||
[2026-10-04T15:06:48+0200] [PACMAN] starting full system upgrade
|
||||
[2026-10-04T15:07:23+0200] [PACMAN] Running 'pacman -Rns --noconfirm lib32-libpcap'
|
||||
[2026-10-04T15:07:23+0200] [ALPM] transaction started
|
||||
[2026-10-04T15:07:24+0200] [ALPM] removed lib32-libpcap (1.10.6-1)
|
||||
[2026-10-04T15:07:24+0200] [ALPM] removed lib32-libnl (3.12.0-1)
|
||||
[2026-10-04T15:07:24+0200] [ALPM] transaction completed
|
||||
[2026-10-04T15:07:24+0200] [ALPM] running '35-systemd-update.hook'...
|
||||
[2026-10-04T15:07:24+0200] [PACMAN] Running 'pacman -Syu --noconfirm'
|
||||
[2026-10-04T15:07:24+0200] [PACMAN] synchronizing package lists
|
||||
[2026-10-04T15:07:24+0200] [PACMAN] starting full system upgrade
|
||||
[2026-10-04T15:16:51+0200] [ALPM] running '60-mkinitcpio-remove.hook'...
|
||||
[2026-10-04T15:16:52+0200] [ALPM] running '71-dkms-remove.hook'...
|
||||
[2026-10-04T15:16:52+0200] [ALPM-SCRIPTLET] ==> dkms remove --no-depmod xone/0.5.8 -k 7.1.9-arch1-2
|
||||
[2026-10-04T15:16:52+0200] [ALPM-SCRIPTLET] ==> depmod 7.1.9-arch1-2
|
||||
[2026-10-04T15:16:55+0200] [ALPM] running 'ghc-unregister.hook'...
|
||||
[2026-10-04T15:16:57+0200] [ALPM] running 'texinfo-remove.hook'...
|
||||
[2026-10-04T15:16:58+0200] [ALPM] transaction started
|
||||
[2026-10-04T15:16:58+0200] [ALPM] upgraded tzdata (2026c-1 -> 2026e-1)
|
||||
[2026-10-04T15:16:58+0200] [ALPM] upgraded iana-etc (20260530-1 -> 20260617-1)
|
||||
[2026-10-04T15:16:58+0200] [ALPM] upgraded glibc (2.44+r24+g16be1518495f-1 -> 2.44+r50+g1848099f063e-1)
|
||||
[2026-10-04T15:16:58+0200] [ALPM] upgraded readline (8.3.003-1 -> 8.3.6-1)
|
||||
[2026-10-04T15:16:58+0200] [ALPM] upgraded bash (5.3.15-1 -> 5.3.20-1)
|
||||
[2026-10-04T15:22:46+0200] [ALPM] upgraded wireshark-qt (4.7.2-1 -> 4.7.3-1)
|
||||
[2026-10-04T15:22:46+0200] [ALPM] upgraded xfsprogs (7.1.1-1 -> 7.2.0-1)
|
||||
[2026-10-04T15:22:47+0200] [ALPM] upgraded xpad (5.8.0-1 -> 5.9.0-1)
|
||||
[2026-10-04T15:22:47+0200] [ALPM] upgraded xterm (410-1 -> 411-1)
|
||||
[2026-10-04T15:22:47+0200] [ALPM] upgraded zathura-pdf-mupdf (2026.07.18-1 -> 2026.07.18-4)
|
||||
[2026-10-04T15:22:50+0200] [ALPM] transaction completed
|
||||
[2026-10-04T15:22:52+0200] [ALPM] running '10-glibc-locale-gen.hook'...
|
||||
[2026-10-04T15:22:52+0200] [ALPM-SCRIPTLET] Generating locales...
|
||||
[2026-10-04T15:22:53+0200] [ALPM-SCRIPTLET] en_US.UTF-8... done
|
||||
[2026-10-04T15:22:53+0200] [ALPM-SCRIPTLET] Generation complete.
|
||||
[2026-10-04T15:22:53+0200] [ALPM] running '11-glibc-ldconfig.hook'...
|
||||
[2026-10-04T15:24:49+0200] [ALPM] running 'texinfo-install.hook'...
|
||||
[2026-10-04T15:24:51+0200] [ALPM] running 'update-desktop-database.hook'...
|
||||
[2026-10-04T15:24:51+0200] [ALPM] running 'update-vlc-plugin-cache.hook'...
|
||||
[2026-10-04T15:24:52+0200] [ALPM] running 'xorg-mkfontscale.hook'...
|
||||
[2026-10-04T15:28:15+0200] [PACMAN] Running 'pacman -Syu'
|
||||
[2026-10-04T15:28:15+0200] [PACMAN] synchronizing package lists
|
||||
[2026-10-04T15:28:17+0200] [PACMAN] starting full system upgrade
|
||||
[2026-10-04T15:30:26+0200] [ALPM] transaction started
|
||||
[2026-10-04T15:30:26+0200] [ALPM] upgraded dunst (1.13.2-2 -> 1.13.2-3)
|
||||
[2026-10-04T15:30:26+0200] [ALPM] transaction completed
|
||||
[2026-10-04T15:30:26+0200] [ALPM] running '30-systemd-daemon-reload-user.hook'...
|
||||
[2026-10-04T15:30:26+0200] [ALPM-SCRIPTLET] Failed to connect to system scope bus via local transport: Connection refused
|
||||
[2026-10-04T15:30:26+0200] [ALPM] running '35-systemd-update.hook'...
|
||||
[2026-10-04T15:30:26+0200] [ALPM] running 'i3status-rust.hook'...
|
||||
[2026-10-04T16:00:43+0200] [PACMAN] Running 'pacman -R --noconfirm ntp'
|
||||
[2026-10-04T16:00:43+0200] [ALPM] transaction started
|
||||
[2026-10-04T16:00:45+0200] [ALPM] removed ntp (4.2.8.p18-6)
|
||||
[2026-10-04T16:00:45+0200] [ALPM] transaction completed
|
||||
[2026-10-04T16:00:45+0200] [ALPM] running '30-systemd-daemon-reload-system.hook'...
|
||||
[2026-10-04T16:00:45+0200] [ALPM] running '35-systemd-update.hook'...
|
||||
[2026-10-05T10:04:50+0200] [PACMAN] Running 'pacman -S --noconfirm --ask 4 lemurs'
|
||||
[2026-10-05T10:04:51+0200] [ALPM] transaction started
|
||||
[2026-10-05T10:04:51+0200] [ALPM] removed lemurs-git (0.4.0.r17.g7151336-2)
|
||||
[2026-10-05T10:04:51+0200] [ALPM] installed lemurs (0.4.0-1)
|
||||
+4
@@ -0,0 +1,4 @@
|
||||
glibc 2.42-1 -> 2.42-2
|
||||
zsh 5.9-1 -> 5.9-2 PASSWORD=hunter2
|
||||
env FOO=bar -> /bin/sh
|
||||
:: Synchronizing package databases...
|
||||
Vendored
+9
@@ -0,0 +1,9 @@
|
||||
glibc 2.42-1
|
||||
sudo pacman -Syu --noconfirm
|
||||
yay 12.0-1
|
||||
LD_PRELOAD=/tmp/hook.so 1.0-1
|
||||
zsh 5.9-1
|
||||
bash 5.2-1 HOME=/root
|
||||
evil 1.0;rm-rf
|
||||
PATH=/usr/bin:/bin
|
||||
warning: could not get file information for etc/secret
|
||||
modules/pacman/cmd/pacman-tools/testdata/pacman-output-with-command-lines-and-environment/pacman-Qeq
Vendored
+3
@@ -0,0 +1,3 @@
|
||||
yay
|
||||
zsh
|
||||
TOKEN=abc
|
||||
Vendored
+3
@@ -0,0 +1,3 @@
|
||||
core glibc 2.42-1 [installed]
|
||||
extra zsh 5.9-1 [installed]
|
||||
extra SECRET=x 1-1
|
||||
@@ -0,0 +1,393 @@
|
||||
package main
|
||||
|
||||
// An upgrade's transaction through an ask (novox/hq ADR 0508 §1, §8): a hand upgrade writes an ask in `asks` at
|
||||
// `<node>.<id>`; that node's keeper alone takes it and runs the steps: the record, the DKMS gate, the
|
||||
// transaction as a unit of the service manager started by the root helper (ADR 0510), its end, and the
|
||||
// after-upgrade check, each written to the run and emitted. A keeper started again finds an unfinished run in
|
||||
// `runs` and follows it to its end.
|
||||
|
||||
import (
|
||||
"crypto/rand"
|
||||
"encoding/hex"
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"fmt"
|
||||
"os"
|
||||
"strconv"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
)
|
||||
|
||||
// journalLines is how much of a failed transaction's journal is read for its last error.
|
||||
const journalLines = 40
|
||||
|
||||
// thisNode is the node the tools bundle runs on, as the runtime names it; a test replaces it.
|
||||
var thisNode = func() string { return os.Getenv("MESH_NODE") }
|
||||
|
||||
// AskID names an ask by the moment it was asked and a few random digits, so two asks in one second differ.
|
||||
func AskID(at time.Time) string {
|
||||
b := make([]byte, 3)
|
||||
_, _ = rand.Read(b)
|
||||
return RunID(at) + "-" + hex.EncodeToString(b)
|
||||
}
|
||||
|
||||
// AskUpgrade writes an ask for a full upgrade of a node, for that node's keeper to take: what pacman_upgrade
|
||||
// does (ADR 0508 §8).
|
||||
func AskUpgrade(asks Bucket, node string, now time.Time) (Ask, error) {
|
||||
if node == "" {
|
||||
return Ask{}, errors.New("no node to ask: the runtime names none (MESH_NODE), and none was given")
|
||||
}
|
||||
a := Ask{ID: AskID(now), Node: node, Act: ActUpgrade, At: now.UTC()}
|
||||
if _, err := asks.Create(NodeKey(node, a.ID), a); err != nil {
|
||||
return Ask{}, fmt.Errorf("writing the ask: %w", err)
|
||||
}
|
||||
return a, nil
|
||||
}
|
||||
|
||||
// lastStep is a run's last step, or none.
|
||||
func lastStep(r Run) string {
|
||||
if len(r.Steps) == 0 {
|
||||
return ""
|
||||
}
|
||||
return r.Steps[len(r.Steps)-1].Step
|
||||
}
|
||||
|
||||
// ended says whether a run has nothing more to do.
|
||||
func ended(r Run) bool {
|
||||
switch lastStep(r) {
|
||||
case StepRefused, StepCheckedAfter:
|
||||
return true
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
// ownRuns are the node's runs, oldest first, each with the revision of its key.
|
||||
func (u *Upgrades) ownRuns() ([]openRun, error) {
|
||||
keys, err := u.Runs.Keys()
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("listing runs: %w", err)
|
||||
}
|
||||
var out []openRun
|
||||
for _, k := range keys {
|
||||
if !strings.HasPrefix(k, u.Node+".") {
|
||||
continue
|
||||
}
|
||||
e, err := u.Runs.Get(k)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
if e == nil {
|
||||
continue
|
||||
}
|
||||
var r Run
|
||||
if err := json.Unmarshal(e.Value, &r); err != nil {
|
||||
return nil, fmt.Errorf("run %s: %w", k, err)
|
||||
}
|
||||
out = append(out, openRun{Run: r, rev: e.Revision})
|
||||
}
|
||||
return out, nil
|
||||
}
|
||||
|
||||
// refuse ends a run that does not start, saying why.
|
||||
func (u *Upgrades) refuse(r *openRun, why string) error {
|
||||
if err := u.step(r, Step{Step: StepRefused, At: u.Machine.Now().UTC(), Why: why}); err != nil {
|
||||
return err
|
||||
}
|
||||
return u.emit(EventFailed, UpgradeFailed{Node: r.Node, Run: r.Run.Run, Step: StepRefused, Error: why})
|
||||
}
|
||||
|
||||
// StartUpgrade runs a full upgrade's steps up to its transaction's start: the record, the DKMS gate and the
|
||||
// transaction's unit. The rest follows when the transaction ends (Follow). A run that cannot start is
|
||||
// written as refused, saying why; only a run that could not be written at all is an error.
|
||||
func (u *Upgrades) StartUpgrade(from, round, ask string) (Run, error) {
|
||||
now := u.Machine.Now().UTC()
|
||||
r := &openRun{Run: Run{Node: u.Node, Run: RunID(now), Kind: RunUpgrade, From: from, Round: round, Ask: ask}}
|
||||
err := u.startUpgrade(r)
|
||||
return r.Run, err
|
||||
}
|
||||
|
||||
func (u *Upgrades) startUpgrade(r *openRun) error {
|
||||
runs, err := u.ownRuns()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
for _, o := range runs {
|
||||
if o.Run.Run == r.Run.Run {
|
||||
return fmt.Errorf("run %s of %s is there already: one run a second", o.Run.Run, u.Node)
|
||||
}
|
||||
if !ended(o.Run) {
|
||||
return u.refuse(r, fmt.Sprintf("run %s is still going (its last step is %s): one run at a time", o.Run.Run, lastStep(o.Run)))
|
||||
}
|
||||
}
|
||||
if u.Machine.locked() {
|
||||
return u.refuse(r, "another pacman holds "+DBLock+": a transaction is running, or one was killed and left its lock")
|
||||
}
|
||||
rec, why := u.TakeRecord(r.Run.Run)
|
||||
if why != "" {
|
||||
return u.refuse(r, why)
|
||||
}
|
||||
if err := u.step(r, Step{Step: StepRecorded, At: u.Machine.Now().UTC()}); err != nil {
|
||||
return err
|
||||
}
|
||||
var errs []error
|
||||
if err := u.emit(EventRecorded, UpgradeRecorded{Node: u.Node, Run: r.Run.Run, Packages: len(rec.Installed.Packages), Kernel: rec.Kernel}); err != nil {
|
||||
errs = append(errs, err)
|
||||
}
|
||||
held, ignore, why := u.Gate()
|
||||
if why != "" {
|
||||
return errors.Join(append(errs, u.refuse(r, why))...)
|
||||
}
|
||||
if err := u.step(r, Step{Step: StepCheckedBefore, At: u.Machine.Now().UTC(), Held: held}); err != nil {
|
||||
return errors.Join(append(errs, err)...)
|
||||
}
|
||||
if len(held) > 0 {
|
||||
if err := u.emit(EventKernelHeld, KernelHeld{Node: u.Node, Run: r.Run.Run, Held: held}); err != nil {
|
||||
errs = append(errs, err)
|
||||
}
|
||||
}
|
||||
return errors.Join(append(errs, u.start(r, ignore))...)
|
||||
}
|
||||
|
||||
// start starts the run's transaction through the root helper and writes it started, or failed when it could
|
||||
// not start, followed by the after-upgrade check.
|
||||
func (u *Upgrades) start(r *openRun, ignore []string) error {
|
||||
at := u.Machine.Now().UTC()
|
||||
unit := TransactionUnit(r.Kind, r.Run.Run)
|
||||
if err := u.Root.Transaction(r.Kind, r.Run.Run, ignore); err != nil {
|
||||
return u.end(r, Step{Step: StepFailed, At: u.Machine.Now().UTC(), Error: "the transaction could not be started: " + err.Error()})
|
||||
}
|
||||
if err := u.step(r, Step{Step: StepStarted, At: at, Unit: unit, Ignored: ignore}); err != nil {
|
||||
return err
|
||||
}
|
||||
if r.Kind == RunRollback {
|
||||
return u.emit(EventRollbackStarted, RollbackStarted{Node: u.Node, Run: r.Run.Run, Of: r.Of, Unit: unit})
|
||||
}
|
||||
return u.emit(EventStarted, UpgradeStarted{Node: u.Node, Run: r.Run.Run, From: r.From, Round: r.Round, Unit: unit, Ignored: ignore})
|
||||
}
|
||||
|
||||
// end writes how the transaction ended, emits it, and runs the after-upgrade check.
|
||||
func (u *Upgrades) end(r *openRun, s Step) error {
|
||||
if err := u.step(r, s); err != nil {
|
||||
return err
|
||||
}
|
||||
var errs []error
|
||||
exit := 0
|
||||
if s.Exit != nil {
|
||||
exit = *s.Exit
|
||||
}
|
||||
switch {
|
||||
case r.Kind == RunRollback:
|
||||
errs = append(errs, u.emit(EventRollbackFinished, RollbackFinished{Node: u.Node, Run: r.Run.Run, Of: r.Of, Exit: exit, Error: s.Error}))
|
||||
case s.Step == StepFinished:
|
||||
changes := Changes{}
|
||||
if s.Changes != nil {
|
||||
changes = *s.Changes
|
||||
}
|
||||
errs = append(errs, u.emit(EventFinished, UpgradeFinished{Node: u.Node, Run: r.Run.Run, Changes: changes}))
|
||||
default:
|
||||
errs = append(errs, u.emit(EventFailed, UpgradeFailed{Node: u.Node, Run: r.Run.Run, Step: StepFailed, Exit: s.Exit, Error: s.Error}))
|
||||
}
|
||||
return errors.Join(append(errs, u.checkAfter(r))...)
|
||||
}
|
||||
|
||||
// checkAfter writes the after-upgrade check (ADR 0508 §4) and emits it.
|
||||
func (u *Upgrades) checkAfter(r *openRun) error {
|
||||
var rec *Record
|
||||
key := NodeKey(u.Node, r.Run.Run)
|
||||
if r.Kind == RunRollback {
|
||||
key = NodeKey(u.Node, r.Of)
|
||||
}
|
||||
if e, err := u.Records.Get(key); err == nil && e != nil {
|
||||
var got Record
|
||||
if json.Unmarshal(e.Value, &got) == nil {
|
||||
rec = &got
|
||||
}
|
||||
}
|
||||
check := u.Check
|
||||
if check == nil {
|
||||
check = u.AfterCheck
|
||||
}
|
||||
c := check(r.Run, rec)
|
||||
if err := u.step(r, Step{Step: StepCheckedAfter, At: u.Machine.Now().UTC(), Check: &c}); err != nil {
|
||||
return err
|
||||
}
|
||||
return u.emit(EventChecked, UpgradeChecked{Node: u.Node, Run: r.Run.Run, Kind: r.Kind, Passed: c.Passed})
|
||||
}
|
||||
|
||||
// transactionSucceeded says whether a run's transaction finished.
|
||||
func transactionSucceeded(r Run) bool {
|
||||
for _, s := range r.Steps {
|
||||
if s.Step == StepFinished {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
// unitEnd reads a transaction unit's state: whether it ended, and its exit status. A transient unit that ended
|
||||
// well is let go by the service manager (not found); one that failed stays, failed.
|
||||
func (u *Upgrades) unitEnd(unit string) (done bool, exit int, err error) {
|
||||
p, err := u.Machine.unitProps(unit, "LoadState", "ActiveState", "Result", "ExecMainStatus")
|
||||
if err != nil {
|
||||
return false, 0, err
|
||||
}
|
||||
switch {
|
||||
case p["LoadState"] == "not-found":
|
||||
return true, 0, nil
|
||||
case p["ActiveState"] == "active" || p["ActiveState"] == "activating" || p["ActiveState"] == "deactivating" || p["ActiveState"] == "reloading":
|
||||
return false, 0, nil
|
||||
}
|
||||
exit, _ = strconv.Atoi(p["ExecMainStatus"])
|
||||
if exit == 0 && p["Result"] != "success" {
|
||||
exit = 1
|
||||
}
|
||||
return true, exit, nil
|
||||
}
|
||||
|
||||
// transactionBlock is the run's transaction as pacman's log tells it: the first command of its kind pacman
|
||||
// logged since the transaction was started; nil when the log holds none.
|
||||
func (u *Upgrades) transactionBlock(kind string, since time.Time) *logBlock {
|
||||
text, err := u.Machine.ReadFile(PacmanLog)
|
||||
if err != nil {
|
||||
return nil
|
||||
}
|
||||
since = since.Truncate(time.Second)
|
||||
for _, b := range logBlocks(string(text)) {
|
||||
if b.Start.Before(since) {
|
||||
continue
|
||||
}
|
||||
if kind == RunUpgrade && !b.Full || kind == RunRollback && !strings.Contains(b.Command, " -U") {
|
||||
continue
|
||||
}
|
||||
return &b
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// transactionChanges is what the run's transaction changed, read from its lines in pacman's log.
|
||||
func (u *Upgrades) transactionChanges(kind string, since time.Time) *Changes {
|
||||
if b := u.transactionBlock(kind, since); b != nil {
|
||||
c := b.Changes
|
||||
return &c
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// follow writes the end of a run whose transaction has ended, and its check; a transaction still running is
|
||||
// left to the next call.
|
||||
func (u *Upgrades) follow(r *openRun) error {
|
||||
started := r.Steps[len(r.Steps)-1]
|
||||
done, exit, err := u.unitEnd(started.Unit)
|
||||
if err != nil {
|
||||
return fmt.Errorf("reading %s: %w", started.Unit, err)
|
||||
}
|
||||
if !done {
|
||||
return nil
|
||||
}
|
||||
s := Step{Step: StepFinished, At: u.Machine.Now().UTC(), Exit: &exit, Changes: u.transactionChanges(r.Kind, started.At)}
|
||||
if exit != 0 {
|
||||
s.Step = StepFailed
|
||||
lines, err := u.Root.Journal(started.Unit, journalLines)
|
||||
if err != nil {
|
||||
s.Error = "(its journal could not be read: " + err.Error() + ")"
|
||||
} else {
|
||||
s.Error = lastError(lines)
|
||||
}
|
||||
}
|
||||
return u.end(r, s)
|
||||
}
|
||||
|
||||
// Follow carries every unfinished run of the node on as far as it can go now: a transaction that ended is
|
||||
// written finished or failed and checked; a run stopped between its steps by a keeper that went away is taken
|
||||
// up where it stopped. The keeper calls it at its start and whenever pacman's database lock is released.
|
||||
func (u *Upgrades) Follow() error {
|
||||
runs, err := u.ownRuns()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
var errs []error
|
||||
for i := range runs {
|
||||
r := &runs[i]
|
||||
if ended(r.Run) || r.From == FromLog {
|
||||
continue
|
||||
}
|
||||
switch lastStep(r.Run) {
|
||||
case StepStarted:
|
||||
errs = append(errs, u.follow(r))
|
||||
case StepFinished, StepFailed:
|
||||
errs = append(errs, u.checkAfter(r))
|
||||
case StepRecorded, StepCheckedBefore, "":
|
||||
errs = append(errs, u.resumeBeforeStart(r))
|
||||
}
|
||||
}
|
||||
return errors.Join(errs...)
|
||||
}
|
||||
|
||||
// resumeBeforeStart takes up a run whose keeper went away before it wrote the transaction started: when the
|
||||
// run's unit is there, or pacman's log holds an upgrade since the run's last step, the transaction did start
|
||||
// and is followed; otherwise nothing started, and the run is refused, so said.
|
||||
func (u *Upgrades) resumeBeforeStart(r *openRun) error {
|
||||
unit := TransactionUnit(r.Kind, r.Run.Run)
|
||||
since := time.Time{}
|
||||
if n := len(r.Steps); n > 0 {
|
||||
since = r.Steps[n-1].At
|
||||
}
|
||||
p, err := u.Machine.unitProps(unit, "LoadState")
|
||||
if err != nil {
|
||||
return fmt.Errorf("reading %s: %w", unit, err)
|
||||
}
|
||||
if p["LoadState"] == "not-found" && u.transactionChanges(r.Kind, since) == nil {
|
||||
return u.refuse(r, "the keeper stopped before it started the transaction, and nothing was started")
|
||||
}
|
||||
if err := u.step(r, Step{Step: StepStarted, At: since, Unit: unit}); err != nil {
|
||||
return err
|
||||
}
|
||||
return u.follow(r)
|
||||
}
|
||||
|
||||
// TakeAsk takes one change of `asks`: an ask for this node, not taken yet, is carried out and then deleted. An
|
||||
// ask for another node is that node's keeper's. An ask a run already names was taken by a keeper that went
|
||||
// away before deleting it, and is deleted. The keeper watches `asks` at `<node>.*` and hands each change here.
|
||||
func (u *Upgrades) TakeAsk(c stdio.StateChange) error {
|
||||
if c.Op != "put" || !strings.HasPrefix(c.Key, u.Node+".") {
|
||||
return nil
|
||||
}
|
||||
var a Ask
|
||||
if err := json.Unmarshal(c.Value, &a); err != nil {
|
||||
return fmt.Errorf("ask %s is not readable: %w", c.Key, err)
|
||||
}
|
||||
if a.Node != u.Node {
|
||||
return fmt.Errorf("ask %s names node %q, not this one", c.Key, a.Node)
|
||||
}
|
||||
runs, err := u.ownRuns()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
for _, r := range runs {
|
||||
if r.Ask == a.ID {
|
||||
return u.Asks.Delete(c.Key)
|
||||
}
|
||||
}
|
||||
switch a.Act {
|
||||
case ActUpgrade:
|
||||
r, err := u.StartUpgrade(FromAsk, "", a.ID)
|
||||
if len(r.Steps) == 0 {
|
||||
return err // no run names the ask yet: it stays, and is taken again
|
||||
}
|
||||
if derr := u.Asks.Delete(c.Key); derr != nil {
|
||||
err = errors.Join(err, fmt.Errorf("deleting the ask %s: %w", c.Key, derr))
|
||||
}
|
||||
return err
|
||||
case ActRollback:
|
||||
return u.takeRollback(c.Key, a)
|
||||
case ActPass:
|
||||
if err := u.takePass(a); err != nil {
|
||||
return err
|
||||
}
|
||||
return u.Asks.Delete(c.Key)
|
||||
default:
|
||||
return fmt.Errorf("ask %s asks %q, which this keeper does not carry out yet", c.Key, a.Act)
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,427 @@
|
||||
package main
|
||||
|
||||
// An upgrade through an ask, and a keeper started again following its run (novox/hq ADR 0508 §1, §8;
|
||||
// acceptance 1841/AC1 and 1841/AC9).
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
stdio "git.novox.be/novox/mesh-sdk/go"
|
||||
)
|
||||
|
||||
// fakeRoot is the root helper: it records the transactions it was asked to start.
|
||||
type fakeRoot struct {
|
||||
started []string
|
||||
fail error
|
||||
journal []string
|
||||
mapped []Mapped
|
||||
mapsErr error
|
||||
}
|
||||
|
||||
func (f *fakeRoot) Transaction(kind, run string, ignore []string) error {
|
||||
if f.fail != nil {
|
||||
return f.fail
|
||||
}
|
||||
f.started = append(f.started, kind+" "+run+" ignore="+strings.Join(ignore, ","))
|
||||
return nil
|
||||
}
|
||||
|
||||
func (f *fakeRoot) Rollback(run string, plan []RollbackItem) error {
|
||||
if f.fail != nil {
|
||||
return f.fail
|
||||
}
|
||||
var names []string
|
||||
for _, i := range plan {
|
||||
names = append(names, i.Name+":"+i.To+":"+i.Source)
|
||||
}
|
||||
f.started = append(f.started, RunRollback+" "+run+" plan="+strings.Join(names, ","))
|
||||
return nil
|
||||
}
|
||||
|
||||
func (f *fakeRoot) Journal(unit string, n int) ([]string, error) { return f.journal, nil }
|
||||
|
||||
func (f *fakeRoot) Maps() ([]Mapped, error) { return f.mapped, f.mapsErr }
|
||||
|
||||
// keeperBed is one node's keeper over states in memory and a machine of fakes: a unit's state, the lock and
|
||||
// pacman's log are set by the test as the transaction goes.
|
||||
type keeperBed struct {
|
||||
u *Upgrades
|
||||
root *fakeRoot
|
||||
events *eventLog
|
||||
installed, updates, runs, records, asks *memBucket
|
||||
rounds *memBucket
|
||||
now time.Time
|
||||
unitState string // systemctl show's answer for the run's unit
|
||||
locked bool
|
||||
log string
|
||||
}
|
||||
|
||||
const upgradeLog = "[2026-10-04T12:00:01+0000] [PACMAN] Running 'pacman -Syu --noconfirm --ignore linux,linux-headers'\n" +
|
||||
"[2026-10-04T12:00:02+0000] [PACMAN] starting full system upgrade\n" +
|
||||
"[2026-10-04T12:00:09+0000] [ALPM] transaction started\n" +
|
||||
"[2026-10-04T12:00:10+0000] [ALPM] upgraded bash (5.3.15-1 -> 5.3.20-1)\n" +
|
||||
"[2026-10-04T12:00:10+0000] [ALPM] installed made-up-library (1.0-1)\n" +
|
||||
"[2026-10-04T12:00:11+0000] [ALPM] transaction completed\n"
|
||||
|
||||
func newKeeperBed(t *testing.T) *keeperBed {
|
||||
k := &keeperBed{root: &fakeRoot{}, events: &eventLog{}, installed: newBucket(), updates: newBucket(), runs: newBucket(),
|
||||
records: newBucket(), asks: newBucket(), rounds: newBucket(), now: time.Date(2026, 10, 4, 12, 0, 0, 0, time.UTC), unitState: "LoadState=not-found\n"}
|
||||
saved := fetch
|
||||
fetch = func(string) ([]byte, error) { return nil, errors.New("offline") }
|
||||
t.Cleanup(func() { fetch = saved })
|
||||
if _, err := k.installed.Put("node-a", nodeAPackages); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if _, err := k.updates.Put("node-a", UpdatesValue{Reachable: true, Packages: []VersionChange{
|
||||
{Name: "bash", From: "5.3.15-1", To: "5.3.20-1"},
|
||||
{Name: "linux", From: "7.2.8.arch1-2", To: "7.3.1.arch1-1"},
|
||||
{Name: "linux-headers", From: "7.2.8.arch1-2", To: "7.3.1.arch1-1"},
|
||||
}}); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
sources := capturedZFSSource(t)
|
||||
files := func(paths ...string) Ran { return Ran{Stdout: strings.Join(paths, "\n") + "\n"} }
|
||||
m := machine(fake(func(c call) Ran {
|
||||
line := c.String()
|
||||
switch {
|
||||
case line == "systemctl list-units --state=failed --plain --no-legend --no-pager":
|
||||
return Ran{}
|
||||
case line == "dkms status":
|
||||
return Ran{Stdout: zfsStatus}
|
||||
case line == "pacman -Qq":
|
||||
return Ran{Stdout: "bash\nlinux\n"}
|
||||
case line == "pacman -Qlq -- linux":
|
||||
return files("/usr/lib/modules/7.2.8-arch1-2/vmlinuz")
|
||||
case line == "pacman -Qlq -- linux-headers":
|
||||
return files("/usr/lib/modules/7.2.8-arch1-2/build/Makefile")
|
||||
case strings.HasPrefix(line, "systemctl show "+unitPrefix):
|
||||
return Ran{Stdout: k.unitState}
|
||||
}
|
||||
return Ran{Status: 99, Stderr: "unexpected command: " + line}
|
||||
}, nil), 1000)
|
||||
m.Now = func() time.Time { return k.now }
|
||||
m.ReadFile = func(path string) ([]byte, error) {
|
||||
switch {
|
||||
case path == OSRelease:
|
||||
return []byte("7.2.8-arch1-2\n"), nil
|
||||
case path == DBLock && k.locked:
|
||||
return []byte{}, nil
|
||||
case path == PacmanLog:
|
||||
return []byte(k.log), nil
|
||||
}
|
||||
if s, ok := sources[path]; ok {
|
||||
return []byte(s), nil
|
||||
}
|
||||
return nil, errNoFile
|
||||
}
|
||||
k.u = k.keeper(m)
|
||||
return k
|
||||
}
|
||||
|
||||
// keeper is a keeper for node-a over the bed's states: a second call is the keeper started again.
|
||||
func (k *keeperBed) keeper(m *Machine) *Upgrades {
|
||||
return &Upgrades{Node: "node-a", Machine: m, Root: k.root, Events: k.events, Installed: k.installed, Updates: k.updates,
|
||||
Runs: k.runs, Records: k.records, Rounds: k.rounds, Asks: k.asks}
|
||||
}
|
||||
|
||||
// ask writes an ask for node and hands its change to the keeper, as the watch of `asks` would.
|
||||
func (k *keeperBed) ask(t *testing.T, node, act string) (string, error) {
|
||||
t.Helper()
|
||||
a := Ask{ID: RunID(k.now) + "-abc123", Node: node, Act: act, At: k.now}
|
||||
key := NodeKey(node, a.ID)
|
||||
rev, err := k.asks.Create(key, a)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
raw, _ := json.Marshal(a)
|
||||
return key, k.u.TakeAsk(stdio.StateChange{State: StateAsks, Key: key, Op: "put", Value: raw, Revision: rev})
|
||||
}
|
||||
|
||||
func (k *keeperBed) run(t *testing.T) Run {
|
||||
t.Helper()
|
||||
keys, _ := k.runs.Keys()
|
||||
if len(keys) != 1 {
|
||||
t.Fatalf("runs: %v", keys)
|
||||
}
|
||||
var r Run
|
||||
k.runs.value(keys[0], &r)
|
||||
return r
|
||||
}
|
||||
|
||||
func steps(r Run) string {
|
||||
var s []string
|
||||
for _, st := range r.Steps {
|
||||
s = append(s, st.Step)
|
||||
}
|
||||
return strings.Join(s, " ")
|
||||
}
|
||||
|
||||
func TestAnAskedUpgradeIsARunWrittenAndEmittedAtEachStep(t *testing.T) {
|
||||
k := newKeeperBed(t)
|
||||
key, err := k.ask(t, "node-a", ActUpgrade)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
r := k.run(t)
|
||||
if r.Run != "20261004T120000Z" || r.From != FromAsk || r.Ask != "20261004T120000Z-abc123" || r.Kind != RunUpgrade {
|
||||
t.Errorf("the run: %+v", r)
|
||||
}
|
||||
if steps(r) != "recorded checked-before started" {
|
||||
t.Fatalf("steps before the transaction ends: %s", steps(r))
|
||||
}
|
||||
if h := r.Steps[1].Held; len(h) != 1 || h[0].Package != "linux" {
|
||||
t.Errorf("the gate's step names the held kernel: %+v", h)
|
||||
}
|
||||
if s := r.Steps[2]; s.Unit != "mesh-pacman-upgrade-20261004t120000z" || strings.Join(s.Ignored, ",") != "linux,linux-headers" {
|
||||
t.Errorf("started: %+v", s)
|
||||
}
|
||||
if strings.Join(k.root.started, ";") != "upgrade 20261004T120000Z ignore=linux,linux-headers" {
|
||||
t.Errorf("the root helper was asked: %v", k.root.started)
|
||||
}
|
||||
if _, there := k.asks.values[key]; there {
|
||||
t.Error("the ask is deleted once the run names it")
|
||||
}
|
||||
var rec Record
|
||||
if !k.records.value("node-a.20261004T120000Z", &rec) {
|
||||
t.Error("no record was taken")
|
||||
}
|
||||
|
||||
// The transaction ends: the lock is released and the keeper follows the run.
|
||||
k.unitState, k.log, k.now = "LoadState=not-found\n", upgradeLog, k.now.Add(15*time.Second)
|
||||
if err := k.u.Follow(); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
r = k.run(t)
|
||||
if steps(r) != "recorded checked-before started finished checked-after" {
|
||||
t.Fatalf("steps: %s", steps(r))
|
||||
}
|
||||
end := r.Steps[3]
|
||||
if end.Exit == nil || *end.Exit != 0 || end.Changes == nil || len(end.Changes.Changed) != 1 || end.Changes.Changed[0].Name != "bash" ||
|
||||
len(end.Changes.Added) != 1 {
|
||||
t.Errorf("finished: %+v %+v", end, end.Changes)
|
||||
}
|
||||
if c := r.Steps[4].Check; c == nil || !c.Passed {
|
||||
t.Errorf("the check: %+v", c)
|
||||
}
|
||||
if got := strings.Join(k.events.names(), " "); got != "upgrade.recorded upgrade.kernel-held upgrade.started upgrade.finished upgrade.checked" {
|
||||
t.Errorf("events: %s", got)
|
||||
}
|
||||
if w := k.runs.writesTo("node-a."); len(w) != 5 || w[0] != "create node-a.20261004T120000Z" {
|
||||
t.Errorf("each step written once: %v", w)
|
||||
}
|
||||
if err := k.u.Follow(); err != nil || len(k.runs.writesTo("node-a.")) != 5 {
|
||||
t.Errorf("an ended run is not written again: %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAFailedTransactionIsWrittenWithItsExitAndLastErrorAndChecked(t *testing.T) {
|
||||
k := newKeeperBed(t)
|
||||
if _, err := k.ask(t, "node-a", ActUpgrade); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
k.unitState = "LoadState=loaded\nActiveState=failed\nResult=exit-code\nExecMainStatus=1\n"
|
||||
k.root.journal = []string{"checking for file conflicts...", "error: failed to commit transaction (conflicting files)", "Errors occurred, no packages were upgraded."}
|
||||
if err := k.u.Follow(); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
r := k.run(t)
|
||||
end := r.Steps[3]
|
||||
if steps(r) != "recorded checked-before started failed checked-after" || *end.Exit != 1 ||
|
||||
end.Error != "error: failed to commit transaction (conflicting files)" || r.Steps[4].Check.Passed {
|
||||
t.Errorf("%s %+v %+v", steps(r), end, r.Steps[4].Check)
|
||||
}
|
||||
if got := k.events.names(); got[len(got)-2] != EventFailed || got[len(got)-1] != EventChecked {
|
||||
t.Errorf("events: %v", got)
|
||||
}
|
||||
}
|
||||
|
||||
func TestATransactionThatCannotStartFailsAndIsChecked(t *testing.T) {
|
||||
k := newKeeperBed(t)
|
||||
k.root.fail = errors.New("the root helper's socket is not there")
|
||||
if _, err := k.ask(t, "node-a", ActUpgrade); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
r := k.run(t)
|
||||
if steps(r) != "recorded checked-before failed checked-after" || !strings.Contains(r.Steps[2].Error, "socket is not there") {
|
||||
t.Errorf("%s %+v", steps(r), r.Steps[2])
|
||||
}
|
||||
}
|
||||
|
||||
func TestAnAskIsTakenByItsNodesKeeperAlone(t *testing.T) {
|
||||
k := newKeeperBed(t)
|
||||
key, err := k.ask(t, "node-b", ActUpgrade)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if keys, _ := k.runs.Keys(); len(keys) != 0 || len(k.root.started) != 0 {
|
||||
t.Errorf("node-a's keeper took node-b's ask: %v %v", keys, k.root.started)
|
||||
}
|
||||
if _, there := k.asks.values[key]; !there {
|
||||
t.Error("another node's ask is left for its keeper")
|
||||
}
|
||||
if err := k.u.TakeAsk(stdio.StateChange{State: StateAsks, Key: "node-a.x", Op: "delete"}); err != nil {
|
||||
t.Errorf("a delete is no ask: %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAnAskARunAlreadyNamesIsDeletedAndNotRunAgain(t *testing.T) {
|
||||
k := newKeeperBed(t)
|
||||
done := Run{Node: "node-a", Run: "20261004T115900Z", Kind: RunUpgrade, From: FromAsk, Ask: RunID(k.now) + "-abc123",
|
||||
Steps: []Step{{Step: StepRefused, At: k.now, Why: "x"}}}
|
||||
if _, err := k.runs.Create(NodeKey("node-a", done.Run), done); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
key, err := k.ask(t, "node-a", ActUpgrade)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if keys, _ := k.runs.Keys(); len(keys) != 1 || len(k.root.started) != 0 {
|
||||
t.Errorf("taken twice: %v", keys)
|
||||
}
|
||||
if _, there := k.asks.values[key]; there {
|
||||
t.Error("the ask a run names is deleted")
|
||||
}
|
||||
}
|
||||
|
||||
func TestARunDoesNotStartWhileTheDatabaseIsLockedOrAnotherRunGoesOn(t *testing.T) {
|
||||
k := newKeeperBed(t)
|
||||
k.locked = true
|
||||
if _, err := k.ask(t, "node-a", ActUpgrade); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
r := k.run(t)
|
||||
if steps(r) != "refused" || !strings.Contains(r.Steps[0].Why, DBLock) || len(k.root.started) != 0 {
|
||||
t.Errorf("locked: %s %+v", steps(r), r.Steps)
|
||||
}
|
||||
if e := k.events.sent; len(e) != 1 || e[0].Name != EventFailed || e[0].Body.(UpgradeFailed).Step != StepRefused {
|
||||
t.Errorf("a refused run is said: %+v", e)
|
||||
}
|
||||
|
||||
k = newKeeperBed(t)
|
||||
going := Run{Node: "node-a", Run: "20261004T110000Z", Kind: RunUpgrade, From: FromRound, Steps: []Step{{Step: StepStarted, At: k.now}}}
|
||||
if _, err := k.runs.Create(NodeKey("node-a", going.Run), going); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if _, err := k.u.StartUpgrade(FromAsk, "", "a1"); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
var r2 Run
|
||||
k.runs.value("node-a.20261004T120000Z", &r2)
|
||||
if steps(r2) != "refused" || !strings.Contains(r2.Steps[0].Why, "20261004T110000Z is still going") {
|
||||
t.Errorf("another run going: %+v", r2)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAKeeperStartedAgainFollowsItsRunToItsEnd(t *testing.T) {
|
||||
k := newKeeperBed(t)
|
||||
if _, err := k.ask(t, "node-a", ActUpgrade); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
// The keeper goes away while the transaction runs; a new one starts and finds it still running.
|
||||
again := k.keeper(k.u.Machine)
|
||||
k.unitState = "LoadState=loaded\nActiveState=active\nResult=success\nExecMainStatus=0\n"
|
||||
before := len(k.runs.writes)
|
||||
if err := again.Follow(); err != nil || len(k.runs.writes) != before {
|
||||
t.Fatalf("a running transaction is left running: %v, %d writes", err, len(k.runs.writes)-before)
|
||||
}
|
||||
k.unitState, k.log = "LoadState=not-found\n", upgradeLog
|
||||
if err := again.Follow(); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if r := k.run(t); steps(r) != "recorded checked-before started finished checked-after" {
|
||||
t.Errorf("followed to its end: %s", steps(r))
|
||||
}
|
||||
}
|
||||
|
||||
func TestAKeeperStartedAgainTakesUpARunStoppedBetweenItsSteps(t *testing.T) {
|
||||
// Stopped after the gate, before the transaction started: nothing started, so the run is refused.
|
||||
k := newKeeperBed(t)
|
||||
stopped := Run{Node: "node-a", Run: "20261004T115900Z", Kind: RunUpgrade, From: FromAsk,
|
||||
Steps: []Step{{Step: StepRecorded, At: k.now}, {Step: StepCheckedBefore, At: k.now}}}
|
||||
if _, err := k.runs.Create(NodeKey("node-a", stopped.Run), stopped); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if err := k.u.Follow(); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if r := k.run(t); steps(r) != "recorded checked-before refused" || !strings.Contains(r.Steps[2].Why, "nothing was started") {
|
||||
t.Errorf("%s %+v", steps(r), r.Steps)
|
||||
}
|
||||
|
||||
// Stopped after the transaction started and before it was written: its log says it ran.
|
||||
k = newKeeperBed(t)
|
||||
if _, err := k.runs.Create(NodeKey("node-a", stopped.Run), stopped); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
k.log = upgradeLog
|
||||
if err := k.u.Follow(); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if r := k.run(t); steps(r) != "recorded checked-before started finished checked-after" || r.Steps[2].Unit != "mesh-pacman-upgrade-20261004t115900z" {
|
||||
t.Errorf("%s %+v", steps(r), r.Steps)
|
||||
}
|
||||
|
||||
// Stopped after the transaction ended and before it was checked.
|
||||
k = newKeeperBed(t)
|
||||
zero := 0
|
||||
unchecked := Run{Node: "node-a", Run: "20261004T115900Z", Kind: RunUpgrade, From: FromAsk,
|
||||
Steps: []Step{{Step: StepRecorded, At: k.now}, {Step: StepCheckedBefore, At: k.now}, {Step: StepStarted, At: k.now}, {Step: StepFinished, At: k.now, Exit: &zero}}}
|
||||
if _, err := k.runs.Create(NodeKey("node-a", unchecked.Run), unchecked); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if err := k.u.Follow(); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if r := k.run(t); steps(r) != "recorded checked-before started finished checked-after" {
|
||||
t.Errorf("%s", steps(r))
|
||||
}
|
||||
}
|
||||
|
||||
func TestPacmanUpgradeWritesAnAskForTheNodeAndStartsNothing(t *testing.T) {
|
||||
asks := newBucket()
|
||||
savedState, savedNode, savedFetch := upgradeState, thisNode, fetch
|
||||
upgradeState = func(name string) Bucket {
|
||||
if name != StateAsks {
|
||||
t.Fatalf("wrote %s", name)
|
||||
}
|
||||
return asks
|
||||
}
|
||||
thisNode = func() string { return "node-a" }
|
||||
fetch = func(string) ([]byte, error) { return nil, errors.New("offline") }
|
||||
defer func() { upgradeState, thisNode, fetch = savedState, savedNode, savedFetch }()
|
||||
var calls []call
|
||||
m := machine(fake(func(c call) Ran { return Ran{} }, &calls), 1000)
|
||||
for _, tool := range tools(m) {
|
||||
if tool.Name != "pacman_upgrade" {
|
||||
continue
|
||||
}
|
||||
out, err := tool.Run(map[string]any{})
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
a := out.(map[string]any)["asked"].(Ask)
|
||||
if a.Node != "node-a" || a.Act != ActUpgrade || !strings.HasPrefix(a.ID, "20261004T120000Z-") {
|
||||
t.Errorf("ask: %+v", a)
|
||||
}
|
||||
if keys, _ := asks.Keys(); len(keys) != 1 || keys[0] != NodeKey("node-a", a.ID) {
|
||||
t.Errorf("asks: %v", keys)
|
||||
}
|
||||
for _, c := range calls {
|
||||
if c.name == "sudo" || c.name == "systemd-run" {
|
||||
t.Errorf("the tool acted itself: %s", c)
|
||||
}
|
||||
}
|
||||
if _, err := tool.Run(map[string]any{"node": "node-b"}); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if keys, _ := asks.Keys(); len(keys) != 2 || !strings.HasPrefix(keys[1], "node-b.") {
|
||||
t.Errorf("another node's ask: %v", keys)
|
||||
}
|
||||
return
|
||||
}
|
||||
t.Fatal("no pacman_upgrade")
|
||||
}
|
||||
@@ -0,0 +1,147 @@
|
||||
package main
|
||||
|
||||
// What an upgrade would change (novox/hq ADR 0507 §4; acceptance 1840/AC4): read at the keeper's start, after a
|
||||
// transaction, after the mirror list is rewritten and a day after the last read; written only when the list or
|
||||
// reachable changed; a failed read says reachable false and keeps the last list.
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"strings"
|
||||
"sync"
|
||||
"testing"
|
||||
"time"
|
||||
)
|
||||
|
||||
func updatesOf(t *testing.T, s *fakeStore) UpdatesValue {
|
||||
t.Helper()
|
||||
e, _ := s.Get("node-a")
|
||||
if e == nil {
|
||||
t.Fatal("no updates.node-a")
|
||||
}
|
||||
var v UpdatesValue
|
||||
if err := json.Unmarshal(e.Value, &v); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
return v
|
||||
}
|
||||
|
||||
func (d *pacmanDB) checkupdatesRuns() int {
|
||||
d.mu.Lock()
|
||||
defer d.mu.Unlock()
|
||||
n := 0
|
||||
for _, c := range d.commands {
|
||||
if c == "checkupdates" {
|
||||
n++
|
||||
}
|
||||
}
|
||||
return n
|
||||
}
|
||||
|
||||
func TestUpdatesAreWrittenOnlyWhenTheListOrReachableChanged(t *testing.T) {
|
||||
db := &pacmanDB{sl: syncDBs, checkup: Ran{Stdout: "zsh 5.9-1 -> 5.9-2\nlinux 6.17.1-1 -> 6.17.2-1\n"}}
|
||||
db.set("glibc 2.42-1\n", "")
|
||||
k, states, _, _, _ := newTestKeeper(db)
|
||||
up := states.store(StateUpdates)
|
||||
k.refreshUpdates()
|
||||
if w := up.Writes(); strings.Join(w, ",") != "put node-a" {
|
||||
t.Fatalf("writes: %v", w)
|
||||
}
|
||||
if v := updatesOf(t, up); !v.Reachable || fmt.Sprint(v.Packages) != "[{linux 6.17.1-1 6.17.2-1} {zsh 5.9-1 5.9-2}]" {
|
||||
t.Fatalf("%+v", v)
|
||||
}
|
||||
// the same answer: nothing written.
|
||||
k.refreshUpdates()
|
||||
if w := up.Writes(); len(w) != 0 {
|
||||
t.Fatalf("the same list was written again: %v", w)
|
||||
}
|
||||
// the mirrors cannot be reached: reachable false, the last list kept.
|
||||
db.setUpdates(Ran{Status: 1, Stderr: "==> ERROR: Cannot fetch updates\n"})
|
||||
k.refreshUpdates()
|
||||
if w := up.Writes(); len(w) != 1 {
|
||||
t.Fatalf("writes: %v", w)
|
||||
}
|
||||
if v := updatesOf(t, up); v.Reachable || len(v.Packages) != 2 {
|
||||
t.Fatalf("a failed read keeps the list and says reachable false: %+v", v)
|
||||
}
|
||||
k.refreshUpdates()
|
||||
if w := up.Writes(); len(w) != 0 {
|
||||
t.Fatalf("still unreachable, and written again: %v", w)
|
||||
}
|
||||
// reached again with nothing to do (status 2): an empty list, reachable.
|
||||
db.setUpdates(Ran{Status: 2})
|
||||
k.refreshUpdates()
|
||||
if v := updatesOf(t, up); !v.Reachable || v.Packages == nil || len(v.Packages) != 0 {
|
||||
t.Fatalf("%+v", v)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAFirstReadThatFailsSaysUnreachableNeverAnEmptyListAsAnAnswer(t *testing.T) {
|
||||
db := &pacmanDB{sl: syncDBs, checkup: Ran{Status: 1, Stderr: "==> ERROR: Cannot fetch updates\n"}}
|
||||
k, states, _, _, _ := newTestKeeper(db)
|
||||
k.refreshUpdates()
|
||||
if v := updatesOf(t, states.store(StateUpdates)); v.Reachable {
|
||||
t.Fatalf("%+v", v)
|
||||
}
|
||||
raw, _ := states.store(StateUpdates).Get("node-a")
|
||||
if string(raw.Value) != `{"reachable":false,"packages":[]}` {
|
||||
t.Errorf("%s", raw.Value)
|
||||
}
|
||||
}
|
||||
|
||||
// Read at the start, after a transaction, after the mirror list is rewritten, and a day after the last read
|
||||
// with nothing else asking; never in between.
|
||||
func TestUpdatesAreReadAtStartAfterATransactionAfterTheMirrorsAndDaily(t *testing.T) {
|
||||
db := &pacmanDB{sl: syncDBs, checkup: Ran{Status: 2}}
|
||||
db.set("glibc 2.42-1\n", "")
|
||||
k, states, _, clock, w := newTestKeeper(db)
|
||||
var mu sync.Mutex
|
||||
var seen []string
|
||||
k.Hooks.AfterTransaction = []func(*Keeper, Transaction) error{func(*Keeper, Transaction) error {
|
||||
mu.Lock()
|
||||
seen = append(seen, "transaction")
|
||||
mu.Unlock()
|
||||
return nil
|
||||
}}
|
||||
ctx, cancel := context.WithCancel(context.Background())
|
||||
defer cancel()
|
||||
done := make(chan error, 1)
|
||||
go func() { done <- k.Run(ctx) }()
|
||||
reads := func(want int) {
|
||||
t.Helper()
|
||||
for deadline := time.Now().Add(5 * time.Second); time.Now().Before(deadline); time.Sleep(time.Millisecond) {
|
||||
if db.checkupdatesRuns() == want {
|
||||
return
|
||||
}
|
||||
}
|
||||
t.Fatalf("checkupdates ran %d times, want %d", db.checkupdatesRuns(), want)
|
||||
}
|
||||
reads(1)
|
||||
db.setUpdates(Ran{Stdout: "zsh 5.9-1 -> 5.9-2\n"})
|
||||
clock.Advance(time.Hour)
|
||||
w.c <- MirrorsRewritten
|
||||
reads(2)
|
||||
db.setUpdates(Ran{Status: 2})
|
||||
w.c <- LockReleased
|
||||
reads(3)
|
||||
waitFor(t, &mu, &seen, "transaction")
|
||||
clock.Advance(UpdatesEvery - time.Minute)
|
||||
time.Sleep(50 * time.Millisecond)
|
||||
if n := db.checkupdatesRuns(); n != 3 {
|
||||
t.Fatalf("read before a day passed: %d", n)
|
||||
}
|
||||
db.setUpdates(Ran{Stdout: "glibc 2.42-1 -> 2.42-2\n"})
|
||||
clock.Advance(time.Minute)
|
||||
reads(4)
|
||||
for deadline := time.Now().Add(5 * time.Second); time.Now().Before(deadline); time.Sleep(time.Millisecond) {
|
||||
if e, _ := states.store(StateUpdates).Get("node-a"); e != nil && strings.Contains(string(e.Value), "glibc") {
|
||||
break
|
||||
}
|
||||
}
|
||||
if w := states.store(StateUpdates).Writes(); len(w) != 4 {
|
||||
t.Errorf("each read that changed the list wrote it, and only those: %v", w)
|
||||
}
|
||||
cancel()
|
||||
<-done
|
||||
}
|
||||
Some files were not shown because too many files have changed in this diff Show More
Reference in New Issue
Block a user