Compare commits

..
Author SHA1 Message Date
jschoubben 0412653920 The store keeps what the records name (hq ADR 0189)
The mesh names what may go from its own build records — a digest it did not
record making is never named, which is what keeps the sweep away from the
images genesis pushed. An artifact stays because a definition the mesh holds
names it, or because it belongs to one of the five most recent successful
builds of its module.

internal/artifacts asks the store to let go of one; internal/inventory
decides and remembers (migration 0055); the sweep runs after a build the mesh
recorded, which is when both the bytes and the keep set moved. Never fatal to
a build.

And the manifest side of while-stopped, refused from the definition alone:
no schedule, run-once, a container the module does not declare, itself.
2026-10-02 21:49:03 +02:00
jschoubben e09a14ba4c A served value may name the consumer it is served to (hq ADR 0188)
${consumer:as} and ${consumer:as:dns} in a serves block are filled per
consumer at resolution, and the one filled value reaches both ends: the
consumer's binding and its ${bound:...} substitutions, and the provider's
contributions entry as `derived`. A fact or alphabet the mesh does not have
is refused at parse; a consumer whose own file already holds the derived
value is refused at resolution, naming the placeholder to write instead.
2026-10-02 21:24:43 +02:00
mesh-admin 1a44d281c2 Merge pull request 'node show cites ADR 0180 for a removed front end (hq ADR 0186)' (#224) from fix/a-ban-list-never-holds-a-neighbour into main 2026-10-02 16:47:25 +00:00
jschoubben 1c8fe65601 node show cites ADR 0180 for a removed front end (hq ADR 0186)
Another session took 0175 while that record was in review; the line printed on every converged
machine was pointing at an unrelated decision.
2026-10-02 18:42:16 +02:00
mesh-admin bea1a1c513 Merge pull request 'A control plane behind its seat's row serves what it can (hq ADR 0185)' (#222) from fix/a-service-asked-to-run-is-still-running into main 2026-10-02 16:21:55 +00:00
jschoubben 21d38c9b0e A control plane behind its seat's row serves what it can (hq ADR 0185)
One verb in the row that this binary cannot run aborted the start, and a stale push that put an
older control plane back took the whole mesh off the bus for ten minutes — recoverable only by a
person running the binary outside its service, because the push that repairs it is one of the verbs
that had stopped being served. Now the verbs it knows are served, the ones it does not answer the
reason, and the start names them once.
2026-10-02 18:16:24 +02:00
mesh-admin 689dd060b0 Merge pull request 'A jail's pattern names <HOST> once, and is refused by name when it does not (hq ADR 0179)' (#221) from fix/a-jails-pattern-names-the-host-once into main 2026-10-02 15:32:14 +00:00
jschoubben 99c4c4ef04 A jail's pattern names <HOST> once, and is refused by name when it does not (hq ADR 0179)
fail2ban expands <HOST> into a named capture group, so a pattern naming it twice is a duplicate
group name: the daemon refuses its whole configuration and exits, and the machine keeps no bans at
all — for every jail, not the one at fault. Hit live on the control node the day the jails shipped.
A jail with no name, no pattern, or a name another of the module's jails took is refused too.
2026-10-02 17:25:34 +02:00
mesh-admin e5e1666f91 Merge pull request 'The intrusion-prevention seat serves status, banned, ban and unban; the proxy logs a refused name (hq ADR 0179)' (#219) from feat/the-intrusion-seat-serves-its-verbs into main 2026-10-02 15:15:41 +00:00
jschoubben bd58d1c3ef The intrusion-prevention seat serves status, banned, ban and unban; the proxy logs a refused name (hq ADR 0179)
Every holder of node-intrusion-prevention owes the four verbs, as the packet filter's holder owes
its three (ADR 0170). The proxy says in its log when a name this mesh does not serve is asked for,
with the asking address last, so its own jail can read it.
2026-10-02 17:02:49 +02:00
mesh-admin d134c89a7c Merge pull request 'The service manager is a node seat with the unit verbs, and a user shape names the account (hq ADR 0176, 0177)' (#218) from feat/the-operators-machine into main 2026-10-02 14:58:50 +00:00
jochen a9307d9f33 The controller seat gains a generic verb: command runs one line of the binary and answers what it printed
Beside the named verbs, `command` takes a command line as the controller's
own shell would — `node account g14 jochen`, `node show ace`, `module list` —
splits it as a shell does (quotes group, backslash escapes, nothing expanded)
and runs it in this binary like every other verb. The named verbs keep their
schemas; this is the whole binary, added because the operator decided any
node may call any tool (hq ADR 0175) and a verb per command was the only
thing keeping the rest behind a shell on the control node. ADR 0154 carries
the dated note. Tests: a plain line, quoted words, an empty line and an
unclosed quote refused.
2026-10-02 16:53:27 +02:00
jochen 5eb1c9c2b7 The service manager is a node seat with the unit verbs, and a user shape names the account (hq ADR 0176, 0177)
node-service-manager joins the mesh's own seats, node-scoped, serving eight
verbs — units, status, start, stop, restart, enable, disable, journal — each
with a schema that takes an optional scope, "system" or the operator
account's "user" manager. Sixteen seats now; the count test says so and names
the record.

${machine:account} resolves in a resource's `name` and `user` as it already
did in path, owner and content: the shell module makes the operator's account
its holder's login shell through the `user` shape, and the desktop's watchers
run as that account through a user-scoped unit (host change alongside).
Neither can name the person. A machine with no account refuses by name.
2026-10-02 16:48:16 +02:00
mesh-admin 37b8edeec6 Merge pull request 'node show says a found firewall that was uninstalled is removed (hq ADR 0175)' (#217) from feat/the-found-front-end-is-uninstalled into main 2026-10-02 14:29:02 +00:00
jschoubben 424406b2d6 node show says a found firewall that was uninstalled is removed (hq ADR 0175) 2026-10-02 16:27:39 +02:00
mesh-admin 90455c61f6 Merge pull request 'The example images build from the Go the manifest pins' (#216) from jschoubben/example-images-go-base into main 2026-10-02 13:56:22 +00:00
jschoubben 1f6793cf0c The example images build from the Go the manifest pins
Four example Dockerfiles named golang:1.25 while go.mod requires 1.26;
the control plane's image takes GO_BASE from the manifest and these did
not, so a build that could not fetch a newer toolchain failed at go mod
download (found running the lab through the mesh).
2026-10-02 15:52:25 +02:00
mesh-admin 6ef522fbda Merge pull request 'Cite hq ADR 0170, not 0169: the firewall seat's record was renumbered' (#215) from fix/adr-0170-cited into main 2026-10-02 12:53:13 +00:00
jschoubben 46f65c12a0 Cite hq ADR 0170, not 0169: the firewall seat's record was renumbered after a collision on hq main 2026-10-02 14:52:24 +02:00
mesh-admin 8f7c02d77a Merge pull request 'The virtualisation capability grants the lab its daemon's socket (hq ADR 0172)' (#214) from jschoubben/the-lab-is-a-module into main 2026-10-02 12:47:45 +00:00
jschoubben a637df01ea The virtualisation capability grants the lab its daemon's socket
The lab raises machines on the virtualisation daemon, and a module may
mount a machine's socket only through the capability that grants it
(novox/hq ADR 0172). Also brings the resolver's tests to the setting
dnsmasq's listen addresses now come from, and to a module left out
rather than refused.
2026-10-02 14:46:17 +02:00
mesh-admin 252eb786a7 Merge pull request 'A filter module's own filter file counts as declared for a mount (hq ADR 0169)' (#213) from feat/the-firewall-seat-serves-its-verbs into main 2026-10-02 12:05:09 +00:00
jschoubben 9d13b0593b A filter module's own filter file counts as declared for a mount (hq ADR 0169)
The nftables module's runtime mounts the file filtering.into names, to reload
the mesh's table; the mount check knew every other declaration of a path and
not this one, and the module's first build was refused for it.
2026-10-02 14:04:35 +02:00
mesh-admin 30d548a762 Merge pull request 'The node-packet-filter seat serves rules, reload and remove (hq ADR 0169)' (#212) from feat/the-firewall-seat-serves-its-verbs into main 2026-10-02 11:28:57 +00:00
jschoubben 550e4c6acb The node-packet-filter seat serves rules, reload and remove (hq ADR 0169)
What a person asks a machine's packet filter whatever filter answers: the
rules as enforced, reload the mesh's own, remove one rule set the mesh did
not write — named as the host reports it under ADR 0168. Every holder serves
all three; a running mesh widens its seat row at the next controller start.
2026-10-02 13:27:04 +02:00
mesh-admin 1587fd97f9 Merge pull request 'The mesh says what filters a converged machine: filters kept per node, shown, named by status, previewed with fates (hq ADR 0168)' (#211) from feat/one-thing-filters-a-converged-machine into main 2026-10-02 10:03:01 +00:00
jschoubben b5df244096 The mesh says what filters a converged machine: filters kept per node, shown by node show, named by status, and previewed with their fates (hq ADR 0168)
A host reports every table and chain that refuses traffic with its owner,
and a converged machine's found firewall's state. The controller keeps both
on the node's record (migration 0054), shows them on node show, names every
converged machine something other than the mesh filters in status — text
and JSON, and such a machine is not well — and the converge preview lists
what filters the machine with the fate of each: retired with the front end,
left as the runtime's, left as a ban, or left in force and not the mesh's.
What was invisible for eleven hours (issues 144, 145) is said by name.
2026-10-02 12:00:12 +02:00
mesh-admin db47bb68e9 Merge pull request 'The controller's tools can set an assignment's settings (hq issue 198)' (#210) from jschoubben/settings-verb into main 2026-10-02 09:42:08 +00:00
jschoubben db84142d38 The controller's tools can set an assignment's settings
Per-machine and mesh-wide settings could be set only from the
controller's command line. The settings verb runs settings set|clear,
passing the values inline, which the command now accepts as well as a
file (novox/hq issue 198).
2026-10-02 11:41:57 +02:00
mesh-admin c9204591bf Merge pull request 'The hub relays the mesh passing through it (hq issue 196)' (#209) from jschoubben/the-hub-relays-the-mesh into main 2026-10-02 09:17:17 +00:00
jschoubben e8e707e013 The hub relays the mesh passing through it
The forward chain judged a packet relayed from one machine of the mesh
to another by this machine's own published ports, so two machines behind
the hub reached each other only on ports the hub published for itself
(novox/hq issue 196). In and out on the tunnel is now accepted, and the
machine it is for filters it.
2026-10-02 11:17:09 +02:00
mesh-admin 0c003774ba Merge pull request 'A take acts on the preview it showed; a setting is judged where it is stored; a kept network and a minted secret are said (hq ADR 0163)' (#206) from feat/a-take-is-a-comparison-the-rest into main 2026-10-02 09:04:05 +00:00
jschoubben fbc3d320ea A take acts on the preview it showed; a setting is judged where it is stored; a kept network and a minted secret are said (hq ADR 0163)
take ends its preview with a digest and --yes names it, as the flip does; a
changed preview or an account older than the flip allows is refused. A module
the machine holds nothing for has nothing to compare, and --yes suffices. A
published port's reach is said as the machine reported it. Every secret the
module holds on the machine is listed with where it came from, and one the
mesh minted for a service whose data was found refuses unless --mint names it.

One judgement of a module's settings against its definition, in the catalogue:
settings set refuses what cannot compose or reaches nothing, naming node,
module, layer and key; Compose leaves out a module whose definition moved
under a stored setting, the envelope says so (left_out), plan and push say it
by name, and the machine is told everything else. A stray setting no longer
refuses the whole machine where it is read (issue 096).

The per-machine setting networks keeps a found network for a taken container,
on an adopted machine only; the container's declaration carries it and the
preview names it (rule 4).
2026-10-02 11:01:09 +02:00
mesh-admin cf495e315f Merge pull request 'The proxy is told its routes and the mesh on the bus, and serves internal names to the mesh only (hq issue 191, ADR 0167)' (#207) from jschoubben/an-internal-only-route into main 2026-10-02 07:42:45 +00:00
mesh-admin 86bc1fad2c Merge pull request 'The controller's tools can issue a module its bus account (hq issue 191)' (#208) from jschoubben/the-controller-issues-a-module-account into main 2026-10-02 00:22:23 +00:00
jschoubben c9eba5f3b8 The controller's tools can issue a module its bus account
A module's account was mintable only from the controller's command line,
so a rollout that gave a module one could not be finished through the
mesh's own tools (novox/hq issue 191). The issue verb runs module issue
for a module on a machine; the caller pushes the machine after.
2026-10-02 01:55:06 +02:00
57 changed files with 3594 additions and 114 deletions
+4 -4
View File
@@ -59,7 +59,7 @@ PROVISIONER_IMAGE ?= mesh-provision-postgres:$(VERSION)
PROVISIONER_DEV_TAG ?= mesh-provision-postgres:development
provisioner-image:
docker build -f examples/postgres-provisioner/Dockerfile \
docker build --build-arg GO_BASE=$(GO_BASE) -f examples/postgres-provisioner/Dockerfile \
-t $(PROVISIONER_IMAGE) -t $(PROVISIONER_DEV_TAG) .
@echo
@docker image inspect $(PROVISIONER_IMAGE) --format 'built {{.RepoTags}} {{.Size}} bytes'
@@ -70,7 +70,7 @@ OBJECTSTORE_IMAGE ?= mesh-provision-objectstore:$(VERSION)
OBJECTSTORE_DEV_TAG ?= mesh-provision-objectstore:development
objectstore-image:
docker build -f examples/objectstore-provisioner/Dockerfile \
docker build --build-arg GO_BASE=$(GO_BASE) -f examples/objectstore-provisioner/Dockerfile \
-t $(OBJECTSTORE_IMAGE) -t $(OBJECTSTORE_DEV_TAG) .
@echo
@docker image inspect $(OBJECTSTORE_IMAGE) --format 'built {{.RepoTags}} {{.Size}} bytes'
@@ -81,7 +81,7 @@ REDIS_PROVISIONER_IMAGE ?= mesh-provision-redis:$(VERSION)
REDIS_PROVISIONER_DEV_TAG ?= mesh-provision-redis:development
redis-provisioner-image:
docker build -f examples/redis-provisioner/Dockerfile \
docker build --build-arg GO_BASE=$(GO_BASE) -f examples/redis-provisioner/Dockerfile \
-t $(REDIS_PROVISIONER_IMAGE) -t $(REDIS_PROVISIONER_DEV_TAG) .
@echo
@docker image inspect $(REDIS_PROVISIONER_IMAGE) --format 'built {{.RepoTags}} {{.Size}} bytes'
@@ -91,7 +91,7 @@ PROXY_IMAGE ?= mesh-route-proxy:$(VERSION)
PROXY_DEV_TAG ?= mesh-route-proxy:development
proxy-image:
docker build -f examples/route-proxy/Dockerfile -t $(PROXY_IMAGE) -t $(PROXY_DEV_TAG) .
docker build --build-arg GO_BASE=$(GO_BASE) -f examples/route-proxy/Dockerfile -t $(PROXY_IMAGE) -t $(PROXY_DEV_TAG) .
@echo
@docker image inspect $(PROXY_IMAGE) --format 'built {{.RepoTags}} {{.Size}} bytes'
+104 -1
View File
@@ -95,11 +95,22 @@ func reportsReaching(t *testing.T, open *stores, reachable []link.Reach, held ..
// filter is not sent to one that has not. The anchor reports one, as a real host does; this
// fixture lacked it from 2026-09-28 and nothing ran the test (issue 177).
Outward: []string{"eth0"},
// And what filters it (ADR 0168): its front end, the runtime's own, and a chain a
// predecessor left in the runtime's user chain.
Filters: anchorFilters,
}); err != nil {
t.Fatal(err)
}
}
// anchorFilters is what the adopted anchor says filters it: ufw's chains, the runtime's, and a
// predecessor's chain the mesh did not write.
var anchorFilters = []link.Filter{
{Where: "table ip filter, chain ufw-reject-input", Owner: "found-firewall", Refuses: "reject"},
{Where: "table ip filter, chain DOCKER", Owner: "runtime", Refuses: `iifname != "docker0" oifname "docker0" drop`},
{Where: "table ip filter, chain DOCKER-USER", Owner: "other", Refuses: `iifname "eth0" tcp dport 6000 drop`},
}
var (
heldContainer = link.Held{ID: "hello-web.server", Module: "hello-web", Kind: "container",
Target: "hello-web", Since: time.Now()}
@@ -143,7 +154,24 @@ func TestTheFlipIsRefusedWhileAFoundContainerIsHeld(t *testing.T) {
func TestTakingNamesWhatItReplaces(t *testing.T) {
open, _ := anAdoptedAnchor(t)
reportsHolding(t, open, heldContainer, heldFile)
said, err := take(t.Context(), open, "anchor", "hello-web", takeOptions{Yes: true})
ctx := t.Context()
// The machine holds something for the module, so the take acts on the preview the operator
// saw and names its digest (novox/hq ADR 0163).
preview, err := take(ctx, open, "anchor", "hello-web", takeOptions{})
if err != nil {
t.Fatal(err)
}
saw := takeDigestIn(t, preview)
if !strings.Contains(preview, "nothing taken; `take anchor hello-web --yes "+saw+"`") {
t.Fatalf("the preview does not say how to act on it:\n%s", preview)
}
if taken, _ := open.inventory.Taken(ctx, "anchor"); len(taken) != 0 {
t.Fatal("the preview took something")
}
if _, err := take(ctx, open, "anchor", "hello-web", takeOptions{Yes: true}); err == nil || !strings.Contains(err.Error(), "name its digest") {
t.Fatalf("--yes without the digest was not refused: %v", err)
}
said, err := take(ctx, open, "anchor", "hello-web", takeOptions{Yes: true, Digest: saw})
if err != nil {
t.Fatal(err)
}
@@ -153,6 +181,67 @@ func TestTakingNamesWhatItReplaces(t *testing.T) {
}
}
// takeDigestIn is the digest a take's preview printed.
func takeDigestIn(t *testing.T, preview string) string {
t.Helper()
for _, line := range strings.Split(preview, "\n") {
if fields := strings.Fields(line); len(fields) == 2 && fields[0] == "preview" {
return fields[1]
}
}
t.Fatalf("the preview printed no digest:\n%s", preview)
return ""
}
// A take acts on the preview the operator saw, and on an account of the machine that is still the
// machine: a changed preview and a stale account refuse (novox/hq ADR 0163, rule 1).
func TestATakeIsRefusedOnAChangedPreviewOrAStaleAccount(t *testing.T) {
open, _ := anAdoptedAnchor(t)
ctx := t.Context()
reportsHolding(t, open, heldContainer, heldFile)
preview, err := take(ctx, open, "anchor", "hello-web", takeOptions{})
if err != nil {
t.Fatal(err)
}
saw := takeDigestIn(t, preview)
// The machine reports again, and what it holds has changed: the found container now carries
// facts the preview never showed.
changed := heldContainer
changed.Facts = map[string]any{"image": "hello:2", "declared_image": "registry.example/hello"}
reportsHolding(t, open, changed, heldFile)
_, err = take(ctx, open, "anchor", "hello-web", takeOptions{Yes: true, Digest: saw})
if err == nil || !strings.Contains(err.Error(), "has changed since preview "+saw) {
t.Fatalf("a changed preview was acted on: %v", err)
}
if taken, _ := open.inventory.Taken(ctx, "anchor"); len(taken) != 0 {
t.Fatal("a refused take took something")
}
// And an account older than the flip allows.
preview, err = take(ctx, open, "anchor", "hello-web", takeOptions{})
if err != nil {
t.Fatal(err)
}
saw = takeDigestIn(t, preview)
saved := reportFreshFor
reportFreshFor = -time.Second
defer func() { reportFreshFor = saved }()
_, err = take(ctx, open, "anchor", "hello-web", takeOptions{Yes: true, Digest: saw})
if err == nil || !strings.Contains(err.Error(), "a take acts only on an account newer than") {
t.Fatalf("a stale account was acted on: %v", err)
}
reportFreshFor = saved
if _, err := take(ctx, open, "anchor", "hello-web", takeOptions{Yes: true, Digest: saw}); err != nil {
t.Fatal(err)
}
// A module the machine holds nothing for has nothing to compare: --yes alone suffices.
if _, err := take(ctx, open, "anchor", "notes", takeOptions{Yes: true}); err == nil {
// notes holds a file, so this one needs the digest too.
t.Fatal("notes holds a found file and was taken without a digest")
}
}
func TestConvergingPreviewsThenChangesAndAdoptingKeepsWhatWasTaken(t *testing.T) {
open, sent := anAdoptedAnchor(t)
ctx := t.Context()
@@ -186,6 +275,20 @@ func TestConvergingPreviewsThenChangesAndAdoptingKeepsWhatWasTaken(t *testing.T)
if strings.Contains(preview, "15672") {
t.Errorf("a loopback listener is in the preview:\n%s", preview)
}
// What filters the machine now, and the fate of each (novox/hq ADR 0168): the predecessor's
// chain is named as not the mesh's and left, so the reader knows before the flip.
for _, want := range []string{
"table ip filter, chain DOCKER-USER",
"NOT THE MESH'S; left in force",
`iifname "eth0" tcp dport 6000 drop`,
"table ip filter, chain ufw-reject-input",
"the found firewall's; retired with it",
"the container runtime's own; left",
} {
if !strings.Contains(preview, want) {
t.Errorf("the preview does not say %q:\n%s", want, preview)
}
}
for _, line := range strings.Split(preview, "\n") {
if strings.Contains(line, "5000") && !strings.Contains(line, "WILL CLOSE") {
t.Errorf("an undeclared published port is not said to close: %s", line)
+299 -23
View File
@@ -29,6 +29,10 @@ func showMode(ctx context.Context, inv *inventory.Inventory, node inventory.Node
if said, err := inv.AdoptionOf(ctx, node.Name); err == nil && len(said.Strays) > 0 {
showStrays(said.Strays)
}
// And what filters it, truthfully (novox/hq ADR 0168): the mesh alone, or not.
if filtering, err := inv.FilteringOf(ctx, node.Name); err == nil {
showFiltering(filtering, false)
}
return nil
}
fmt.Printf(" mode adopted since %s\n",
@@ -72,10 +76,65 @@ func showMode(ctx context.Context, inv *inventory.Inventory, node inventory.Node
}
}
showStrays(said.Strays)
if filtering, err := inv.FilteringOf(ctx, node.Name); err == nil {
showFiltering(filtering, true)
}
fmt.Printf(" as of %s\n", said.At.Local().Format(time.DateTime))
return nil
}
// showFiltering says what filters a machine, with owners (novox/hq ADR 0168), and for a converged
// machine the state of the firewall it was found with. A machine that has not said is not said to
// be filtered by anything.
func showFiltering(f inventory.Filtering, adopted bool) {
if len(f.Filters) == 0 && f.FoundFirewall == nil {
return
}
if fw := f.FoundFirewall; fw != nil && !adopted {
switch {
case fw.Active:
fmt.Printf(" found firewall %s is ACTIVE on this converged machine; the next apply retires it again\n", fw.Kind)
case fw.RetiredBy == "removed":
fmt.Printf(" found firewall %s, removed: the mesh's filter is what filters this machine (novox/hq ADR 0180)\n", fw.Kind)
case fw.RetiredBy == inventory.FilterMesh || fw.RetiredBy == "mesh":
fmt.Printf(" found firewall %s, retired by the mesh; its configuration stays on disk\n", fw.Kind)
case fw.RetiredBy != "":
fmt.Printf(" found firewall %s, found inactive — not by the mesh\n", fw.Kind)
default:
fmt.Printf(" found firewall %s, inactive\n", fw.Kind)
}
}
if len(f.Filters) == 0 {
return
}
if f.Alone() {
fmt.Printf(" filtered by the mesh alone (%s)\n", filterSummary(f.Filters))
return
}
fmt.Printf(" filtered by NOT the mesh alone: %d rule set(s) the mesh did not write refuse traffic here\n", len(f.Others()))
for _, x := range f.Filters {
if x.Owner == inventory.FilterOther || x.Owner == inventory.FilterFoundFirewall {
fmt.Printf(" %-17s %s — %s: %s\n", "", x.Where, x.Owner, x.Refuses)
}
}
fmt.Printf(" %-17s and its own: %s\n", "", filterSummary(f.Filters))
}
// filterSummary counts a machine's filters by owner: "mesh 2, runtime 3, ban 1".
func filterSummary(filters []inventory.Filter) string {
counts := map[string]int{}
for _, x := range filters {
counts[x.Owner]++
}
var parts []string
for _, owner := range []string{inventory.FilterMesh, inventory.FilterRuntime, inventory.FilterBan, inventory.FilterFoundFirewall, inventory.FilterOther} {
if n := counts[owner]; n > 0 {
parts = append(parts, fmt.Sprintf("%s %d", owner, n))
}
}
return strings.Join(parts, ", ")
}
// showStrays says what a machine runs that the mesh neither wrote nor holds (ADR 0163).
func showStrays(strays []inventory.Stray) {
if len(strays) == 0 {
@@ -156,11 +215,25 @@ const DefaultFilter = "nftables"
// take is a module's cutover on an adopted node: the operator's act, done when that module's data
// has moved. From the next push its resources converge there like any other, replacing what the
// node found and holds for it.
//
// **Previewed, and the preview is a comparison** (novox/hq ADR 0163): for every held thing the
// module would replace, what runs beside what the module declares, and the difference; the
// module's secrets on the machine and where each came from; its settings on the machine. Without
// --yes the comparison is printed and nothing changes. `--yes <digest>` cuts over exactly what was
// previewed, the way the flip is confirmed: the preview ends with a digest of what it said, and a
// take naming an older one, or acting on an account of the machine older than the flip allows, is
// refused. A module the machine holds nothing for has nothing to compare, and `--yes` suffices.
// takeOptions is what a take was told about the differences it may pass (novox/hq ADR 0163).
type takeOptions struct {
Yes bool
Yes bool
// Digest is the preview's, named with --yes; required whenever the machine holds something
// for the module.
Digest string
Downgrade bool
Replace map[string]bool
// Mint names the secrets the service shall take a new value for, although the mesh minted
// one and the service already has its own (rule 2).
Mint map[string]bool
}
func take(ctx context.Context, open *stores, node, module string, opts takeOptions) (string, error) {
@@ -179,45 +252,128 @@ func take(ctx context.Context, open *stores, node, module string, opts takeOptio
}
// The comparison first (novox/hq ADR 0163): every held thing the module would replace, beside
// what the module declares, and the differences that refuse unless named.
reported, err := inv.AdoptionOf(ctx, node)
c, err := comparisonFor(ctx, open, node, module)
if err != nil {
return "", err
}
preview, refusals := comparisonOf(reported.Held, module, opts)
preview, refusals, saw := comparisonOf(module, c, opts)
if len(refusals) > 0 {
return "", fmt.Errorf("taking %s on %s is refused:\n %s\n%s", module, node,
strings.Join(refusals, "\n "), preview)
}
holds := len(heldOf(c.reported, module)) > 0
if holds {
preview += "\n preview " + saw
}
if !opts.Yes {
return preview + fmt.Sprintf("\nnothing taken; `take %s %s --yes` cuts it over as previewed", node, module), nil
if !holds {
return preview + fmt.Sprintf("\nnothing taken; `take %s %s --yes` declares it as the mesh's own", node, module), nil
}
return preview + fmt.Sprintf("\nnothing taken; `take %s %s --yes %s` cuts it over as previewed", node, module, saw), nil
}
if holds {
// The take acts on the preview the operator saw, and on an account of the machine that
// is still the machine: the same two refusals the flip makes.
if age := time.Since(c.reported.At); age > reportFreshFor {
return preview, fmt.Errorf("%s last said what it holds %s ago, and a take acts only on "+
"an account newer than %s: run `push %s --wait 2m`, then preview again",
node, age.Round(time.Second), reportFreshFor, node)
}
if opts.Digest == "" {
return preview, fmt.Errorf("taking %s on %s acts on the preview you saw: name its digest, "+
"`take %s %s --yes %s`, once you have read it", module, node, node, module, saw)
}
if opts.Digest != saw {
return preview, fmt.Errorf("what taking %s on %s would replace has changed since preview %s "+
"(it is now %s): read the preview above, and run `take %s %s --yes %s` if it is "+
"what you want", module, node, opts.Digest, saw, node, module, saw)
}
}
if err := inv.Take(ctx, node, module); err != nil {
return "", err
}
said := fmt.Sprintf("%s is taken on %s", module, node)
if preview != "" {
if holds {
said += "; the next push replaces what the node found and holds for it:\n" + preview
}
return said + fmt.Sprintf("\n run `push %s` to cut it over", node), nil
}
// comparison is everything a take puts beside what the module declares: the machine's account of
// what it holds and what is reachable on it, the module's secrets on the machine, its settings
// there, and which found networks a setting keeps for each of its containers (by held id).
type comparison struct {
reported inventory.Adoption
secrets []inventory.SecretState
layers []catalogue.Layer
keeps map[string][]string
// settingsRefused is why the module's settings cannot compose with its definition, when
// they cannot — the module would be left out of the declaration (rule 6).
settingsRefused string
}
func comparisonFor(ctx context.Context, open *stores, node, module string) (comparison, error) {
inv := open.inventory
var c comparison
var err error
if c.reported, err = inv.AdoptionOf(ctx, node); err != nil {
return c, err
}
if c.secrets, err = inv.SecretsOf(ctx, node, module); err != nil {
return c, err
}
if c.layers, err = inv.SettingsFor(ctx, node, module); err != nil {
return c, err
}
shelf, err := inv.Catalogue(ctx)
if err != nil {
return c, err
}
if m, known := shelf[module]; known && len(c.layers) > 0 {
if err := catalogue.JudgeSettings(m, c.layers, true); err != nil {
c.settingsRefused = err.Error()
}
if kept, err := catalogue.KeptNetworks(m, c.layers, true); err == nil && len(kept) > 0 {
c.keeps = map[string][]string{}
for id, networks := range kept {
c.keeps[module+"."+id] = networks
}
}
}
return c, nil
}
// heldOf is what a node holds for one module.
func heldOf(reported inventory.Adoption, module string) []inventory.Held {
var out []inventory.Held
for _, h := range reported.Held {
if h.Module == module {
out = append(out, h)
}
}
return out
}
// comparisonOf is a take's preview: for every held thing of the module, what runs beside what the
// module declares, and the refusals the differences earn unless the take named them
// (novox/hq ADR 0163): an image older than the one running, a declared file that differs from the
// found one. A narrowed port and a shared network are said and not refused.
func comparisonOf(held []inventory.Held, module string, opts takeOptions) (string, []string) {
// module declares; its secrets and its settings on the machine; and the refusals the differences
// earn unless the take named them (novox/hq ADR 0163): an image older than the one running, a
// declared file that differs from the found one, a secret the mesh minted for a service whose data
// was found. A narrowed port and a shared network are said and not refused. The digest is of what
// the preview says, so anything in it changing changes the digest.
func comparisonOf(module string, c comparison, opts takeOptions) (preview string, refusals []string, digest string) {
var b strings.Builder
var refusals []string
held := heldOf(c.reported, module)
foundData := false
for _, h := range held {
if h.Module != module {
continue
if h.Kind == "container" || h.Kind == "directory" {
foundData = true
}
fmt.Fprintf(&b, " %s", heldLine(h))
if h.Kept != "" {
fmt.Fprintf(&b, ", original kept at %s", h.Kept)
}
b.WriteString("\n")
for _, line := range comparisonLines(h) {
for _, line := range comparisonLinesWith(h, c.keeps[h.ID], c.reported) {
fmt.Fprintf(&b, " %s\n", line)
}
f := factsOf(h)
@@ -232,7 +388,52 @@ func comparisonOf(held []inventory.Held, module string, opts takeOptions) (strin
h.Target, h.Target))
}
}
return b.String(), refusals
// The module's secrets on the machine (rule 2 and 3): a service whose data was found already
// has a value for each, so one the mesh minted and nobody accepted refuses unless --mint says
// the service shall take a new one.
for _, sec := range c.secrets {
name := sec.Name
if sec.Local != "" {
name += " (" + sec.Local + ")"
}
what := "own secret"
accept := fmt.Sprintf("`secret accept <node> %s %s`", module, sec.Name)
if !sec.Own() {
what = "secret from " + sec.Provider
accept = fmt.Sprintf("`secret accept <node> %s %s --provider %s`", module, sec.Name, sec.Provider)
if sec.Local != "" {
accept = strings.TrimSuffix(accept, "`") + " --local " + sec.Local + "`"
}
}
switch {
case sec.Origin == inventory.OriginAccepted:
fmt.Fprintf(&b, " %s %s: accepted from a person, carried in as it is\n", what, name)
case opts.Mint[sec.Name]:
fmt.Fprintf(&b, " %s %s: minted by the mesh; the service takes the new value, as --mint said\n", what, name)
case foundData:
fmt.Fprintf(&b, " %s %s: MINTED by the mesh and not accepted — the running service already has one\n", what, name)
refusals = append(refusals, fmt.Sprintf("%s: the mesh minted a value and the service whose data was found "+
"already uses its own; %s carries the existing value in, or `--mint %s` says the service shall take "+
"the new one", name, accept, sec.Name))
default:
fmt.Fprintf(&b, " %s %s: minted by the mesh\n", what, name)
}
}
// And its settings on this machine, composed against its definition (rule 1, rule 6).
for _, layer := range c.layers {
keys := make([]string, 0, len(layer.Values))
for k := range layer.Values {
keys = append(keys, k)
}
sort.Strings(keys)
fmt.Fprintf(&b, " settings from %s: %s\n", layer.From, strings.Join(keys, ", "))
}
if c.settingsRefused != "" {
fmt.Fprintf(&b, " SETTINGS DO NOT COMPOSE with the module's definition, so the push leaves it out: %s\n", c.settingsRefused)
}
preview = strings.TrimRight(b.String(), "\n")
sum := sha256.Sum256([]byte(preview))
return preview, refusals, hex.EncodeToString(sum[:])[:12]
}
// facts is a held thing's facts as the preview reads them.
@@ -286,6 +487,13 @@ func factsOf(h inventory.Held) facts {
// comparisonLines says a held thing's facts the way a person weighs them.
func comparisonLines(h inventory.Held) []string {
return comparisonLinesWith(h, nil, inventory.Adoption{})
}
// comparisonLinesWith is comparisonLines knowing which found networks this machine's setting keeps
// for the container (rule 4) and what the machine reports reachable, so a published port's reach
// is said beside the port (rule 1).
func comparisonLinesWith(h inventory.Held, keeps []string, reported inventory.Adoption) []string {
f := factsOf(h)
var out []string
if f.image != "" || f.declaredImage != "" {
@@ -311,14 +519,46 @@ func comparisonLines(h inventory.Held) []string {
}
sort.Strings(names)
for _, n := range names {
if members := f.networks[n]; len(members) > 0 {
out = append(out, fmt.Sprintf("on the network %s with %s, which may reach it by name and will not once it moves to the module's own network",
members := f.networks[n]
if len(members) == 0 {
continue
}
if slices.Contains(keeps, n) {
out = append(out, fmt.Sprintf("on the network %s with %s — kept by this machine's setting, so they still reach it by name once taken",
n, strings.Join(members, ", ")))
continue
}
out = append(out, fmt.Sprintf("on the network %s with %s, which may reach it by name and will not once it moves to the module's own network"+
" (`settings set %s --node <node>` with {%q: {<container>: [%q]}} keeps it)",
n, strings.Join(members, ", "), h.Module, catalogue.NetworksSetting, n))
}
for _, n := range keeps {
if _, found := f.networks[n]; !found {
out = append(out, fmt.Sprintf("keeps the network %s by this machine's setting, which the found container is not on", n))
}
}
if len(f.ports) > 0 || len(f.declaredPorts) > 0 {
out = append(out, fmt.Sprintf("publishes %s; the module declares %s",
orNone(strings.Join(f.ports, " ")), orNone(strings.Join(f.declaredPorts, " "))))
// How far each published port reaches now, as the machine reported it: the listener the
// runtime publishes for this container. The found firewall's and the guard's rules are
// not read; what they let through is said as what was reported reachable.
var reach []string
for _, r := range reported.Reachable {
if r.By == h.Target && r.Published {
reach = append(reach, fmt.Sprintf("%s:%d (%s, container port %d)", r.Address, r.Port, r.Protocol, r.ContainerPort))
}
}
switch {
case len(reach) > 0:
line := "reachable now at " + strings.Join(reach, ", ")
if reported.Firewall != "" && reported.Firewall != "none" {
line += ", behind the found firewall (" + reported.Firewall + "), whose rules are not read"
}
out = append(out, line)
case len(f.ports) > 0 && len(reported.Reachable) > 0:
out = append(out, "not reported reachable on the machine")
}
}
if len(f.mounts) > 0 || len(f.declaredVolumes) > 0 {
out = append(out, fmt.Sprintf("mounts %s; the module declares %s",
@@ -480,7 +720,11 @@ func converge(ctx context.Context, open *stores, node string, yes bool, digest s
}
derived := derivedFilter{rules: rules, foundation: with.Foundation, mesh: with.Mesh,
outward: plan.PublicDomain != "", outwardLinks: with.OutwardLinks}
preview, saw := previewOf(node, reported, derived, plan, taken, filter, runs[filter])
filtering, err := inv.FilteringOf(ctx, node)
if err != nil {
return "", err
}
preview, saw := previewOf(node, reported, filtering, derived, plan, taken, filter, runs[filter])
preview += "\n\n preview " + saw
if !yes {
return preview + fmt.Sprintf("\n\nNothing has changed. Run `converge %s --yes %s` to do "+
@@ -550,7 +794,7 @@ func converge(ctx context.Context, open *stores, node string, yes bool, digest s
// previewOf is what converging a node will change, before it changes it, and a short digest of
// what it said: every reachable thing and its fate, the modules the flip takes and the filter. The
// digest is what the flip is asked to act on, so it changes whenever any of those would.
func previewOf(node string, reported inventory.Adoption, derived derivedFilter,
func previewOf(node string, reported inventory.Adoption, filtering inventory.Filtering, derived derivedFilter,
plan catalogue.Resolution, taken []string, filter string, filterAssigned bool) (string, string) {
var said []string
var b strings.Builder
@@ -642,6 +886,30 @@ func previewOf(node string, reported inventory.Adoption, derived derivedFilter,
fmt.Fprintf(&b, " the found firewall (%s) is disabled, never flushed: its configuration stays on disk\n", fw)
}
said = append(said, fmt.Sprintf("filter %s assigned=%t firewall=%s", filter, filterAssigned, fw))
// What filters the machine now, and the fate of each (novox/hq ADR 0168): the found firewall
// retired, the runtime's own and bans left, and what the mesh did not write left and named —
// so the reader knows before the flip that the machine will not be filtered by the mesh alone.
if len(filtering.Filters) > 0 {
b.WriteString("\n what filters the machine now, and what the flip does to each:\n")
for _, x := range filtering.Filters {
fate := "left: " + x.Owner + "'s"
switch x.Owner {
case inventory.FilterMesh:
fate = "the mesh's guard; replaced by its filter"
case inventory.FilterFoundFirewall:
fate = "the found firewall's; retired with it"
case inventory.FilterRuntime:
fate = "the container runtime's own; left"
case inventory.FilterBan:
fate = "a ban list; left"
case inventory.FilterOther:
fate = "NOT THE MESH'S; left in force — the machine is not filtered by the mesh alone until you remove it"
}
fmt.Fprintf(&b, " %-50s %s\n", x.Where, fate)
fmt.Fprintf(&b, " %-50s %s\n", "", x.Refuses)
said = append(said, "filter "+x.Owner+" "+x.Where)
}
}
// Sorted: the same account, reported in another order, is the same preview.
sort.Strings(said)
sum := sha256.Sum256([]byte(strings.Join(said, "\n")))
@@ -767,21 +1035,29 @@ func adopt(ctx context.Context, open *stores, node string) (string, error) {
// takeCommand, convergeCommand and adoptCommand are the command line's adapters to the acts above.
func takeCommand(ctx context.Context, args []string) error {
set := flag.NewFlagSet("take", flag.ContinueOnError)
yes := set.Bool("yes", false, "cut over as previewed; without it the comparison is printed and nothing is taken")
yes := set.Bool("yes", false, "cut over as previewed, naming the digest the preview printed after it; "+
"without it the comparison is printed and nothing is taken")
downgrade := set.Bool("downgrade", false, "take it although the module's image is older than the one running")
var replace stringList
var replace, mint stringList
set.Var(&replace, "replace", "a found file's path whose content the module may replace although it differs (repeatable; * for every one)")
set.Var(&mint, "mint", "a secret the service shall take the mesh's minted value for, although it already has its own (repeatable)")
positionals, err := parseAround(set, args)
if err != nil {
return err
}
if len(positionals) != 2 {
return errors.New("take <node> <module> [--yes] [--downgrade] [--replace <path>]...")
if len(positionals) < 2 || len(positionals) > 3 || (len(positionals) == 3 && !*yes) {
return errors.New("take <node> <module> [--yes <digest>] [--downgrade] [--replace <path>]... [--mint <secret>]...")
}
opts := takeOptions{Yes: *yes, Downgrade: *downgrade, Replace: map[string]bool{}, Mint: map[string]bool{}}
if len(positionals) == 3 {
opts.Digest = positionals[2]
}
opts := takeOptions{Yes: *yes, Downgrade: *downgrade, Replace: map[string]bool{}}
for _, r := range replace {
opts.Replace[r] = true
}
for _, m := range mint {
opts.Mint[m] = true
}
return runAct(ctx, func(open *stores) (string, error) { return take(ctx, open, positionals[0], positionals[1], opts) })
}
+3 -3
View File
@@ -102,7 +102,7 @@ func commands(who Authenticator) http.Handler {
}))
// Adoption (novox/hq ADR 0100): the same acts as `take`, `converge` and `adopt`.
mux.HandleFunc("POST /take", acting(who, true, func(ctx context.Context, open *stores, in request) (string, error) {
return take(ctx, open, in.Node, in.Module, takeOptions{Yes: true})
return take(ctx, open, in.Node, in.Module, takeOptions{Yes: in.Yes, Digest: in.Digest})
}))
mux.HandleFunc("POST /converge", acting(who, false, func(ctx context.Context, open *stores, in request) (string, error) {
return converge(ctx, open, in.Node, in.Yes, in.Digest, in.Filter)
@@ -124,8 +124,8 @@ func commands(who Authenticator) http.Handler {
type request struct {
Node string `json:"node"`
Module string `json:"module"`
// Yes, Digest and Filter are converge's: do it rather than preview it, the digest of the
// preview it acts on, and which module loads the mesh's filter.
// Yes, Digest and Filter are converge's and take's: do it rather than preview it, the digest
// of the preview it acts on, and (converge) which module loads the mesh's filter.
Yes bool `json:"yes,omitempty"`
Digest string `json:"digest,omitempty"`
Filter string `json:"filter,omitempty"`
+8
View File
@@ -533,6 +533,10 @@ func takeIn(ctx context.Context, inv *inventory.Inventory, result link.BuildResu
if err := inv.RegisterModule(ctx, manifest, recorded); err != nil {
return manifest, kept, err
}
// The keep set just moved, and new bytes just landed (novox/hq ADR 0189). Asked here rather
// than on a timer of its own: this is the only moment either is true. Never fatal — the build
// worked and the module is registered.
collect(ctx, inv)
return manifest, kept, nil
}
@@ -607,6 +611,10 @@ type answers struct {
// a consequence of the refusals above: a node that does not resolve is not on the network, and
// a mesh whose hub is that node has no hub.
network string
// filtered is every converged machine that is not filtered by the mesh alone (novox/hq ADR
// 0168): what filters it beyond the mesh's own, the runtime's plumbing and bans, by name — a
// predecessor's chain, a found firewall in force again. Such a machine is not "all well".
filtered map[string]inventory.Filtering
// untaken is, per machine, each assigned module whose resources the machine is holding as it
// found them, and how many — a module that was assigned, sent, and is running none of what it
// declares because nothing has taken it (novox/hq ADR 0100, 04-ISSUES/125).
+84
View File
@@ -0,0 +1,84 @@
package main
import (
"context"
"errors"
"fmt"
"os"
"github.com/novox/mesh-controller/internal/artifacts"
"github.com/novox/mesh-controller/internal/inventory"
)
// Letting the artifact store go of what the mesh no longer keeps (novox/hq ADR 0189, issue 108).
//
// **Run where the records change.** A build is the moment new bytes landed in the store and the
// moment the keep set moved, so it is the moment to say what may go — and it needs no timer of
// its own. Reclaiming the bytes is the store's own nightly step; this only decides.
//
// Never fatal to a build. The build succeeded, the module is registered, and a store that could
// not be reached is a thing to say rather than a reason to undo any of that. The next build asks
// again, and the references it could not collect are still uncollected, so nothing is lost by
// having failed.
// collect asks the store to let go of everything the mesh made and no longer keeps, and records
// what it let go of. Says what it did and what it could not; returns nothing, because nothing
// upstream should branch on it.
func collect(ctx context.Context, inv *inventory.Inventory) {
references, err := inv.ToCollect(ctx)
if err != nil {
fmt.Fprintf(os.Stderr, "could not work out what the artifact store may let go of: %v\n", err)
return
}
if len(references) == 0 {
return
}
shelf, err := inv.Catalogue(ctx)
if err != nil {
fmt.Fprintf(os.Stderr, "could not read the catalogue to find the artifact store: %v\n", err)
return
}
// As the mesh reaches it from the network. Empty means the store is not on the network — on a
// mesh being raised it is not yet, and there the store holds one build of anything and has
// nothing to collect.
address, err := artifactStoreAddress(ctx, inv, shelf, "")
if err != nil || address == "" {
if err != nil {
fmt.Fprintf(os.Stderr, "could not find the artifact store to collect from: %v\n", err)
}
return
}
store := artifacts.Store{Address: address}
var done []string
var refused int
for _, reference := range references {
switch err := store.LetGo(ctx, reference); {
case err == nil, errors.Is(err, artifacts.Gone):
// Gone is the outcome wanted, already true. Recorded so the next sweep does not ask
// again for ever.
done = append(done, reference)
default:
refused++
if refused == 1 {
// Once per sweep. A store that refuses one refuses all of them, and a hundred
// identical lines would bury the reason.
fmt.Fprintf(os.Stderr, "the artifact store kept %s: %v\n", reference, err)
}
}
}
if len(done) > 0 {
if err := inv.MarkCollected(ctx, done); err != nil {
// Said, and that is all: the artifacts are gone either way, and the only cost of an
// unrecorded collection is that the next sweep asks about them again.
fmt.Fprintf(os.Stderr, "the store let go of %d artifact(s) and the record of it did not keep: %v\n",
len(done), err)
return
}
fmt.Fprintf(os.Stderr, "the artifact store let go of %d artifact(s) the mesh no longer keeps\n",
len(done))
}
if refused > 0 {
fmt.Fprintf(os.Stderr, "%d artifact(s) were not collected; the next build asks again\n", refused)
}
}
+2 -1
View File
@@ -180,7 +180,8 @@ func usage() {
api --issuer URL [--listen A] assign and unassign over http, for a surface that is not here
assign <node> <module> put a module on a node
unassign <node> <module> take it off
take <node> <module> cut a module over on an adopted node, once its data has moved
take <node> <module> preview a module's cutover on an adopted node: what runs beside
what it declares; --yes <digest> cuts it over as previewed
converge <node> [--yes <digest>] [--filter nftables] preview, then make, an adopted node converged
adopt <node> return a converged node to adopted; what was taken stays taken
settings set <module> <file> what a module's config should say, for the whole mesh
+7 -3
View File
@@ -352,10 +352,14 @@ func settingsCommand(ctx context.Context, args []string) error {
switch args[0] {
case "set":
if len(positionals) != 2 {
return errors.New("settings set <module> <settings.json> [--node <node>]")
return errors.New("settings set <module> <settings.json | {…}> [--node <node>]")
}
raw, err := os.ReadFile(positionals[1])
if err != nil {
// A file, or the values themselves when they begin with `{` — which is how the mesh's own
// `settings` tool passes them, having no file to hand over (novox/hq issue 198).
var raw []byte
if strings.HasPrefix(strings.TrimSpace(positionals[1]), "{") {
raw = []byte(positionals[1])
} else if raw, err = os.ReadFile(positionals[1]); err != nil {
return err
}
var values map[string]any
+49 -15
View File
@@ -186,8 +186,12 @@ func planFor(ctx context.Context, open *stores, nodeName string) (catalogue.Reso
// Settings for everything that resolved, including modules nobody assigned directly: a
// requirement pulled in by something else is still configurable, and finding out that it is
// not only when you try would be an arbitrary line nobody could predict.
//
// A setting that reaches nothing, or cannot compose with the definition it was stored for,
// no longer refuses the machine here: it is judged where it is stored, and a definition that
// moved under it costs that module its place in the declaration, said by name (novox/hq ADR
// 0163, rule 6 — see Compose).
settings := catalogue.SettingsBy{}
var stray []string
for _, m := range resolved.Modules {
layers, err := inv.SettingsFor(ctx, nodeName, m.Module)
if err != nil {
@@ -197,18 +201,6 @@ func planFor(ctx context.Context, open *stores, nodeName string) (catalogue.Reso
continue
}
settings[m.Module] = layers
stray = append(stray, catalogue.UnusedSettings(m, layers)...)
}
if len(stray) > 0 {
// Somebody set something that reaches no file. Said here rather than discovered by the
// machine not behaving differently, which is the slowest way there is.
//
// Marked like a set that will not compose, and for the same reason: it is a standing fact
// about this node's own configuration, not a question the mesh could not answer. A gatherer
// passes over it as it always did — one node's stray setting must not stop every other node
// being described (novox/hq 04-ISSUES/152).
return catalogue.Resolution{}, nil, notResolvable{fmt.Errorf(
"these settings reach nothing:\n - %s", strings.Join(stray, "\n - "))}
}
return resolved, settings, nil
}
@@ -405,7 +397,32 @@ func declarationWith(ctx context.Context, open *stores, node string,
return sendable{}, err
}
return sendable{Resources: composed.Resources, Adoption: adoption,
Received: composed.Received, Mesh: with.Mesh}, nil
Received: composed.Received, Mesh: with.Mesh,
LeftOut: sortedKeysOf(composed.LeftOut), leftOutWhy: composed.LeftOut}, nil
}
// sortedKeysOf is a map's keys, sorted — so what a declaration says it left out does not move
// for a reordering nobody made.
func sortedKeysOf(m map[string]string) []string {
if len(m) == 0 {
return nil
}
out := make([]string, 0, len(m))
for k := range m {
out = append(out, k)
}
sort.Strings(out)
return out
}
// reportLeftOut says which of a machine's modules its declaration leaves out and why (novox/hq ADR
// 0163, rule 6), one line each: the machine is told everything else, and is told it was left out.
func reportLeftOut(node string, declared sendable) {
for _, m := range declared.LeftOut {
fmt.Printf("%s: %s left out — a setting stored for it cannot compose with its definition; "+
"what the machine holds for it is kept and its containers are untouched. %s\n",
node, m, declared.leftOutWhy[m])
}
}
// renderingFor is everything a node's declaration is composed with, and the node's record.
@@ -445,7 +462,10 @@ func renderingFor(ctx context.Context, open *stores, node string,
for _, m := range plan.Modules {
g, err := catalogue.GivenPorts(m, settings[m.Module])
if err != nil {
return catalogue.Rendering{}, inventory.Node{}, err
// A given port its definition no longer publishes: the module is left out of the
// declaration, by name, when it is composed (novox/hq ADR 0163, rule 6) — never the
// machine refused here for it.
continue
}
if g != nil {
given[m.Module] = g
@@ -1000,6 +1020,20 @@ func planCommand(ctx context.Context, args []string) error {
return nil
}
// Which modules a push would leave out, and why — said before the plan, since the plan is of
// what the machine would be told (novox/hq ADR 0163, rule 6). Judged, never composed: `plan`
// without --json allocates nothing.
if record, err := open.inventory.NodeByName(ctx, args[0]); err == nil {
left := plan.LeftOut(settings, record.Adopted)
reportLeftOut(args[0], sendable{LeftOut: sortedKeysOf(left), leftOutWhy: left})
}
// And a setting that reaches nothing — refused where it is stored, and said here for one
// stored before its definition moved from under it.
for _, m := range plan.Modules {
for _, stray := range catalogue.UnusedSettings(m, settings[m.Module]) {
fmt.Printf("%s: a setting reaches nothing — %s\n", args[0], stray)
}
}
fmt.Printf("%s would run:\n", args[0])
for _, m := range plan.Modules {
fmt.Printf(" %-20s %s\n", m.Module, plan.Because[m.Module])
+19 -3
View File
@@ -131,10 +131,17 @@ func serve(ctx context.Context) error {
// And the mesh's own verbs, as the seat this control plane holds (novox/hq ADR 0154). Served
// from the store's row, so what the seat declares is what is answered.
handlers, err := seatToolHandlers()
handlers, behind, err := seatToolHandlers()
if err != nil {
return err
}
if len(behind) > 0 {
// Said once, loudly, and then served anyway (novox/hq ADR 0185): the mesh keeps answering
// while whatever put an older control plane here is undone.
fmt.Printf("this control plane is behind the %s row: it cannot run %s. "+
"Those answer the reason when called; everything else is served as usual\n",
catalogue.ControllerSeatName, strings.Join(behind, ", "))
}
bus, isNATS := server.Bus().(link.OverNATS)
if !isNATS {
return errors.New("the mesh's verbs are served over the bus, and this control plane is not on it")
@@ -353,7 +360,11 @@ func pushCommand(ctx context.Context, args []string) error {
// The private network is in here with everything else. It used to be composed separately
// and prepended, which meant every machine with an address was on it and no machine could
// be kept off. It is a module now, so it arrives the way a module does.
return declarationWith(held, open, node, plan, settings, gens, Allocating)
declared, err := declarationWith(held, open, node, plan, settings, gens, Allocating)
if err == nil {
reportLeftOut(node, declared)
}
return declared, err
})
sentDigest := map[string]string{}
@@ -454,7 +465,11 @@ func pushCommand(ctx context.Context, args []string) error {
return sendable{}, err
}
reportUnhostable(node, plan)
return declarationWith(held, open, node, plan, settings, gens, Allocating)
declared, err := declarationWith(held, open, node, plan, settings, gens, Allocating)
if err == nil {
reportLeftOut(node, declared)
}
return declared, err
},
func(s readyNode, body []byte) error {
if err := link.Declare(ctx, server.Bus(), ident, s.node, body,
@@ -666,6 +681,7 @@ func sendTo(ctx context.Context, open *stores, names []string) error {
refusals = append(refusals, fmt.Sprintf("%s:\n%v", name, err))
continue
}
reportLeftOut(name, declared)
sending = append(sending, readyNode{name, declared})
}
if len(refusals) > 0 {
+31
View File
@@ -3,6 +3,7 @@ package main
import (
"encoding/json"
"fmt"
"github.com/novox/mesh-controller/internal/inventory"
"sort"
"time"
)
@@ -66,6 +67,21 @@ type meshStatus struct {
// **A document without this said an outage was a well mesh.** Read from what each machine
// reported, so it is the machine's account and not the mesh's take-time listing.
Untaken []machineUntaken `json:"untaken,omitempty"`
// Filtered is every converged machine that is not filtered by the mesh alone (novox/hq ADR
// 0168), one entry per rule set the mesh did not write — the found firewall in force again,
// or a chain nobody speaks for. Absent when every converged machine is filtered by the mesh
// alone. A document without this called a machine well while a predecessor's chain refused
// what the mesh declared open.
Filtered []machineFiltered `json:"filtered,omitempty"`
}
// machineFiltered is one rule set on a converged machine that the mesh did not write and that
// refuses traffic: where it is, whose the host reads it as, and what it refuses.
type machineFiltered struct {
Node string `json:"node"`
Where string `json:"where"`
Owner string `json:"owner"`
Refuses string `json:"refuses"`
}
// machineUntaken is one module a machine is holding rather than running, and how many resources of
@@ -173,6 +189,21 @@ func statusAsJSON(asked answers) ([]byte, error) {
machineUntaken{Node: name, Module: m, Held: asked.untaken[name][m]})
}
}
filteredNodes := make([]string, 0, len(asked.filtered))
for name := range asked.filtered {
filteredNodes = append(filteredNodes, name)
}
sort.Strings(filteredNodes)
for _, name := range filteredNodes {
f := asked.filtered[name]
if fw := f.FoundFirewall; fw != nil && fw.Active {
out.Filtered = append(out.Filtered, machineFiltered{Node: name, Where: "the found firewall",
Owner: inventory.FilterFoundFirewall, Refuses: fw.Kind + " is in force again"})
}
for _, x := range f.Others() {
out.Filtered = append(out.Filtered, machineFiltered{Node: name, Where: x.Where, Owner: x.Owner, Refuses: x.Refuses})
}
}
for name := range asked.refused {
out.Unresolved = append(out.Unresolved, machineUnresolved{
Node: name, Problem: asked.refused[name]})
+43
View File
@@ -167,3 +167,46 @@ func TestAMachineFailingTheSameWayIsSaidToBeStuck(t *testing.T) {
t.Fatalf("one failure is not stuck: %v", once)
}
}
// A converged machine something other than the mesh filters is named, per rule set, and is not
// well (novox/hq ADR 0168); one filtered by the mesh alone is not in the list.
func TestAMachineNotFilteredByTheMeshAloneIsNamedAndNotWell(t *testing.T) {
alone := inventory.Filtering{Filters: []inventory.Filter{
{Where: "table inet mesh, chain input", Owner: inventory.FilterMesh, Refuses: "policy drop"},
{Where: "table ip filter, chain DOCKER", Owner: inventory.FilterRuntime, Refuses: "drop"},
{Where: "table ip filter, chain f2b-sshd", Owner: inventory.FilterBan, Refuses: "ip saddr 192.0.2.1 reject"},
}}
if !alone.Alone() {
t.Fatal("the mesh's own, the runtime's and a ban are not the mesh alone")
}
notAlone := inventory.Filtering{
Filters: append(alone.Filters, inventory.Filter{Where: "chain HAL-MESH-ONLY (iptables-legacy)",
Owner: inventory.FilterOther, Refuses: `-A HAL-MESH-ONLY -m comment --comment "not public" -j DROP`}),
FoundFirewall: &inventory.FoundFirewall{Kind: "ufw", Active: true},
}
asked := answers{nodes: []inventory.Node{{Name: "home-server"}, {Name: "laptop"}},
filtered: map[string]inventory.Filtering{"home-server": notAlone}}
if asked.well() {
t.Fatal("a machine not filtered by the mesh alone reads as well")
}
body, err := statusAsJSON(asked)
if err != nil {
t.Fatal(err)
}
var parsed struct {
Filtered []map[string]string `json:"filtered"`
}
if err := json.Unmarshal(body, &parsed); err != nil {
t.Fatal(err)
}
if len(parsed.Filtered) != 2 {
t.Fatalf("filtered: %v", parsed.Filtered)
}
if parsed.Filtered[0]["node"] != "home-server" || parsed.Filtered[0]["owner"] != inventory.FilterFoundFirewall ||
parsed.Filtered[1]["where"] != "chain HAL-MESH-ONLY (iptables-legacy)" || parsed.Filtered[1]["owner"] != inventory.FilterOther {
t.Fatalf("filtered: %v", parsed.Filtered)
}
if body, _ := statusAsJSON(answers{nodes: asked.nodes}); strings.Contains(string(body), `"filtered"`) {
t.Fatal("a mesh filtered by itself alone carries a filtered list")
}
}
+124 -7
View File
@@ -48,6 +48,21 @@ func argvFor(verb string, args map[string]any) ([]string, error) {
return nil
}
switch verb {
case "command":
// The generic verb: the command line as given, split as a shell would split it, with
// nothing added — the named verbs add flags a caller cannot reach; this one is the whole
// binary and says so in its description (novox/hq ADR 0154, 0175).
if err := need("command"); err != nil {
return nil, err
}
argv, err := splitCommandLine(str("command"))
if err != nil {
return nil, err
}
if len(argv) == 0 {
return nil, errors.New("command names no command")
}
return argv, nil
case "status":
return []string{"status", "--json"}, nil
case "nodes":
@@ -129,6 +144,33 @@ func argvFor(verb string, args map[string]any) ([]string, error) {
// Half of either shape: the command says its usage, which names both shapes, and that is
// the answer the caller needs.
return []string{"rotate"}, nil
case "settings":
// `settings set|clear` at a shell (novox/hq issue 198). The values travel as an argument
// because a tool has no file to hand the command; the command reads either.
if err := need("module"); err != nil {
return nil, err
}
argv := []string{"settings", "set", str("module")}
switch {
case str("clear") == "true":
argv = []string{"settings", "clear", str("module")}
case str("values") != "":
argv = append(argv, str("values"))
}
// Neither values nor clear: the command says its usage, which names both, and that is the
// answer the caller needs — the same as `rotate` given half of either shape.
if n := str("node"); n != "" {
argv = append(argv, "--node", n)
}
return argv, nil
case "issue":
// The same act as `module issue` at a shell (novox/hq design 25 §4): the account is minted
// into the mesh's records and delivered at the machine's next push, which is the caller's to
// ask for — so the mesh is never pushed as a side effect of a credential.
if err := need("node", "module"); err != nil {
return nil, err
}
return []string{"module", "issue", str("module"), "--node", str("node")}, nil
case "build":
if err := need("repository"); err != nil {
return nil, err
@@ -188,13 +230,15 @@ func runVerb(ctx context.Context, argv []string) (verbAnswer, error) {
}
// seatToolHandlers are the handlers for every verb the mesh-controller seat declares, from the
// store's row, so a verb the row does not carry is not served and a verb it carries that this binary
// cannot run is said at start rather than at the first call.
func seatToolHandlers() (map[string]link.ToolHandler, error) {
// store's row, so a verb the row does not carry is not served. A verb it carries that this binary
// cannot run is named at start and answers the reason when called — never a refusal to serve, which
// would take the whole control plane down for one word (novox/hq ADR 0185).
func seatToolHandlers() (map[string]link.ToolHandler, []string, error) {
seat, known := catalogue.SeatNamed(catalogue.ControllerSeatName)
if !known {
return nil, fmt.Errorf("this mesh defines no %s seat", catalogue.ControllerSeatName)
return nil, nil, fmt.Errorf("this mesh defines no %s seat", catalogue.ControllerSeatName)
}
var behind []string
handlers := map[string]link.ToolHandler{}
for _, v := range seat.Serves {
verb := v.Name
@@ -205,8 +249,27 @@ func seatToolHandlers() (map[string]link.ToolHandler, error) {
continue
}
if _, err := argvFor(verb, sampleArguments(v)); err != nil {
return nil, fmt.Errorf("the %s seat's row declares %q, which this control plane cannot run: %w",
catalogue.ControllerSeatName, verb, err)
// **A row ahead of this binary is not a reason to go silent.**
//
// The row is the store's and a control plane follows it (novox/hq ADR 0154), so a verb
// this build does not know means the row was widened by a newer one — the ordinary
// state of a roll-out, and of a push that put an older control plane back. Refusing to
// serve at all made that transient fatal: on 2026-10-02 one unknown verb took the whole
// mesh off the bus for ten minutes, and the way back was a human running the binary by
// hand, because the thing that would have repaired it is the thing that was down
// (novox/hq 04-ISSUES/201, ADR 0185).
//
// So the verbs this binary knows are served, and this one answers the reason instead of
// nothing: a caller gets a sentence naming the fault, and everything else keeps working
// — including the push that replaces this binary with the one whose verb it is.
behind = append(behind, verb)
reason := err
handlers[verb] = func(context.Context, json.RawMessage) (any, error) {
return nil, fmt.Errorf("%s is in this mesh's %s row and the control plane running "+
"here cannot run it: %w. It is a verb of a newer build; this one is behind",
verb, catalogue.ControllerSeatName, reason)
}
continue
}
handlers[verb] = func(ctx context.Context, raw json.RawMessage) (any, error) {
args := map[string]any{}
@@ -222,7 +285,7 @@ func seatToolHandlers() (map[string]link.ToolHandler, error) {
return runVerb(ctx, argv)
}
}
return handlers, nil
return handlers, behind, nil
}
// seatTools is what `tools` answers: every seat with a protocol, and the tools each serves, from the
@@ -262,3 +325,57 @@ func sampleArguments(v catalogue.Verb) map[string]any {
}
return sample
}
// splitCommandLine splits a command line into words the way a POSIX shell does for the simple
// cases a controller command needs: spaces separate, single or double quotes group, a backslash
// escapes the next character inside double quotes or outside any. No expansion of anything.
func splitCommandLine(line string) ([]string, error) {
var words []string
var cur strings.Builder
inWord := false
quote := rune(0)
runes := []rune(line)
for i := 0; i < len(runes); i++ {
r := runes[i]
switch {
case quote == '\'':
if r == '\'' {
quote = 0
} else {
cur.WriteRune(r)
}
case quote == '"':
if r == '"' {
quote = 0
} else if r == '\\' && i+1 < len(runes) {
i++
cur.WriteRune(runes[i])
} else {
cur.WriteRune(r)
}
case r == '\'' || r == '"':
quote = r
inWord = true
case r == '\\' && i+1 < len(runes):
i++
cur.WriteRune(runes[i])
inWord = true
case r == ' ' || r == '\t' || r == '\n':
if inWord {
words = append(words, cur.String())
cur.Reset()
inWord = false
}
default:
cur.WriteRune(r)
inWord = true
}
}
if quote != 0 {
return nil, fmt.Errorf("command has an unclosed %c quote", quote)
}
if inWord {
words = append(words, cur.String())
}
return words, nil
}
+111 -1
View File
@@ -1,6 +1,7 @@
package main
import (
"context"
"strings"
"testing"
@@ -73,6 +74,38 @@ func TestRotateTakesAProvisionOrAnOwnSecret(t *testing.T) {
}
}
// `settings` is `settings set|clear` at a shell, with the values passed inline (novox/hq issue 198).
func TestSettingsSetsOrClearsALayer(t *testing.T) {
argv, err := argvFor("settings", map[string]any{"module": "dnsmasq", "values": `{"a":1}`, "node": "ace"})
if err != nil || strings.Join(argv, " ") != `settings set dnsmasq {"a":1} --node ace` {
t.Fatalf("set on a machine: %v %v", argv, err)
}
argv, _ = argvFor("settings", map[string]any{"module": "dnsmasq", "clear": "true"})
if strings.Join(argv, " ") != "settings clear dnsmasq" {
t.Fatalf("clear for the mesh: %v", argv)
}
argv, _ = argvFor("settings", map[string]any{"module": "dnsmasq"})
if strings.Join(argv, " ") != "settings set dnsmasq" {
t.Fatalf("a set with no values falls to the command's usage: %v", argv)
}
}
// `issue` is `module issue` at a shell: the module and the machine, and nothing that would push. A
// module's bus account was mintable only from the controller's command line, so an agent working
// through the tools could not finish a rollout that gave a module one (novox/hq issue 191).
func TestIssueGivesAModuleItsAccountOnAMachine(t *testing.T) {
argv, err := argvFor("issue", map[string]any{"node": "ace", "module": "route-proxy"})
if err != nil {
t.Fatal(err)
}
if strings.Join(argv, " ") != "module issue route-proxy --node ace" {
t.Fatalf("issue runs %v", argv)
}
if _, err := argvFor("issue", map[string]any{"module": "route-proxy"}); err == nil {
t.Error("an account was issued without saying which machine reads it")
}
}
// A required argument missing is refused in the verb's own words, before anything runs.
func TestAVerbMissingWhatItNeedsIsRefused(t *testing.T) {
if _, err := argvFor("node", map[string]any{}); err == nil || !strings.Contains(err.Error(), `node needs "node"`) {
@@ -98,10 +131,13 @@ func TestActsDoNotBlockTheCall(t *testing.T) {
// What `tools` answers is the seats' records, with each verb's schema.
func TestToolsAnswersTheSeatsRecords(t *testing.T) {
handlers, err := seatToolHandlers()
handlers, behind, err := seatToolHandlers()
if err != nil {
t.Fatal(err)
}
if len(behind) != 0 {
t.Fatalf("this build cannot run %v of its own seat's verbs", behind)
}
if len(handlers) != len(catalogue.ControllerVerbs) {
t.Fatalf("%d handlers for %d verbs", len(handlers), len(catalogue.ControllerVerbs))
}
@@ -139,3 +175,77 @@ func TestAJSONVerbsAnswerIsItsStandardOutput(t *testing.T) {
t.Fatalf("stderr and stdout are both what the command said: %s", answer.Output)
}
}
// `command` is the generic verb: the command line as given, split as a shell would, nothing added —
// so an operator's `node account g14 jochen` is one call through the console rather than a shell on
// the control node (novox/hq ADR 0154, ADR 0175).
func TestCommandRunsTheLineAsGiven(t *testing.T) {
argv, err := argvFor("command", map[string]any{"command": "node account g14 jochen"})
if err != nil || strings.Join(argv, " ") != "node account g14 jochen" {
t.Fatalf("a plain line: %v %v", argv, err)
}
argv, err = argvFor("command", map[string]any{"command": `settings set dnsmasq '{"a": "b c"}' --node ace`})
if err != nil || len(argv) != 6 || argv[3] != `{"a": "b c"}` {
t.Fatalf("a quoted word stays one word: %q %v", argv, err)
}
argv, err = argvFor("command", map[string]any{"command": `node add "the box" --adopted`})
if err != nil || len(argv) != 4 || argv[2] != "the box" {
t.Fatalf("double quotes group: %q %v", argv, err)
}
if _, err := argvFor("command", map[string]any{"command": " "}); err == nil {
t.Fatal("an empty line was accepted")
}
if _, err := argvFor("command", map[string]any{"command": `node "unclosed`}); err == nil {
t.Fatal("an unclosed quote was accepted")
}
}
// A verb in the row that this binary cannot run does not take the control plane off the bus: the
// rest are served, the unknown one answers the reason, and the start-up names it (novox/hq ADR
// 0185). One unknown word cost the mesh ten minutes of silence on 2026-10-02, recoverable only by
// a person running the binary by hand — the push that would have repaired it needs the control
// plane that was down.
func TestARowAheadOfThisBuildIsServedAnyway(t *testing.T) {
seat, known := catalogue.SeatNamed(catalogue.ControllerSeatName)
if !known {
t.Fatal("no controller seat")
}
// The row as a newer control plane would have written it: every verb this build knows, and one
// it does not.
widened := seat
widened.Serves = append(append([]catalogue.Verb{}, seat.Serves...),
catalogue.Verb{Name: "teleport", Description: "a verb from a build that does not exist yet"})
rows := catalogue.DefaultSeats()
for i := range rows {
if rows[i].Name == catalogue.ControllerSeatName {
rows[i] = widened
}
}
catalogue.UseSeats(rows)
t.Cleanup(func() { catalogue.UseSeats(catalogue.DefaultSeats()) })
handlers, behind, err := seatToolHandlers()
if err != nil {
t.Fatalf("a row with one unknown verb refused to serve at all: %v", err)
}
if len(behind) != 1 || behind[0] != "teleport" {
t.Fatalf("the verbs this build cannot run were reported as %v", behind)
}
if len(handlers) != len(widened.Serves) {
t.Fatalf("%d handlers for %d verbs in the row", len(handlers), len(widened.Serves))
}
for _, known := range []string{"status", "nodes", "push"} {
if handlers[known] == nil {
t.Errorf("%s is not served although this build knows it", known)
}
}
_, err = handlers["teleport"](context.Background(), nil)
if err == nil {
t.Fatal("the unknown verb answered as though it had run")
}
for _, want := range []string{"teleport", "cannot run it", "behind"} {
if !strings.Contains(err.Error(), want) {
t.Errorf("the answer does not say %q: %v", want, err)
}
}
}
+11
View File
@@ -31,6 +31,14 @@ type sendable struct {
// the same composition as its received files, and every machine's private-network address.
Received map[string]map[string][]catalogue.Contribution
Mesh []string
// LeftOut is every module of the machine's set left out of this declaration because a stored
// setting cannot compose with its definition (novox/hq ADR 0163, rule 6), sorted. The host
// keeps that module's held things and touches none of its containers; a machine is told
// everything or nothing about what it IS told, and what it is not told is said. Absent from
// the body when empty, so a declaration that leaves nothing out is byte for byte what it was.
LeftOut []string
// leftOutWhy is why each was, for push and plan to say; never on the wire.
leftOutWhy map[string]string
}
// adoptionEnvelope is what an adopted node is told about its mode. Taken is every module taken on
@@ -51,6 +59,9 @@ func (s sendable) Body() ([]byte, error) {
if s.Sequence > 0 {
envelope["sequence"] = s.Sequence
}
if len(s.LeftOut) > 0 {
envelope["left_out"] = s.LeftOut
}
// An empty declaration is deliberate here — the node owns nothing the mesh put there
// (novox/hq issue 127) — and the host refuses an empty body unless it is told the emptiness
// is meant, so a truncated or mis-composed body is never mistaken for "own nothing".
+59
View File
@@ -382,3 +382,62 @@ func TestAnEmptyDeclarationSaysOwnsNothing(t *testing.T) {
t.Fatalf("a non-empty declaration must not mark owns_nothing; got %v", env)
}
}
// A setting is judged where it is stored, and an impossible one costs a module, not a machine
// (novox/hq ADR 0163, rule 6): stored while it composed, a setting whose definition then moved from
// under it leaves that module out of the declaration — said in the envelope, so the host keeps the
// module's things — and the machine is told everything else.
func TestADefinitionMovingUnderAStoredSettingLeavesThatModuleOutNotTheMachine(t *testing.T) {
open := aMesh(t)
ctx := t.Context()
web := helloWeb()
web.Resources[1]["ports"] = []any{"8080"}
register(t, open, web)
register(t, open, catalogue.Manifest{Module: "notes", Version: "1",
Resources: []map[string]any{{"id": "conf", "type": "file", "path": "/etc/notes.conf", "content": "x"}}})
for _, m := range []string{"hello-web", "notes"} {
if _, err := assign(ctx, open, "laptop", m); err != nil {
t.Fatal(err)
}
}
// Judged where it is stored: a port the module does not publish is refused by name.
err := open.inventory.SetSettings(ctx, "laptop", "hello-web",
map[string]any{catalogue.PortsSetting: map[string]any{"9999": 10000}})
if err == nil || !strings.Contains(err.Error(), "hello-web on laptop") || !strings.Contains(err.Error(), "9999") {
t.Fatalf("an impossible setting was stored: %v", err)
}
if err := open.inventory.SetSettings(ctx, "laptop", "hello-web",
map[string]any{catalogue.PortsSetting: map[string]any{"8080": 10000}}); err != nil {
t.Fatal(err)
}
if declared := composed(t, open, "laptop"); len(declared.LeftOut) != 0 {
t.Fatalf("a setting that composes left a module out: %v", declared.LeftOut)
}
// The definition moves: the container publishes another port now.
web.Version = "2"
web.Resources[1]["ports"] = []any{"9090"}
register(t, open, web)
declared := composed(t, open, "laptop")
if len(declared.LeftOut) != 1 || declared.LeftOut[0] != "hello-web" {
t.Fatalf("hello-web is not left out: %v", declared.LeftOut)
}
if !strings.Contains(declared.leftOutWhy["hello-web"], "no container of its publishes 8080") {
t.Fatalf("why it was left out is not said: %v", declared.leftOutWhy)
}
if hasID(declared.Resources, "hello-web.server") || !hasID(declared.Resources, "notes.conf") {
t.Fatalf("the machine was not told everything else: %v", declared.Resources)
}
body, err := declared.Body()
if err != nil {
t.Fatal(err)
}
var env map[string]any
if err := json.Unmarshal(body, &env); err != nil {
t.Fatal(err)
}
left, _ := env["left_out"].([]any)
if len(left) != 1 || left[0] != "hello-web" {
t.Fatalf("the envelope does not say what was left out: %v", env)
}
}
+55 -1
View File
@@ -227,6 +227,28 @@ func printStatus(asked answers) error {
" not readable from a commit; that needs a version the host reports as ordered\n\n")
}
if len(asked.filtered) > 0 {
// A converged machine is filtered by the mesh alone, and the mesh says truthfully which
// (novox/hq ADR 0168). One that is not — a predecessor's chain still refusing, a found
// firewall in force again — is said here, and is not well.
machines := make([]string, 0, len(asked.filtered))
for name := range asked.filtered {
machines = append(machines, name)
}
sort.Strings(machines)
fmt.Printf("%d converged machine(s) are not filtered by the mesh alone:\n", len(machines))
for _, name := range machines {
f := asked.filtered[name]
if fw := f.FoundFirewall; fw != nil && fw.Active {
fmt.Printf(" %-12s the found firewall (%s) is in force again; the next apply retires it\n", name, fw.Kind)
}
for _, x := range f.Others() {
fmt.Printf(" %-12s %s (%s): %s\n", name, x.Where, x.Owner, x.Refuses)
}
}
fmt.Printf("\n the mesh wrote none of these and removes none; `node show <node>` lists every filter with its owner\n\n")
}
if len(asked.untaken) > 0 {
// **Before the adopted line, and it breaks "all well".** An adopted machine is a state
// somebody chose and can leave alone; a module assigned to one and never taken is work
@@ -357,6 +379,13 @@ func theThreeQuestions(ctx context.Context, open *stores) (answers, error) {
if err != nil {
return answers{}, err
}
// And which converged machines something other than the mesh filters (novox/hq ADR 0168), as
// each last reported — the account that was missing when a predecessor's chain refused what the
// mesh declared open for eleven hours (04-ISSUES/144, 145).
out.filtered, err = filteredMachines(ctx, inv, out.nodes)
if err != nil {
return answers{}, err
}
out.plans, err = inv.RecentPlans(ctx, 5)
if err != nil {
return answers{}, err
@@ -406,6 +435,30 @@ func theThreeQuestions(ctx context.Context, open *stores) (answers, error) {
//
// A machine that reports no holds contributes nothing, so a converged mesh answers an empty map and
// the caller prints nothing.
// filteredMachines is every converged machine not filtered by the mesh alone, with what it last said
// filters it (novox/hq ADR 0168). An adopted machine keeps its found firewall by design and is not
// counted; a machine that has not said is not said to be filtered by anything.
func filteredMachines(ctx context.Context, inv *inventory.Inventory, nodes []inventory.Node) (
map[string]inventory.Filtering, error) {
out := map[string]inventory.Filtering{}
for _, n := range nodes {
if n.Adopted {
continue
}
f, err := inv.FilteringOf(ctx, n.Name)
if err != nil {
return nil, fmt.Errorf("what filters %s cannot be read: %w", n.Name, err)
}
if len(f.Filters) == 0 && f.FoundFirewall == nil {
continue
}
if !f.Alone() {
out[n.Name] = f
}
}
return out, nil
}
func untakenModules(ctx context.Context, inv *inventory.Inventory, nodes []inventory.Node) (
map[string]map[string]int, error) {
@@ -442,7 +495,8 @@ func untakenModules(ctx context.Context, inv *inventory.Inventory, nodes []inven
// read as success for the whole of the edge cut-over outage (novox/hq 04-ISSUES/125).
func (a answers) well() bool {
return len(a.wrong) == 0 && len(a.quiet) == 0 && len(a.behind) == 0 &&
len(a.waiting) == 0 && len(a.refused) == 0 && a.network == "" && len(a.untaken) == 0
len(a.waiting) == 0 && len(a.refused) == 0 && a.network == "" && len(a.untaken) == 0 &&
len(a.filtered) == 0
}
// hostSplit is which machines report which host version, for every version more than one machine
+110 -16
View File
@@ -4,26 +4,42 @@ import (
"strings"
"testing"
"github.com/novox/mesh-controller/internal/catalogue"
"github.com/novox/mesh-controller/internal/inventory"
)
// What the forge's take compares, as a machine would report it.
func aForgeComparison() comparison {
return comparison{reported: inventory.Adoption{
Firewall: "ufw",
Held: []inventory.Held{
{ID: "forge.server", Module: "forge", Kind: "container", Target: "forge", Facts: map[string]any{
"image": "forge:1.27.3", "image_created": "2026-09-17T10:00:00Z",
"declared_image": "forge:1.22.6", "declared_image_created": "2026-08-20T10:00:00Z", "downgrade": true,
"networks": map[string]any{"predecessor_default": []any{"office", "db"}},
"ports": []any{"3000/tcp>0.0.0.0:3000"}, "declared_ports": []any{"3000:3000"},
}},
{ID: "forge.config", Module: "forge", Kind: "file", Target: "/etc/forge/app.ini", Kept: "/var/lib/mesh/kept/app.ini",
Facts: map[string]any{"differs": true, "difference": []any{"- private scope: local", "+ upstream: public"}}},
{ID: "other.server", Module: "other", Kind: "container", Target: "other"},
},
Reachable: []inventory.Reach{
{Protocol: "tcp", Address: "0.0.0.0", Port: 3000, By: "forge", Published: true, ContainerPort: 3000},
{Protocol: "tcp", Address: "0.0.0.0", Port: 22, By: "sshd"},
},
}}
}
// A take is a comparison (novox/hq ADR 0163): the preview puts what runs beside what the module
// declares, and an older image or a differing file refuses unless named.
func TestATakePreviewsTheComparisonAndRefusesWhatIsNotNamed(t *testing.T) {
held := []inventory.Held{
{ID: "forge.server", Module: "forge", Kind: "container", Target: "forge", Facts: map[string]any{
"image": "forge:1.27.3", "image_created": "2026-09-17T10:00:00Z",
"declared_image": "forge:1.22.6", "declared_image_created": "2026-08-20T10:00:00Z", "downgrade": true,
"networks": map[string]any{"predecessor_default": []any{"office", "db"}},
"ports": []any{"3000/tcp>0.0.0.0:3000"}, "declared_ports": []any{"3000:3000"},
}},
{ID: "forge.config", Module: "forge", Kind: "file", Target: "/etc/forge/app.ini", Kept: "/var/lib/mesh/kept/app.ini",
Facts: map[string]any{"differs": true, "difference": []any{"- private scope: local", "+ upstream: public"}}},
{ID: "other.server", Module: "other", Kind: "container", Target: "other"},
}
preview, refusals := comparisonOf(held, "forge", takeOptions{})
c := aForgeComparison()
preview, refusals, saw := comparisonOf("forge", c, takeOptions{})
for _, want := range []string{"runs forge:1.27.3 (made 2026-09-17)", "declares forge:1.22.6 (made 2026-08-20)", "DOWNGRADE",
"on the network predecessor_default with office, db", "publishes 3000/tcp>0.0.0.0:3000; the module declares 3000:3000",
"on the network predecessor_default with office, db", "will not once it moves to the module's own network",
"publishes 3000/tcp>0.0.0.0:3000; the module declares 3000:3000",
// How far the port reaches now, as the machine reported it (rule 1).
"reachable now at 0.0.0.0:3000 (tcp, container port 3000), behind the found firewall (ufw)",
"- private scope: local", "original kept at /var/lib/mesh/kept/app.ini"} {
if !strings.Contains(preview, want) {
t.Errorf("the preview lacks %q:\n%s", want, preview)
@@ -35,15 +51,93 @@ func TestATakePreviewsTheComparisonAndRefusesWhatIsNotNamed(t *testing.T) {
if len(refusals) != 2 || !strings.Contains(refusals[0], "--downgrade") || !strings.Contains(refusals[1], "--replace /etc/forge/app.ini") {
t.Fatalf("the downgrade and the differing file refuse, each naming its override: %v", refusals)
}
if len(saw) != 12 {
t.Fatalf("the preview's digest is %q", saw)
}
// Named, they pass.
if _, refusals := comparisonOf(held, "forge", takeOptions{Downgrade: true, Replace: map[string]bool{"/etc/forge/app.ini": true}}); len(refusals) != 0 {
if _, refusals, _ := comparisonOf("forge", c, takeOptions{Downgrade: true, Replace: map[string]bool{"/etc/forge/app.ini": true}}); len(refusals) != 0 {
t.Fatalf("named differences still refused: %v", refusals)
}
if _, refusals := comparisonOf(held, "forge", takeOptions{Downgrade: true, Replace: map[string]bool{"*": true}}); len(refusals) != 0 {
if _, refusals, _ := comparisonOf("forge", c, takeOptions{Downgrade: true, Replace: map[string]bool{"*": true}}); len(refusals) != 0 {
t.Fatalf("replace * did not cover the file: %v", refusals)
}
// A held thing with no facts yet — a host older than this — refuses nothing and says what it can.
if preview, refusals := comparisonOf(held, "other", takeOptions{}); len(refusals) != 0 || !strings.Contains(preview, "container other") {
if preview, refusals, _ := comparisonOf("other", c, takeOptions{}); len(refusals) != 0 || !strings.Contains(preview, "container other") {
t.Fatalf("a factless hold: %q %v", preview, refusals)
}
// The digest is of what the preview says: a fact changing changes it.
c.reported.Held[0].Facts["image"] = "forge:1.27.4"
if _, _, again := comparisonOf("forge", c, takeOptions{}); again == saw {
t.Fatal("the found image changed and the digest did not")
}
}
// A secret the mesh minted for a service whose data was found refuses: the running service already
// has a value (rule 2). Accepted, it is carried in; `--mint` says the service shall take the new one.
func TestAMintedSecretForFoundDataRefusesUnlessAcceptedOrMinted(t *testing.T) {
c := aForgeComparison()
c.secrets = []inventory.SecretState{
{Name: "admin", Origin: inventory.OriginMade},
{Name: "postgres-database", Origin: inventory.OriginMade, Provider: "anchor"},
{Name: "broker", Origin: inventory.OriginAccepted},
}
preview, refusals, _ := comparisonOf("forge", c, takeOptions{Downgrade: true, Replace: map[string]bool{"*": true}})
for _, want := range []string{
"own secret admin: MINTED by the mesh and not accepted",
"secret from anchor postgres-database: MINTED by the mesh and not accepted",
"own secret broker: accepted from a person, carried in as it is",
} {
if !strings.Contains(preview, want) {
t.Errorf("the preview lacks %q:\n%s", want, preview)
}
}
if len(refusals) != 2 {
t.Fatalf("two minted secrets refuse: %v", refusals)
}
if !strings.Contains(refusals[0], "`secret accept <node> forge admin`") || !strings.Contains(refusals[0], "`--mint admin`") {
t.Errorf("the own secret's refusal names accepting it and minting it: %s", refusals[0])
}
if !strings.Contains(refusals[1], "`secret accept <node> forge postgres-database --provider anchor`") {
t.Errorf("the required secret's refusal names its provider: %s", refusals[1])
}
preview, refusals, _ = comparisonOf("forge", c, takeOptions{Downgrade: true, Replace: map[string]bool{"*": true},
Mint: map[string]bool{"admin": true, "postgres-database": true}})
if len(refusals) != 0 || !strings.Contains(preview, "admin: minted by the mesh; the service takes the new value, as --mint said") {
t.Fatalf("--mint did not pass the minted secrets: %v\n%s", refusals, preview)
}
// With no found data — only a file held — the service has no value of its own, and a minted
// secret is simply said.
c.reported.Held = c.reported.Held[1:2]
if _, refusals, _ := comparisonOf("forge", c, takeOptions{Replace: map[string]bool{"*": true}}); len(refusals) != 0 {
t.Fatalf("a minted secret refused with no data found: %v", refusals)
}
}
// A found network a per-machine setting keeps is named in the preview (rule 4), and the module's
// settings are said with where each came from, composed or not (rules 1 and 6).
func TestTheKeptNetworkAndTheSettingsAreInThePreview(t *testing.T) {
c := aForgeComparison()
c.keeps = map[string][]string{"forge.server": {"predecessor_default"}}
c.layers = []catalogue.Layer{
{From: catalogue.MeshWideLayer, Values: map[string]any{"site": "x"}},
{From: "anchor", Values: map[string]any{catalogue.NetworksSetting: map[string]any{"server": []any{"predecessor_default"}}}},
}
preview, _, _ := comparisonOf("forge", c, takeOptions{Downgrade: true, Replace: map[string]bool{"*": true}})
for _, want := range []string{
"on the network predecessor_default with office, db — kept by this machine's setting, so they still reach it by name once taken",
"settings from the mesh: site",
"settings from anchor: networks",
} {
if !strings.Contains(preview, want) {
t.Errorf("the preview lacks %q:\n%s", want, preview)
}
}
if strings.Contains(preview, "will not once it moves") {
t.Errorf("a kept network is still said to be lost:\n%s", preview)
}
c.settingsRefused = "forge: ports is a { port: machine-port } map"
preview, _, _ = comparisonOf("forge", c, takeOptions{Downgrade: true, Replace: map[string]bool{"*": true}})
if !strings.Contains(preview, "SETTINGS DO NOT COMPOSE with the module's definition, so the push leaves it out: forge: ports") {
t.Errorf("settings that cannot compose are not said:\n%s", preview)
}
}
+4 -1
View File
@@ -10,7 +10,10 @@
# The client is copied from the vendor's own image rather than installed from a distribution:
# `apk add mc` on Alpine installs Midnight Commander, which is a different program with the same
# name, and the failure would be a provisioner that starts cleanly and cannot do anything.
FROM golang:1.25-alpine AS build
# The Go it builds with is the one the manifest pins (build.on GO_BASE), passed by the Makefile and the
# build machine alike; the default only serves a hand build, and matches go.mod.
ARG GO_BASE=golang:1.26-alpine
FROM ${GO_BASE} AS build
WORKDIR /src
COPY go.mod go.sum ./
RUN go mod download
+4 -1
View File
@@ -3,7 +3,10 @@
# Built here so a machine can be given it by the mesh rather than by somebody putting a binary on
# it. Static and FROM scratch for the same reason the control plane's image is: it is fetched by
# digest and run on a machine, and everything in it is something a person would have to audit.
FROM golang:1.25-alpine AS build
# The Go it builds with is the one the manifest pins (build.on GO_BASE), passed by the Makefile and the
# build machine alike; the default only serves a hand build, and matches go.mod.
ARG GO_BASE=golang:1.26-alpine
FROM ${GO_BASE} AS build
WORKDIR /src
COPY go.mod go.sum ./
RUN go mod download
+4 -1
View File
@@ -2,7 +2,10 @@
#
# FROM scratch, like the postgres one and unlike the bucket one: it speaks the store's own wire
# protocol directly and needs no client in the image.
FROM golang:1.25-alpine AS build
# The Go it builds with is the one the manifest pins (build.on GO_BASE), passed by the Makefile and the
# build machine alike; the default only serves a hand build, and matches go.mod.
ARG GO_BASE=golang:1.26-alpine
FROM ${GO_BASE} AS build
WORKDIR /src
COPY go.mod go.sum ./
RUN go mod download
+4 -1
View File
@@ -2,7 +2,10 @@
#
# Static and FROM scratch like the control plane's image, and for the same reason: it is fetched
# by digest and run on a machine, so everything in it is something a person would have to audit.
FROM golang:1.25-alpine AS build
# The Go it builds with is the one the manifest pins (build.on GO_BASE), passed by the Makefile and the
# build machine alike; the default only serves a hand build, and matches go.mod.
ARG GO_BASE=golang:1.26-alpine
FROM ${GO_BASE} AS build
WORKDIR /src
COPY go.mod go.sum ./
RUN go mod download
+5
View File
@@ -732,6 +732,11 @@ func handler(held *table) http.Handler {
// And since a host may now be routed only on some paths, those are a third thing:
// saying "no route for this name" while listing that very name as served is a
// contradiction an operator would have to disbelieve the proxy to get past.
// **Said in the log as well as to the client.** A name this mesh does not serve, asked
// for from outside, is what a scanner does, and the machine's intrusion prevention reads
// this proxy's log for exactly that line (novox/hq ADR 0179): the address last, as the
// jail's filter expects it.
log.Printf("refused: no route for %q, asked from %s", r.Host, r.RemoteAddr)
w.Header().Set("Content-Type", "text/plain; charset=utf-8")
w.WriteHeader(http.StatusNotFound)
if !hidden && held.routed(r.Host) {
+99
View File
@@ -0,0 +1,99 @@
// Package artifacts speaks to the mesh's artifact store over its own door.
//
// Only what the mesh needs that nothing else does: letting go of something it put there
// (novox/hq ADR 0189, issue 108). Pushing is the builder's, through the container runtime; reading
// is every machine's, through its runtime. This is the one operation that belongs to the thing
// holding the records, because it is the only one that is a decision rather than a transfer.
package artifacts
import (
"context"
"fmt"
"net/http"
"strings"
"time"
"github.com/novox/mesh-controller/internal/catalogue"
)
// Store is the artifact store at an address, as this machine reaches it.
type Store struct {
// Address is `host:port` — the store as the caller reaches it now, composed and never
// recorded (novox/hq 04-ISSUES/102).
Address string
// HTTP is the client used; nil is a client with a modest timeout.
HTTP *http.Client
}
// Gone is the answer when the store does not hold it: the outcome wanted, already true.
var Gone = fmt.Errorf("the store does not hold it")
// LetGo asks the store to drop one artifact the mesh recorded making.
//
// Takes a reference as the mesh records it — `artifact-store://<module>/<artifact>@sha256:…` for
// an image, `…/blobs/sha256:…` for an archive — because that is the identity every record uses,
// and composes the address here at the moment of use.
//
// Returns Gone when the store answers that it does not have it. That is not a failure: the sweep
// wants the artifact absent, and it is. It is distinguished from success only so a caller can say
// which of the two happened.
func (s Store) LetGo(ctx context.Context, reference string) error {
path, kept := catalogue.InArtifactStore(reference)
if !kept {
// Nothing the mesh put in its own store. Refused rather than attempted: composing a
// delete for a reference of unknown shape is how a sweep reaches something that is not
// the mesh's.
return fmt.Errorf("%s is not a reference into the mesh's artifact store", reference)
}
if s.Address == "" {
return fmt.Errorf("this mesh has no artifact store on its network to ask about %s", reference)
}
repository, kind, digest, err := split(path)
if err != nil {
return err
}
url := "http://" + s.Address + "/v2/" + repository + "/" + kind + "/" + digest
request, err := http.NewRequestWithContext(ctx, http.MethodDelete, url, nil)
if err != nil {
return err
}
client := s.HTTP
if client == nil {
client = &http.Client{Timeout: 30 * time.Second}
}
response, err := client.Do(request)
if err != nil {
return err
}
defer response.Body.Close()
switch response.StatusCode {
case http.StatusAccepted, http.StatusOK, http.StatusNoContent:
return nil
case http.StatusNotFound:
return Gone
case http.StatusMethodNotAllowed:
// The registry was started without deletion enabled. Said plainly, because the remedy is
// a setting on the store's module and not anything about this artifact.
return fmt.Errorf(
"the artifact store refuses deletion: its server was started without it enabled "+
"(REGISTRY_STORAGE_DELETE_ENABLED), so nothing can be collected until the store "+
"module is applied again (novox/hq ADR 0189). Asking about %s", reference)
default:
return fmt.Errorf("the artifact store answered %s for %s", response.Status, reference)
}
}
// split reads a recorded path into the repository, which endpoint names the thing, and the digest.
//
// Two shapes, which are the two the mesh records: `<repository>@sha256:<hex>` is a manifest, and
// `<repository>/blobs/sha256:<hex>` is a blob.
func split(path string) (repository, kind, digest string, err error) {
if before, after, ok := strings.Cut(path, "@sha256:"); ok {
return before, "manifests", "sha256:" + after, nil
}
if before, after, ok := strings.Cut(path, "/blobs/sha256:"); ok {
return before, "blobs", "sha256:" + after, nil
}
return "", "", "", fmt.Errorf("%q names nothing the store holds by digest", path)
}
+94
View File
@@ -0,0 +1,94 @@
package artifacts
import (
"context"
"errors"
"net/http"
"net/http/httptest"
"strings"
"testing"
"github.com/novox/mesh-controller/internal/catalogue"
)
// Asking the store to let go of what the mesh no longer keeps (novox/hq ADR 0189, issue 108).
//
// A fake store records what it was asked to delete, so what is asserted is the mesh's decision
// and the shape of the request — not the registry's behaviour, which is the registry's to test.
func fakeStore(t *testing.T, answer int) (Store, *[]string) {
t.Helper()
var asked []string
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if r.Method != http.MethodDelete {
t.Errorf("the store was asked %s %s; collecting is a delete", r.Method, r.URL.Path)
}
asked = append(asked, r.URL.Path)
w.WriteHeader(answer)
}))
t.Cleanup(server.Close)
return Store{Address: strings.TrimPrefix(server.URL, "http://")}, &asked
}
func TestAnImageAndAnArchiveAreAskedForAtTheirOwnEndpoints(t *testing.T) {
// The two shapes the mesh records: a manifest by digest, and a blob by digest. They are
// different endpoints, and asking at the wrong one answers 404 — which this would then
// record as collected, leaving the bytes on disk for ever while the record says otherwise.
store, asked := fakeStore(t, http.StatusAccepted)
ctx := context.Background()
image := catalogue.ArtifactStoreScheme + "web/app@sha256:abc123"
archive := catalogue.ArtifactStoreScheme + "web/config/blobs/sha256:def456"
if err := store.LetGo(ctx, image); err != nil {
t.Fatal(err)
}
if err := store.LetGo(ctx, archive); err != nil {
t.Fatal(err)
}
want := []string{"/v2/web/app/manifests/sha256:abc123", "/v2/web/config/blobs/sha256:def456"}
if len(*asked) != 2 || (*asked)[0] != want[0] || (*asked)[1] != want[1] {
t.Fatalf("the store was asked %v; want %v", *asked, want)
}
}
func TestAStoreThatDoesNotHaveItAnswersGone(t *testing.T) {
// The outcome wanted, already true. Told apart from success only so the sweep can say which
// happened; both are recorded, because retrying for ever is the thing to avoid.
store, _ := fakeStore(t, http.StatusNotFound)
err := store.LetGo(context.Background(), catalogue.ArtifactStoreScheme+"web/app@sha256:abc123")
if !errors.Is(err, Gone) {
t.Fatalf("a store that does not hold it answered %v, want Gone", err)
}
}
func TestAStoreWithDeletionOffSaysSoAndNamesTheRemedy(t *testing.T) {
// The registry answers 405 when it was started without deletion enabled. The remedy is a
// setting on the store's module, and saying "405" would send somebody to the wrong place.
store, _ := fakeStore(t, http.StatusMethodNotAllowed)
err := store.LetGo(context.Background(), catalogue.ArtifactStoreScheme+"web/app@sha256:abc123")
if err == nil {
t.Fatal("a store that refuses deletion was read as success")
}
if !strings.Contains(err.Error(), "REGISTRY_STORAGE_DELETE_ENABLED") {
t.Fatalf("the refusal does not name the remedy: %v", err)
}
}
func TestAReferenceThatIsNotTheMeshsOwnIsNeverAsked(t *testing.T) {
// The whole safety of the sweep is that it names only what the mesh recorded putting there.
// A reference of another shape — a vendor's image, a package version — is refused rather
// than composed into a delete somewhere that is not the mesh's store.
store, asked := fakeStore(t, http.StatusAccepted)
for _, reference := range []string{
"docker.io/library/registry@sha256:abc123",
"registry@sha256:abc123",
"1.4.2",
} {
if err := store.LetGo(context.Background(), reference); err == nil {
t.Errorf("%s was asked about; it is not a reference into the mesh's store", reference)
}
}
if len(*asked) != 0 {
t.Fatalf("the store was asked about %v", *asked)
}
}
+12 -4
View File
@@ -46,7 +46,7 @@ func boundUsed(content string) [][2]string {
// Three facts the mesh states about any provision, plus whatever the provider said it serves. A
// module may not reach a binding it does not have — the same boundary as a secret, for the same
// reason.
func knownFor(m Manifest, needs []Needed, node string) map[string]map[string]string {
func knownFor(m Manifest, needs []Needed, node string) (map[string]map[string]string, error) {
out := map[string]map[string]string{}
for _, want := range m.Wants() {
for i := range needs {
@@ -54,12 +54,20 @@ func knownFor(m Manifest, needs []Needed, node string) map[string]map[string]str
if n.Name != want || n.For != m.Module {
continue
}
as := ConsumerIdentity(node, IdentitySource(m.Slug, m.Module))
values := map[string]string{
"at": n.At,
"from": n.From,
"as": ConsumerIdentity(node, IdentitySource(m.Slug, m.Module)),
"as": as,
}
for key, value := range n.Serves {
// What the provider derives for this consumer rather than for all of them
// (novox/hq ADR 0188). Filled here, the one place a provision and the module
// requiring it are both in hand.
served, err := ServedTo(n.Serves, as)
if err != nil {
return nil, fmt.Errorf("%s requires %s: %w", m.Module, want, err)
}
for key, value := range served {
// The provider's own vocabulary. Rendered plainly: a port is 5432, not 5432.000000,
// which is what a float would write and what a connection string would refuse.
values[key] = plainly(value)
@@ -67,7 +75,7 @@ func knownFor(m Manifest, needs []Needed, node string) map[string]map[string]str
out[want] = values
}
}
return out
return out, nil
}
// withOwnNames adds a module's own composed names to what it may name from one binding:
+290
View File
@@ -0,0 +1,290 @@
package catalogue
import (
"fmt"
"regexp"
"sort"
"strings"
)
// What a provider derives for one consumer, said once in the provider's definition and delivered
// to both ends (novox/hq ADR 0188, issue 124).
//
// A `serves` block is otherwise literal: the same values for every consumer. Where the provider
// *names the resource* — a bucket, a database, a vhost — the name is derived from who is asking,
// and before this the mesh had no channel for it. The provider recomputed it in its own code and
// every consumer transcribed it into its own definition by hand, which is a copy of somebody
// else's rule kept in agreement by nobody. One of three transcriptions was wrong for months.
//
// **The mesh learns no protocol here; it spells its own name in an alphabet it already knows.**
// The only fact a served value may name is the identity the mesh itself minted for the consumer,
// in one of two alphabets: as it was minted, and as a DNS label. Everything a provider wants
// around it — a prefix, a suffix, a separator — it writes around the placeholder, because a
// served value is a string.
// consumerFact is `${consumer:<fact>}` or `${consumer:<fact>:<alphabet>}`.
var consumerFact = regexp.MustCompile(`\$\{consumer:([a-z][a-z0-9-]*)(?::([a-z][a-z0-9-]*))?\}`)
// consumerFacts are what a served value may name about the consumer it is being derived for.
// One entry, deliberately: the identity is the one thing about a consumer the mesh itself chose,
// so it is the one thing the mesh can hand to a provider without either end guessing.
var consumerFacts = []string{"as"}
// consumerAlphabets are the ways the mesh will write that identity. `dns` is the mesh's own
// identifier with its separator written `-` instead of `_` — the whole of the difference between
// the alphabet the mesh mints in and the one buckets, vhosts and hostnames accept.
var consumerAlphabets = []string{"dns"}
// ServedTo fills a provider's served values for one consumer.
//
// `as` is the identity the mesh minted for that consumer — the same string it is told to present
// as a login. Values with no placeholder are returned exactly as they were, and a block with no
// placeholder at all is returned unchanged, so this costs nothing for the providers that derive
// nothing.
//
// Only strings carry placeholders. A number, a boolean or a nested object is a value the provider
// stated outright, and is left alone.
func ServedTo(serves map[string]any, as string) (map[string]any, error) {
if len(serves) == 0 {
return serves, nil
}
var out map[string]any
for _, key := range sortedAnyKeys(serves) {
text, ok := serves[key].(string)
if !ok || !strings.Contains(text, "${consumer:") {
continue
}
filled, err := consumerInto(text, as)
if err != nil {
return nil, fmt.Errorf("the value served as %q: %w", key, err)
}
if out == nil {
// Copied only once something actually changes: the caller's map is the manifest's,
// and a provider that derives nothing must not have it rewritten underneath it.
out = make(map[string]any, len(serves))
for k, v := range serves {
out[k] = v
}
}
out[key] = filled
}
if out == nil {
return serves, nil
}
return out, nil
}
// consumerInto replaces every `${consumer:…}` in one value.
//
// **A fact or an alphabet the mesh does not have is refused, not left standing.** Written through,
// the literal `${consumer:as}` would reach a configuration file and be read as a bucket name,
// failing somewhere that names neither the module nor the mesh — the same reasoning `${bound:…}`
// is refused by (boundInto).
func consumerInto(value, as string) (string, error) {
var failed error
out := consumerFact.ReplaceAllStringFunc(value, func(match string) string {
parts := consumerFact.FindStringSubmatch(match)
fact, alphabet := parts[1], parts[2]
if fact != "as" {
if failed == nil {
failed = fmt.Errorf(
"says %s, and the mesh states %s about a consumer", match, orNothing(consumerFacts))
}
return match
}
switch alphabet {
case "":
return as
case "dns":
return asDNSLabel(as)
default:
if failed == nil {
failed = fmt.Errorf(
"says %s, and the mesh writes an identity as %s", match, orNothing(consumerAlphabets))
}
return match
}
})
if failed != nil {
return "", failed
}
return out, nil
}
// asDNSLabel writes a minted identity as a DNS label.
//
// The mesh's identities are already lower-case letters, digits and `_` (ConsumerIdentity), and
// already short enough for the tightest backend they reach (CheckIdentity, twenty characters). So
// this is the separator and nothing else — no lower-casing of what is already lower case, no
// truncation to a limit the identity is already inside, no padding of a name that is already long
// enough. Each of those would be the mesh guessing at a rule it has not been given.
func asDNSLabel(as string) string {
return strings.ReplaceAll(as, "_", "-")
}
// CheckServes refuses a `serves` block that names a consumer fact or an alphabet the mesh does not
// have, when the definition is parsed rather than when a consumer is resolved.
//
// A provision nobody consumes yet still has its rule read: a definition that would be refused the
// first time somebody required it is a definition that is wrong now.
func CheckServes(m Manifest) []string {
var problems []string
for _, provision := range sortedServes(m.Serves) {
for _, key := range sortedAnyKeys(m.Serves[provision]) {
text, ok := m.Serves[provision][key].(string)
if !ok {
continue
}
// A probe identity, because what is checked is the shape of the statement and not
// what any consumer is called.
if _, err := consumerInto(text, "mesh_node_module"); err != nil {
problems = append(problems, fmt.Sprintf(
"%s serves %s, and the value it serves as %q %s", m.Module, provision, key, err))
}
}
}
return problems
}
func sortedServes(serves map[string]map[string]any) []string {
out := make([]string, 0, len(serves))
for k := range serves {
out = append(out, k)
}
sort.Strings(out)
return out
}
func sortedAnyKeys(values map[string]any) []string {
out := make([]string, 0, len(values))
for k := range values {
out = append(out, k)
}
sort.Strings(out)
return out
}
// derivedFor is what the provider on this machine derives for one consumer of one provision
// (novox/hq ADR 0188).
//
// Settled first, then derived: an operator may set a prefix on what the provider serves and the
// mesh still fills the consumer's half of it ([ADR 0174]). Only the keys that actually name the
// consumer are returned — the rest of a `serves` block is the same for every consumer and is
// already in the provider's own definition, so repeating it here would be a second copy to go
// stale.
//
// The first module in the resolved order that says it serves the provision answers, which is the
// choice servedOnThisMachine makes for the consumer's half. Nothing serving it on this machine is
// not an error: a contribution can reach a machine whose provider is a record or an adapter, and
// then there is nothing derived to tell.
func (r Resolution) derivedFor(provision, as string, settings SettingsBy) (map[string]any, error) {
for _, m := range r.Modules {
serves, said := m.Serves[provision]
if !said {
continue
}
var names map[string]any
for key, value := range serves {
if text, ok := value.(string); ok && strings.Contains(text, "${consumer:") {
if names == nil {
names = map[string]any{}
}
names[key] = value
}
}
if names == nil {
return nil, nil
}
settled, err := Settle(names, settings[m.Module])
if err != nil {
return nil, fmt.Errorf("%s serving %s: %w", m.Module, provision, err)
}
derived, err := ServedTo(settled, as)
if err != nil {
return nil, fmt.Errorf("%s serving %s to %s: %w", m.Module, provision, as, err)
}
return derived, nil
}
return nil, nil
}
// notTranscribed refuses a consumer's file that writes out the value its provider derives for it,
// instead of asking for it (novox/hq ADR 0188, issue 124).
//
// **What would have caught the one wrong instance.** The object store's three consumers each wrote
// their bucket into their own configuration by hand. One of them named a predecessor's bucket, and
// nothing compared it to what the provider would actually create: the module would have
// authenticated successfully and been refused on every object, which reads like a credential fault
// and is not one. It looked authoritative for months.
//
// The test is exact and costs one string search: a definition whose file already contains the
// value the mesh is about to derive for it has written down somebody else's rule. It cannot be a
// coincidence — a derived value carries the identity the mesh minted for this very consumer on
// this very machine, which nothing else would spell out — and it cannot be checked afterwards,
// because after substitution every consumer's file contains it legitimately.
//
// Only values that actually name the consumer are judged. A provider that serves a constant under
// the same key serves the same constant to everyone, and a consumer repeating it is redundant
// rather than wrong.
func notTranscribed(resource map[string]any, known map[string]map[string]string, module string) error {
if fmt.Sprint(resource["type"]) != "file" {
return nil
}
content, ok := resource["content"].(string)
if !ok || content == "" {
return nil
}
for _, provision := range sortedKnown(known) {
values := known[provision]
identity := values["as"]
if identity == "" {
continue
}
for _, key := range sortedStringKeys(values) {
if key == "as" {
// The login is not derived from itself, and a consumer that must present it in a
// connection string legitimately has it from `${bound:…}` — which is what it will
// be after substitution, so this would judge the substitution, not the module.
continue
}
value := values[key]
if value == "" || !namesTheConsumer(value, identity) {
continue
}
if !strings.Contains(content, value) {
continue
}
return fmt.Errorf(
"%s writes %q into %v, and that is exactly what %s derives for it — a definition "+
"keeping its own copy of somebody else's naming rule is one that can disagree "+
"with it, silently. Say ${bound:%s:%s} and be told",
module, value, resource["id"], provision, provision, key)
}
}
return nil
}
// namesTheConsumer is whether a derived value was built from this consumer's identity — in the
// alphabet it was minted in, or as a DNS label. A value that does not contain it was not derived
// from it, whatever else it may be.
func namesTheConsumer(value, identity string) bool {
return strings.Contains(value, identity) || strings.Contains(value, asDNSLabel(identity))
}
func sortedKnown(known map[string]map[string]string) []string {
out := make([]string, 0, len(known))
for k := range known {
out = append(out, k)
}
sort.Strings(out)
return out
}
func sortedStringKeys(values map[string]string) []string {
out := make([]string, 0, len(values))
for k := range values {
out = append(out, k)
}
sort.Strings(out)
return out
}
+111 -8
View File
@@ -249,13 +249,32 @@ type Composed struct {
Resources []map[string]any
Owner map[string]string
Received map[string]map[string][]Contribution
// LeftOut is every module of this machine's set that was left out of its declaration, and
// why (novox/hq ADR 0163, rule 6): a setting stored for it that its definition can no longer
// compose. Its held things are kept and its containers untouched — the machine is told so —
// and it is told everything else.
LeftOut map[string]string
}
// LeftOut is which of this machine's modules a declaration composed with these settings leaves
// out, and why (novox/hq ADR 0163, rule 6): each whose stored settings its definition can no longer
// compose. Empty when every module composes. The same judgement SetSettings makes before storing.
func (r Resolution) LeftOut(settings SettingsBy, adopted bool) map[string]string {
out := map[string]string{}
for _, m := range r.Modules {
if err := JudgeSettings(m, settings[m.Module], adopted); err != nil {
out[m.Module] = err.Error()
}
}
return out
}
// Compose is Declaration with the owner of every resource said.
func (r Resolution) Compose(with Rendering) (Composed, error) {
owner := map[string]string{}
received := map[string]map[string][]Contribution{}
resources, err := r.compose(with, owner, received)
leftOut := map[string]string{}
resources, err := r.compose(with, owner, received, leftOut)
if err != nil {
return Composed{}, err
}
@@ -267,7 +286,7 @@ func (r Resolution) Compose(with Rendering) (Composed, error) {
"sealed": with.BusMembership, "mode": "0600",
})
}
return Composed{Resources: resources, Owner: owner, Received: received}, nil
return Composed{Resources: resources, Owner: owner, Received: received, LeftOut: leftOut}, nil
}
// BusMembershipID names the resource carrying a machine's membership for the new bus, and
@@ -277,7 +296,25 @@ func BusMembershipID() string { return "bus-membership" }
const BusMembershipPath = "/var/lib/mesh/membership-next.json"
func (r Resolution) compose(with Rendering, owner map[string]string,
received map[string]map[string][]Contribution) ([]map[string]any, error) {
received map[string]map[string][]Contribution, leftOut map[string]string) ([]map[string]any, error) {
// **A setting is judged where it is stored, and an impossible one costs a module, not a
// machine** (novox/hq ADR 0163, rule 6). A definition that moved under a stored setting makes
// this module uncomposable; it is left out of the declaration — its held things kept, its
// containers untouched, the machine told so by name — and the machine is told everything else.
// Before placing, because a placement is a setting too.
left := r.LeftOut(with.Settings, with.Adopted)
kept := make([]Manifest, 0, len(r.Modules))
for _, m := range r.Modules {
if why, isLeft := left[m.Module]; isLeft {
if leftOut != nil {
leftOut[m.Module] = why
}
continue
}
kept = append(kept, m)
}
r.Modules = kept
// Every manifest is placed first (novox/hq ADR 0112): the maps naming where its bindings,
// credentials and contributions land are resolved against this node's directories, so every
// reader below — the binding files, the sealed secrets, the grant paths a contribution
@@ -591,7 +628,16 @@ func (r Resolution) compose(with Rendering, owner map[string]string,
if err != nil {
return nil, err
}
file, err := boundFile(*found, m.Binds[to], ConsumerIdentity(r.Node, IdentitySource(m.Slug, m.Module)), own)
as := ConsumerIdentity(r.Node, IdentitySource(m.Slug, m.Module))
// What the provider derives for THIS consumer, filled here where the consumer is
// known (novox/hq ADR 0188). The same fill knownFor does below, so the binding file
// and the module's `${bound:…}` substitutions cannot say different things.
told := *found
told.Serves, err = ServedTo(told.Serves, as)
if err != nil {
return nil, fmt.Errorf("%s is told about %s: %w", m.Module, to, err)
}
file, err := boundFile(told, m.Binds[to], as, own)
if err != nil {
return nil, err
}
@@ -657,7 +703,10 @@ func (r Resolution) compose(with Rendering, owner map[string]string,
return nil, err
}
// And what its bindings say, for the half of a connection that is not secret.
known := knownFor(m, r.Needs, r.Node)
known, err := knownFor(m, r.Needs, r.Node)
if err != nil {
return nil, err
}
// A requirement answered on this same machine is not in r.Needs — its binding file is
// written from `here` (above) — and so `${bound:…}` could not name it, though the file
// beside it said the same facts. Filled from the same answer, so the two cannot disagree.
@@ -674,7 +723,11 @@ func (r Resolution) compose(with Rendering, owner map[string]string,
}
local := *answered
local.For = m.Module
for provision, values := range knownFor(m, []Needed{local}, r.Node) {
here, err := knownFor(m, []Needed{local}, r.Node)
if err != nil {
return nil, err
}
for provision, values := range here {
known[provision] = values
}
}
@@ -699,6 +752,17 @@ func (r Resolution) compose(with Rendering, owner map[string]string,
// And the machine underneath, which no binding of its own can tell it.
thisMachine := machineFacts(r, with.Names, with.MeshRange)
// **A definition that already holds the answer transcribed it** (novox/hq ADR 0188).
// Judged over what the module itself declares, and before anything is substituted: the
// mesh's own generated files — the binding, the contributions — legitimately carry the
// derived value, and after substitution so does every consumer's file, so this is the one
// moment the two can be told apart.
for _, own := range m.Resources {
if err := notTranscribed(own, known, m.Module); err != nil {
return nil, err
}
}
// Which of this module's files carry a secret, for the rule that a container may not read
// one of them as its environment without saying so (ADR 0086, issue 041).
secretFiles := secretFilesOf(resources)
@@ -706,6 +770,10 @@ func (r Resolution) compose(with Rendering, owner map[string]string,
// Which of this module's resources its preparation runs before, if it prepares anything.
prepareBefore := preparationTarget(m)
// Which found networks this machine's setting keeps for each of its containers (novox/hq
// ADR 0163, rule 4); judged above, so an invalid one is not here.
keptNetworks, _ := KeptNetworks(m, with.Settings[m.Module], with.Adopted)
for _, unsettled := range resources {
resource, err := ApplySettings(unsettled, with.Settings[m.Module])
if err != nil {
@@ -715,6 +783,16 @@ func (r Resolution) compose(with Rendering, owner map[string]string,
for k, v := range resource {
copied[k] = v
}
if networks, keeps := keptNetworks[fmt.Sprint(copied["id"])]; keeps {
// The container also joins the found network the setting names, so a neighbour
// that resolves it there keeps resolving it. Passed to the host as its own field,
// which it joins after the container is made.
joins := make([]any, 0, len(networks))
for _, n := range networks {
joins = append(joins, n)
}
copied["networks"] = joins
}
if err := refuseSecretsInEnvironment(copied, secretFiles, m.Module); err != nil {
return nil, err
}
@@ -960,8 +1038,15 @@ func (r Resolution) filtersHere() string {
// computed for this machine, and each module's per-node exposure. The same answer whether the node
// is adopted or converged — the one loads it as a filter, the other declares it as openings.
func (r Resolution) Rules(with Rendering) ([]Rule, error) {
// A module whose settings cannot compose is left out of the declaration (novox/hq ADR 0163,
// rule 6), and out of the filter with it: nothing of it is declared, so nothing of it is let
// through.
left := r.LeftOut(with.Settings, with.Adopted)
exposure := map[string]map[int]string{}
for _, m := range r.Modules {
if _, isLeft := left[m.Module]; isLeft {
continue
}
e, err := Exposure(m, with.Settings[m.Module])
if err != nil {
return nil, err
@@ -1042,6 +1127,19 @@ type Contribution struct {
// requirement's name — everything providing `reverse-proxy` understands the same shape, which
// is what makes swapping one for another cost nothing.
Values map[string]any `json:"values"`
// Derived is what this provider's own definition said it derives for this consumer, already
// derived (novox/hq ADR 0188).
//
// **The provider is told, rather than recomputing it.** A served value may name the consumer's
// identity — a bucket named for who is asking, a database prefixed with it — and before this
// the rule lived twice: once in the provisioner's code, once transcribed into every consumer's
// definition. The mesh fills the provider's own statement here and delivers the same filled
// value to the consumer, so the two cannot disagree: there is no second computation to
// disagree with.
//
// Only the keys that are per-consumer. The rest of what the provider serves is the same for
// everyone and is in its own definition, where it already is.
Derived map[string]any `json:"derived,omitempty"`
}
// grantPath is where one consumer's sealed credential lands on the providing machine.
@@ -1133,12 +1231,17 @@ func (r Resolution) contributions(settings SettingsBy, grants []Grant,
// told about it and withdraws the login on its next pass.
continue
}
as := holderAs(ConsumerIdentity(g.Consumer, IdentitySource(g.Slug, g.From)), g.Local)
derived, err := r.derivedFor(g.Provision, as, settings)
if err != nil {
return nil, err
}
out[g.Provision] = append(out[g.Provision], Contribution{
From: g.From, Node: g.Consumer, At: g.At, Values: g.Values,
From: g.From, Node: g.Consumer, At: g.At, Values: g.Values, Derived: derived,
// One holder per local name: the identity the consumer is known by, and the local name
// after it where the module keeps several (ADR 0094). Not a login any backend checks —
// a secret is not a login — so the identity limit does not apply to the suffix.
As: holderAs(ConsumerIdentity(g.Consumer, IdentitySource(g.Slug, g.From)), g.Local),
As: as,
Secret: grantPath(directories[g.Provision], g.Consumer, holderAs(g.From, g.Local)),
})
if granted[g.Provision] == nil {
@@ -0,0 +1,297 @@
package catalogue
import (
"encoding/json"
"strings"
"testing"
)
// What a provider derives for each consumer, said once and delivered to both ends
// (novox/hq ADR 0188, issue 124).
//
// The failure these are written against: the object store's provisioner derived each consumer's
// bucket from the login the mesh minted, in its own code, and the mesh had no channel to tell the
// consumer which bucket that was — so all three consumers wrote the answer into their own
// definitions by hand. Two were right. One named a predecessor's bucket and would have
// authenticated successfully and been refused on every object. Each of them also named the
// machine the module happens to run on, which a definition may not do.
// store is an object store in the shape minio has: it serves a region and a port to everyone, and
// a bucket named for whoever is asking.
func store() Manifest {
return Manifest{
Module: "store", Version: "1",
Provides: FromAnywhere("s3-bucket"),
Listens: []Listening{{Port: 9000, Protocol: "tcp", From: FromMesh}},
Serves: map[string]map[string]any{"s3-bucket": {
"region": "eu-west",
"bucket": "${consumer:as:dns}",
}},
Receives: map[string]string{"s3-bucket": "/var/lib/store/grants/mesh.json"},
Grants: map[string]string{"s3-bucket": "/var/lib/store/grants"},
Resources: []map[string]any{{
"id": "server", "type": "container", "name": "store", "ports": []any{"9000"},
}},
}
}
// files is a consumer that writes the bucket into its own configuration — which is the thing it
// could not do before, and had to transcribe.
func files() Manifest {
return Manifest{
Module: "files", Version: "1", Slug: "files",
Requires: []string{"s3-bucket"},
Binds: map[string]string{"s3-bucket": "/var/lib/files/store.json"},
Secrets: map[string]string{"s3-bucket": "/var/lib/files/store.secret"},
Resources: []map[string]any{{
"id": "env", "type": "file", "path": "/var/lib/files/env", "mode": "0600",
"content": "BUCKET=${bound:s3-bucket:bucket}\nREGION=${bound:s3-bucket:region}\n",
}},
}
}
// pics is a second consumer of the same provider on the same machine: two derivations, neither
// the other's.
func pics() Manifest {
return Manifest{
Module: "pics", Version: "1", Slug: "pics",
Requires: []string{"s3-bucket"},
Binds: map[string]string{"s3-bucket": "/var/lib/pics/store.json"},
Secrets: map[string]string{"s3-bucket": "/var/lib/pics/store.secret"},
Resources: []map[string]any{{
"id": "env", "type": "file", "path": "/var/lib/pics/env", "mode": "0600",
"content": "BUCKET=${bound:s3-bucket:bucket}\n",
}},
}
}
// The three places the derived value lands must agree, because agreeing is the whole point: the
// consumer's own file, the binding it reads as JSON, and the provider's contributions entry.
func TestADerivedValueReachesBothEndsAndAgrees(t *testing.T) {
r, err := Resolve(shelf(store(), files()), []string{"store", "files"}, reachable(), World{})
if err != nil {
t.Fatal(err)
}
out, err := r.Declaration(Rendering{Grants: []Grant{{
Provision: "s3-bucket", Consumer: "workstation", From: "files", Slug: "files",
Values: map[string]any{}, Sealed: "c2VhbGVk",
}}})
if err != nil {
t.Fatal(err)
}
// The mesh minted this identity for the consumer; the bucket is that identity as a DNS label.
// Derived here with the mesh's own function, so the test cannot agree with a wrong rule.
as := ConsumerIdentity("workstation", IdentitySource("files", "files"))
want := strings.ReplaceAll(as, "_", "-")
if want == as || !strings.Contains(as, "_") {
t.Fatalf("the mesh's identity %q has no separator to rewrite; this test proves nothing", as)
}
env := fileNamed(out, "files.env")
if env == nil {
t.Fatalf("the consumer was given no file: %v", out)
}
if got := env["content"].(string); !strings.Contains(got, "BUCKET="+want+"\n") {
t.Errorf("the consumer's own file was not told the bucket:\n%s\nwant BUCKET=%s", got, want)
}
binding := fileNamed(out, "files.bound-s3-bucket")
if binding == nil {
t.Fatalf("the consumer was given no binding: %v", out)
}
var said struct {
Serves map[string]any `json:"serves"`
}
if err := json.Unmarshal([]byte(binding["content"].(string)), &said); err != nil {
t.Fatal(err)
}
if said.Serves["bucket"] != want {
t.Errorf("the binding says the bucket is %q, want %q", said.Serves["bucket"], want)
}
// And what is the same for everybody is still the same for everybody.
if said.Serves["region"] != "eu-west" {
t.Errorf("the binding lost what the provider serves to all: %v", said.Serves)
}
given := storeGrants(t, out)
if len(given) != 1 {
t.Fatalf("the provider was told about %d consumer(s): %v", len(given), given)
}
if given[0].Derived["bucket"] != want {
t.Errorf("the provider was told the bucket is %v, and the consumer was told %q — "+
"the two ends disagree, which is the whole failure", given[0].Derived["bucket"], want)
}
// Only the per-consumer half. The region is the same for everyone and is already in the
// provider's own definition; repeating it here would be a copy to go stale.
if _, carried := given[0].Derived["region"]; carried {
t.Errorf("the provider was handed back what it already says for everyone: %v", given[0].Derived)
}
}
// Two consumers of one provider on one machine get two buckets, and neither gets the other's.
func TestTwoConsumersOfOneProviderGetTheirOwnDerivation(t *testing.T) {
r, err := Resolve(shelf(store(), files(), pics()),
[]string{"store", "files", "pics"}, reachable(), World{})
if err != nil {
t.Fatal(err)
}
out, err := r.Declaration(Rendering{Grants: []Grant{
{Provision: "s3-bucket", Consumer: "workstation", From: "files", Slug: "files",
Values: map[string]any{}, Sealed: "c2VhbGVk"},
{Provision: "s3-bucket", Consumer: "workstation", From: "pics", Slug: "pics",
Values: map[string]any{}, Sealed: "c2VhbGVk"},
}})
if err != nil {
t.Fatal(err)
}
forFiles := strings.ReplaceAll(ConsumerIdentity("workstation", IdentitySource("files", "files")), "_", "-")
forPics := strings.ReplaceAll(ConsumerIdentity("workstation", IdentitySource("pics", "pics")), "_", "-")
if forFiles == forPics {
t.Fatal("the two consumers were given the same identity; this test proves nothing")
}
if got := fileNamed(out, "files.env")["content"].(string); !strings.Contains(got, "BUCKET="+forFiles+"\n") {
t.Errorf("files was not given its own bucket:\n%s", got)
}
if got := fileNamed(out, "pics.env")["content"].(string); !strings.Contains(got, "BUCKET="+forPics+"\n") {
t.Errorf("pics was not given its own bucket:\n%s", got)
}
var buckets []any
for _, g := range storeGrants(t, out) {
buckets = append(buckets, g.Derived["bucket"])
}
if len(buckets) != 2 || buckets[0] == buckets[1] {
t.Errorf("the provider was told %v; it must be told one bucket per consumer", buckets)
}
}
// An operator may still set what the provider serves, and the mesh still derives the rest: the
// setting is laid on first, then the consumer's half is filled.
func TestASettingComposesWithADerivedValue(t *testing.T) {
r, err := Resolve(shelf(store(), files()), []string{"store", "files"}, reachable(), World{})
if err != nil {
t.Fatal(err)
}
out, err := r.Declaration(Rendering{
Settings: SettingsBy{"store": {{From: "the operator",
Values: map[string]any{"bucket": "team-${consumer:as:dns}"}}}},
Grants: []Grant{{Provision: "s3-bucket", Consumer: "workstation", From: "files", Slug: "files",
Values: map[string]any{}, Sealed: "c2VhbGVk"}},
})
if err != nil {
t.Fatal(err)
}
want := "team-" + strings.ReplaceAll(ConsumerIdentity("workstation", IdentitySource("files", "files")), "_", "-")
if got := fileNamed(out, "files.env")["content"].(string); !strings.Contains(got, "BUCKET="+want+"\n") {
t.Errorf("the operator's prefix did not survive the derivation:\n%s\nwant BUCKET=%s", got, want)
}
if given := storeGrants(t, out); given[0].Derived["bucket"] != want {
t.Errorf("the provider was told %v, the consumer %q", given[0].Derived["bucket"], want)
}
}
// A fact or an alphabet the mesh does not have is refused where the definition is, not where a
// consumer happens to be resolved — and the refusal says what may be said instead.
func TestAServedValueNamingSomethingTheMeshDoesNotHaveIsRefused(t *testing.T) {
for _, c := range []struct{ value, says string }{
{"${consumer:node}", "as"},
{"${consumer:as:punycode}", "dns"},
} {
m := store()
m.Serves["s3-bucket"]["bucket"] = c.value
raw, err := json.Marshal(m)
if err != nil {
t.Fatal(err)
}
_, err = ParseManifest(raw)
if err == nil {
t.Fatalf("%s was accepted", c.value)
}
if !strings.Contains(err.Error(), c.value) {
t.Errorf("the refusal of %s does not quote it: %v", c.value, err)
}
if !strings.Contains(err.Error(), c.says) {
t.Errorf("the refusal of %s does not say what may be said (%q): %v", c.value, c.says, err)
}
}
}
// `dns` is checked against an identity the mesh actually mints, not an invented string.
func TestTheDNSAlphabetIsTheMintedIdentityWithItsSeparatorRewritten(t *testing.T) {
as := ConsumerIdentity("anchor", IdentitySource("ncloud", "nextcloud"))
if err := CheckIdentity("anchor", IdentitySource("ncloud", "nextcloud")); err != nil {
t.Fatalf("the mesh would not mint this identity at all: %v", err)
}
label := asDNSLabel(as)
if strings.Contains(label, "_") {
t.Errorf("%q is not a DNS label", label)
}
if strings.ReplaceAll(label, "-", "_") != as {
t.Errorf("%q is not %q with its separator rewritten", label, as)
}
}
// The check that would have caught the one wrong instance: a consumer that writes the derived
// value into its own definition instead of asking for it is refused, whether it transcribed the
// right answer or a predecessor's.
func TestAConsumerThatTranscribesWhatItsProviderDerivesIsRefused(t *testing.T) {
as := ConsumerIdentity("workstation", IdentitySource("files", "files"))
transcribed := strings.ReplaceAll(as, "_", "-")
m := files()
m.Resources = []map[string]any{{
"id": "env", "type": "file", "path": "/var/lib/files/env", "mode": "0600",
// Exactly what the provider will create — correct today, and a copy of a rule that is
// not this module's.
"content": "BUCKET=" + transcribed + "\n",
}}
r, err := Resolve(shelf(store(), m), []string{"store", "files"}, reachable(), World{})
if err != nil {
t.Fatal(err)
}
_, err = r.Declaration(Rendering{Grants: []Grant{{
Provision: "s3-bucket", Consumer: "workstation", From: "files", Slug: "files",
Values: map[string]any{}, Sealed: "c2VhbGVk",
}}})
if err == nil {
t.Fatal("a definition holding its own copy of the provider's naming rule was accepted")
}
if !strings.Contains(err.Error(), "${bound:s3-bucket:bucket}") {
t.Errorf("the refusal does not say what to write instead: %v", err)
}
// And a constant the provider serves to everyone is not a transcription: repeating it is
// redundant, not wrong, and refusing it would be the mesh policing style.
m.Resources = []map[string]any{{
"id": "env", "type": "file", "path": "/var/lib/files/env", "mode": "0600",
"content": "REGION=eu-west\n",
}}
r, err = Resolve(shelf(store(), m), []string{"store", "files"}, reachable(), World{})
if err != nil {
t.Fatal(err)
}
if _, err := r.Declaration(Rendering{Grants: []Grant{{
Provision: "s3-bucket", Consumer: "workstation", From: "files", Slug: "files",
Values: map[string]any{}, Sealed: "c2VhbGVk",
}}}); err != nil {
t.Errorf("a value the provider serves to everyone was judged a transcription: %v", err)
}
}
func storeGrants(t *testing.T, out []map[string]any) []Contribution {
t.Helper()
for _, r := range out {
if r["path"] != "/var/lib/store/grants/mesh.json" {
continue
}
var parsed struct {
Given []Contribution `json:"given"`
}
if err := json.Unmarshal([]byte(r["content"].(string)), &parsed); err != nil {
t.Fatal(err)
}
return parsed.Given
}
t.Fatalf("the provider was given no contributions file: %v", out)
return nil
}
+12
View File
@@ -444,6 +444,18 @@ func AsNftables(rules []Rule, mesh []string, outward bool, foundation []int,
b.WriteString("\t\t# this machine's own guests reaching outward: not a port opened to anybody\n")
b.WriteString(fmt.Sprintf("\t\tiifname != { %s } accept\n", inward))
}
// **The mesh passing through, not arriving.** A machine the mesh routes through — the hub, for
// every path between machines that are not co-located (novox/hq ADR 0007) — relays a packet that
// came in on the tunnel and leaves on it again, addressed to another machine of the mesh. That is
// no port of this machine's: the machine it is for filters it against its own rules. Without
// this, the chain below judged a relayed packet by this machine's own published ports, so two
// machines behind the hub reached each other only on ports the hub happened to publish for itself
// (novox/hq issue 196). In and out on the tunnel both: a packet off the tunnel for this machine's
// own containers leaves by a bridge, and still meets the rules below.
if tunnel != "" {
b.WriteString("\t\t# the mesh passing through to another of its machines, which filters it itself\n")
b.WriteString(fmt.Sprintf("\t\tiifname %q oifname %q accept\n", tunnel, tunnel))
}
if len(rules) > 0 {
b.WriteString("\n")
+31
View File
@@ -887,3 +887,34 @@ func TestAPublicPortNeedsNoGuestLine(t *testing.T) {
t.Fatalf("a public port was given a guest line it does not need:\n%s", nft)
}
}
// **The hub relays the mesh** (novox/hq ADR 0007, issue 196). Two machines that are not co-located
// reach each other through the hub, so the hub forwards a packet that arrives on the tunnel and
// leaves on it. The forward chain judged that packet by the hub's own published ports, and two
// machines behind the hub reached each other only on the ports the hub happened to publish.
//
// Measured: from one home machine to another through the hub, 17 of 55 ports answered, and they
// were exactly the hub's own; the SYN for the rest never left the hub.
func TestTheMeshPassingThroughIsRelayedNotJudgedAsThisMachines(t *testing.T) {
nft := AsNftables(nil, []string{"10.42.0.1", "10.42.0.2"}, false, nil, []string{"eth0"}, "mesh0")
relay := `iifname "mesh0" oifname "mesh0" accept`
if !strings.Contains(chainBody(t, nft, "forward"), relay) {
t.Errorf("the forward chain does not relay the mesh through this machine:\n%s", chainBody(t, nft, "forward"))
}
// Relaying is not receiving: nothing in the input chain opens because of it.
if strings.Contains(chainBody(t, nft, "input"), "oifname") {
t.Errorf("the input chain names an outgoing interface, which no packet for this machine has:\n%s",
chainBody(t, nft, "input"))
}
// And off the tunnel into this machine's own containers is still judged: the tunnel is not
// accepted wholesale, only in and out on it.
if strings.Contains(chainBody(t, nft, "forward"), `iifname "mesh0" accept`) {
t.Errorf("the forward chain accepts everything off the tunnel:\n%s", chainBody(t, nft, "forward"))
}
// A machine with no tunnel relays nothing, and names no interface it does not have.
alone := AsNftables(nil, nil, false, nil, []string{"eth0"}, "")
if strings.Contains(alone, "oifname") {
t.Errorf("a machine with no tunnel was given a relay rule:\n%s", alone)
}
}
+5 -2
View File
@@ -102,8 +102,11 @@ func accountHomeOf(account, home string) string {
func machineInto(resource map[string]any, facts map[string]string, module string) error {
// Content, and now the path and owner too: a module that writes into a person's home names it
// with ${machine:account-home} and ${machine:account}, which it cannot know until assigned
// (novox/hq to-be 29), the same reason its content names ${machine:address}.
for _, field := range []string{"path", "owner", "content"} {
// (novox/hq to-be 29), the same reason its content names ${machine:address}. And the name a
// `user` shape sets the login shell of, and the user a user-scoped unit or a process runs as:
// the shell module makes the operator's account its holder's login shell, and the desktop's
// watchers run as that account (novox/hq ADR 0176, ADR 0177) — neither can name the person.
for _, field := range []string{"path", "owner", "content", "name", "user"} {
s, ok := resource[field].(string)
if !ok {
continue
+126
View File
@@ -1360,6 +1360,10 @@ func ParseManifest(raw []byte) (Manifest, error) {
"%s serves %q to whoever requires it, and does not provide it", m.Module, to))
}
}
// A served value may be derived for the consumer it is served to (novox/hq ADR 0188). Read
// here, where the definition is, rather than when somebody first requires it: a rule that
// would be refused at the first consumer is wrong from the moment it is written.
problems = append(problems, CheckServes(m)...)
for to, where := range m.Binds {
if !placedOrAbsolute(where) {
problems = append(problems, fmt.Sprintf(
@@ -1507,6 +1511,13 @@ func ParseManifest(raw []byte) (Manifest, error) {
}
}
}
// **A scheduled step may hold this module's own containers still while it runs**
// (novox/hq ADR 0189). What the host judges is the declaration it receives — whether each
// id is a container placed on that machine; what belongs here is what only the definition
// shows: that the ids are this module's, that they are containers, and that the step is
// scheduled. A module naming a neighbour's container would be a module that can stop the
// mesh, and the manifest is where that is visible.
problems = append(problems, whileStoppedProblems(m, r, hasSchedule(r))...)
}
for name, own := range m.OwnSecrets {
if !placedOrAbsolute(own.Path) {
@@ -1667,6 +1678,7 @@ func ParseManifest(raw []byte) (Manifest, error) {
problems = append(problems, m.undeclaredMounts()...)
problems = append(problems, m.unknownDirRefs()...)
problems = append(problems, m.unknownAccessRefs()...)
problems = append(problems, m.jailProblems()...)
for i, r := range m.Resources {
id, _ := r["id"].(string)
@@ -1769,6 +1781,46 @@ func (m Manifest) MachineSide(port int) (at int, mayAssign bool) {
var facilitiesOf = map[string][]string{
// Both spellings: /var/run is a link to /run on every machine the mesh runs on.
"container-runtime": {"/var/run/docker.sock", "/run/docker.sock"},
// The virtualisation daemon's socket, for the lab (novox/hq ADR 0172): it raises machines there.
"virtualisation": {"/var/lib/incus/unix.socket"},
}
// jailProblems is every jail this module declares that the machine's intrusion prevention would
// refuse (novox/hq ADR 0179).
//
// **Because one bad pattern stops every jail, not its own.** fail2ban expands `<HOST>` into a named
// capture group, so a pattern naming it twice is a duplicate group name, and the daemon refuses the
// whole configuration and exits — the machine keeps no bans at all, for any jail, including the one
// watching its ssh. Caught live on the control node the day this was built, where a proxy's pattern
// matched two shapes of refusal in one line. A pattern matches one shape; several shapes are several
// patterns, one per line, as fail2ban's own filters are written.
func (m Manifest) jailProblems() []string {
var problems []string
seen := map[string]bool{}
for _, j := range m.Jails {
switch {
case strings.TrimSpace(j.Name) == "":
problems = append(problems, m.Module+" declares a jail with no name")
case seen[j.Name]:
problems = append(problems, m.Module+" declares two jails called "+strconv.Quote(j.Name))
}
seen[j.Name] = true
if strings.TrimSpace(j.Failregex) == "" {
problems = append(problems, m.Module+"'s jail "+strconv.Quote(j.Name)+" says nothing a failed attempt looks like")
}
for _, line := range strings.Split(j.Failregex, "\n") {
if strings.TrimSpace(line) == "" {
continue
}
if n := strings.Count(line, "<HOST>"); n > 1 {
problems = append(problems, fmt.Sprintf("%s's jail %s names <HOST> %d times in one pattern; "+
"fail2ban reads it as one capture group and refuses the whole configuration, so the machine "+
"keeps no bans at all — write one pattern per shape, each naming <HOST> once",
m.Module, strconv.Quote(j.Name), n))
}
}
}
return problems
}
// undeclaredMounts is every bind-mount source no declaration covers — see the check above.
@@ -1810,6 +1862,12 @@ func (m Manifest) undeclaredMounts() []string {
claim(p)
}
}
// The file a filter module's rule set is written to is declared by `filtering.into`: the mesh
// writes it, the module loads it, and the module's runtime may read it back to reload the
// mesh's own table (novox/hq ADR 0170).
if m.Filtering != nil {
claim(m.Filtering.Into)
}
// Under a declared directory is declared: a module that says where its data lives has said so
// for what it puts inside.
covers := func(path string) bool {
@@ -1991,3 +2049,71 @@ func (o OwnSecrets) Paths() map[string]string {
// InstancesInterchangeable is the one value of a definition's `instances`: the module is the same
// on every machine, so any instance may answer for the module.
const InstancesInterchangeable = "interchangeable"
// WhileStopped is the resource key naming the containers a scheduled step holds still while it
// runs (novox/hq ADR 0189). Carried to the host unchanged, like `schedule`.
const WhileStopped = "while-stopped"
// hasSchedule is whether a resource declares a cadence, as a string.
func hasSchedule(r map[string]any) bool {
s, _ := r["schedule"].(string)
return s != ""
}
// whileStoppedProblems judges one container's maintenance window against its own definition
// (novox/hq ADR 0189).
//
// Three things the manifest is the only place to see: that the step is scheduled (a one-time
// offline job says *before* rather than *instead of* — at apply the host already has a window,
// because the declaration is applied in order and a run-once step gates what follows); that every
// id it names is **this module's own** container; and that it does not name itself.
//
// The host checks the fourth — that the container is actually placed on that machine — because
// that is a fact about the declaration and not about the definition.
func whileStoppedProblems(m Manifest, r map[string]any, scheduled bool) []string {
raw, present := r[WhileStopped]
if !present {
return nil
}
ids, ok := raw.([]any)
if !ok {
return []string{fmt.Sprintf(
"%s declares %s on %v as a %T; it is a list of this module's container ids",
m.Module, WhileStopped, r["id"], raw)}
}
var problems []string
if len(ids) > 0 && !scheduled {
problems = append(problems, fmt.Sprintf(
"%s declares %s on %v, which has no schedule. A maintenance window is for a recurring "+
"step: at apply the mesh already has one, because a run-once step gates what is "+
"declared after it (novox/hq ADR 0189)", m.Module, WhileStopped, r["id"]))
}
containers := map[string]bool{}
for _, own := range m.Resources {
if fmt.Sprint(own["type"]) == "container" {
containers[fmt.Sprint(own["id"])] = true
}
}
for _, each := range ids {
id, ok := each.(string)
if !ok {
problems = append(problems, fmt.Sprintf(
"%s declares %s on %v naming a %T; each entry is a container's id",
m.Module, WhileStopped, r["id"], each))
continue
}
if id == fmt.Sprint(r["id"]) {
problems = append(problems, fmt.Sprintf(
"%s declares %s on %v naming itself", m.Module, WhileStopped, r["id"]))
continue
}
if !containers[id] {
problems = append(problems, fmt.Sprintf(
"%s declares %s on %v naming %q, which is not a container this module declares. "+
"A step may hold still its own module's containers and nobody else's — one "+
"that could quiesce a neighbour could stop the mesh",
m.Module, WhileStopped, r["id"], id))
}
}
return problems
}
+10
View File
@@ -98,3 +98,13 @@ func TestAMountOfABoundFactIsAccepted(t *testing.T) {
t.Fatalf("a mount of the file the mesh writes a binding to was refused: %v", err)
}
}
// The file a filter module's rule set is written to is declared by `filtering.into` (novox/hq ADR
// 0169): the module's runtime mounts it to reload the mesh's own table, and nothing else declares it.
func TestAMountOfTheFilterFileIsDeclaredByFilteringInto(t *testing.T) {
_, err := ParseManifest([]byte(`{"module":"nftables","filtering":{"into":"/etc/nftables.conf"},` +
`"resources":[` + strings.Replace(aContainerMounting, "%s", "/etc/nftables.conf", 1) + `]}`))
if err != nil {
t.Fatalf("a filter module mounting its own filter file was refused: %v", err)
}
}
@@ -0,0 +1,60 @@
package catalogue
import (
"strings"
"testing"
)
// A `user` shape and a user-scoped unit name the operator account the way a home file does
// (novox/hq ADR 0176, ADR 0177): with ${machine:account}, resolved when the module is assigned.
func TestAUserShapeAndAUserScopedUnitNameTheAccount(t *testing.T) {
facts := map[string]string{"account": "ops", "account-home": "/home/ops"}
login := map[string]any{"type": "user", "id": "login", "name": "${machine:account}", "shell": "/usr/bin/zsh"}
if err := machineInto(login, facts, "zsh"); err != nil {
t.Fatal(err)
}
if login["name"] != "ops" {
t.Fatalf("the user shape did not learn the account: %v", login["name"])
}
watcher := map[string]any{"type": "service", "id": "watcher", "unit": "i3-reload-watcher.service",
"scope": "user", "user": "${machine:account}"}
if err := machineInto(watcher, facts, "i3"); err != nil {
t.Fatal(err)
}
if watcher["user"] != "ops" {
t.Fatalf("the user-scoped unit did not learn the account: %v", watcher["user"])
}
// A machine with no operator account refuses rather than writing the literal.
err := machineInto(map[string]any{"type": "user", "id": "login", "name": "${machine:account}"},
map[string]string{"address": "10.0.0.1"}, "zsh")
if err == nil || !strings.Contains(err.Error(), "${machine:account}") {
t.Fatalf("a user shape on a machine with no account was not refused by name: %v", err)
}
}
// The service manager is a seat of the mesh's own with the unit verbs as its contract (novox/hq
// ADR 0177): every verb described, with a schema, taking a scope.
func TestTheServiceManagerSeatServesTheUnitVerbs(t *testing.T) {
seat, ok := SeatNamed("node-service-manager")
if !ok {
t.Fatal("node-service-manager is not a seat the mesh defines")
}
if seat.Scope != ScopeNode {
t.Fatalf("the service manager is a role each machine has once, and the seat is %s-scoped", seat.Scope)
}
want := []string{"units", "status", "start", "stop", "restart", "enable", "disable", "journal"}
var got []string
for _, v := range seat.Serves {
got = append(got, v.Name)
if v.Description == "" || v.Input == nil {
t.Fatalf("%s is promised without a description or a schema", v.Name)
}
props, _ := v.Input["properties"].(map[string]any)
if _, has := props["scope"]; !has {
t.Fatalf("%s takes no scope, and a user unit could not be asked for", v.Name)
}
}
if strings.Join(got, ",") != strings.Join(want, ",") {
t.Fatalf("the seat serves %v, not %v", got, want)
}
}
+14 -5
View File
@@ -75,17 +75,26 @@ func TestAnAssignmentPlacesDirectoriesAndAccesses(t *testing.T) {
}
// An access declared by id and placed by nobody resolves to nowhere, and that is refused with the
// setting to write — not mounted as the literal, not skipped.
// setting to write — not mounted as the literal, not skipped. The refusal costs the module its
// place in the declaration, not the machine its declaration (novox/hq ADR 0163, rule 6).
func TestAnUnplacedAccessIsRefusedByName(t *testing.T) {
got, err := Resolve(shelf(placeable()), []string{"arr"}, workstation(), World{})
if err != nil {
t.Fatal(err)
}
_, err = got.Declaration(placedBy(map[string]any{
with := placedBy(map[string]any{
AccessesSetting: map[string]any{"series": "/storage/media/series"},
}))
if err == nil || !strings.Contains(err.Error(), `"spool"`) || !strings.Contains(err.Error(), AccessesSetting) {
t.Fatalf("an access nobody placed was not refused by name: %v", err)
})
composed, err := got.Compose(with)
if err != nil {
t.Fatal(err)
}
why := composed.LeftOut["arr"]
if why == "" || !strings.Contains(why, `"spool"`) || !strings.Contains(why, AccessesSetting) {
t.Fatalf("an access nobody placed was not refused by name: %v", composed.LeftOut)
}
if _, declared := byID(composed.Resources)["arr.server"]; declared {
t.Fatal("the module with the unplaced access was declared anyway")
}
}
+16 -6
View File
@@ -48,7 +48,9 @@ func TestTheResolverForwardsToFixedUpstreamsAndNeverReadsResolvConf(t *testing.T
}
for _, want := range []string{
"\nno-resolv\n", "\nserver=1.1.1.1\n", "\nserver=8.8.8.8\n",
"\nlisten-address=127.0.0.1\n", "\nlisten-address=${machine:address}\n", "\nbind-dynamic\n",
// Loopback is the mesh-wide setting's default; a machine answering its own LAN adds its
// address there (novox/hq issue 198).
"\nlisten-address=${setting:listen-addresses}\n", "\nlisten-address=${machine:address}\n", "\nbind-dynamic\n",
"\ndomain-needed\n", "\nbogus-priv\n",
"\nconf-file=" + m.Facts["node-zones"].Path + "\n",
} {
@@ -113,7 +115,8 @@ func TestTheResolverAndWhatAsksItComposeOnOneMachine(t *testing.T) {
// issue 111) — the resolver's zones read only the second, and in this scenario the two
// happen to be the same map, since nothing routed is part of it.
Names: twoMachines, Machines: twoMachines, Suffix: "internal",
Needed: map[string]map[string]string{"dnsmasq": {"broker": "sealed"}},
Needed: map[string]map[string]string{"dnsmasq": {"broker": "sealed"}},
Settings: SettingsBy{"dnsmasq": {{From: "the mesh", Values: map[string]any{"listen-addresses": "127.0.0.1"}}}},
})
if err != nil {
t.Fatal(err)
@@ -207,9 +210,16 @@ func TestTheResolverOnAMachineOffTheNetworkIsRefused(t *testing.T) {
if err != nil {
t.Fatal(err)
}
_, err = got.Declaration(Rendering{Names: twoMachines, Suffix: "internal",
Needed: map[string]map[string]string{"dnsmasq": {"broker": "sealed"}}})
if err == nil || !strings.Contains(err.Error(), "${machine:address}") {
t.Fatalf("a machine off the network was composed a resolver, or refused for another reason: %v", err)
// Left out of the declaration and said, rather than composed listening nowhere: a module that
// cannot compose on a machine is kept as it is there, with the reason (hq ADR 0163).
composed, err := got.Compose(Rendering{Names: twoMachines, Suffix: "internal",
Needed: map[string]map[string]string{"dnsmasq": {"broker": "sealed"}},
Settings: SettingsBy{"dnsmasq": {{From: "the mesh", Values: map[string]any{"listen-addresses": "127.0.0.1"}}}}})
if err == nil && !strings.Contains(composed.LeftOut["dnsmasq"], "${machine:address}") {
t.Fatalf("a machine off the network was composed a resolver, or left out for another reason: %v",
composed.LeftOut)
}
if err != nil && !strings.Contains(err.Error(), "${machine:address}") {
t.Fatalf("a machine off the network was refused for another reason: %v", err)
}
}
+75 -2
View File
@@ -99,8 +99,48 @@ var defaultSeats = []Seat{
{Name: "mesh-build-machine", Scope: ScopeMesh,
Accepts: []string{"build"}, Emits: []string{"started", "built", "log.*"}, Decision: "novox/hq ADR 0121"},
{Name: "node-dns-resolver", Scope: ScopeNode, Decision: "novox/hq ADR 0121"},
{Name: "node-intrusion-prevention", Scope: ScopeNode, Decision: "novox/hq ADR 0121"},
{Name: "node-packet-filter", Scope: ScopeNode, Decision: "novox/hq ADR 0121"},
// The intrusion prevention's verbs (novox/hq ADR 0179): what a person asks a machine's ban list
// whatever keeps it — who is banned and why, ban one address, let one go. Every holder serves all
// four; the jails themselves are composed from the modules the machine runs (to-be 31).
{Name: "node-intrusion-prevention", Scope: ScopeNode, Decision: "novox/hq ADR 0121",
Serves: []Verb{
{Name: "status", Description: "Every jail on this machine with how many it is watching and " +
"holding now, and the totals since the jail started; one jail's detail when named.",
Input: schema(map[string]string{"jail": "one jail (optional)"}, nil)},
{Name: "banned", Description: "Every address banned on this machine right now, with the jail " +
"that holds it and when the ban ends.",
Input: schema(map[string]string{"jail": "one jail (optional)"}, nil)},
{Name: "ban", Description: "Ban one address in one jail now, for the jail's ban time — an " +
"operator's act on the live ban list, which the mesh never writes itself.",
Input: schema(map[string]string{"ip": "the address", "jail": "the jail to hold it"}, []string{"ip", "jail"})},
{Name: "unban", Description: "Let one address go, from one jail or from every jail when none is named.",
Input: schema(map[string]string{"ip": "the address", "jail": "one jail (optional)"}, []string{"ip"})},
}},
// The packet filter's verbs (novox/hq ADR 0170): what a person asks a machine's filter whatever
// filter answers — the rules as enforced, reload the mesh's own, remove one thing the mesh did
// not write. Every holder serves all three; what differs by filter is the holder's own tools.
{Name: "node-packet-filter", Scope: ScopeNode, Decision: "novox/hq ADR 0121",
Serves: []Verb{
{Name: "rules", Description: "The packet filter as this machine enforces it now: the nftables " +
"ruleset and, where the tool exists, the legacy filter's listings. Narrowed to one table or " +
"chain when asked.",
Input: schema(map[string]string{"table": "one nftables table, as `family name` (optional)",
"chain": "one chain of that table (optional)"}, nil)},
{Name: "reload", Description: "Load the mesh's own filter again from the file the mesh writes, " +
"and answer with the mesh's table as loaded.",
Input: schema(map[string]string{}, nil)},
{Name: "remove", Description: "Remove one rule set the mesh did not write, named exactly as the " +
"host reports it (novox/hq ADR 0168) — `chain X (iptables-legacy)` or `table ip6 filter, chain " +
"DOCKER-USER`. Refuses the mesh's tables, the runtime's own chains, a built-in chain and an " +
"active found firewall's chains. An operator's act, by name, never a flush.",
Input: schema(map[string]string{"where": "the rule set, as `node show` lists it"}, []string{"where"})},
}},
// The machine's service manager (novox/hq ADR 0177). The host applies every declared unit,
// system or user scope; the holder answers questions and operator acts about them, each verb
// taking the unit and an optional scope. The holder runs nothing of its own: its verbs are
// served by the node tools runtime (ADR 0175).
{Name: "node-service-manager", Scope: ScopeNode, Decision: "novox/hq ADR 0177",
Serves: serviceManagerVerbs()},
// Deferred (novox/hq ADR 0121): renaming to mesh-private-network is a scope + server/client
// model change, not a rename, so it stays until that is built.
{Name: "the-private-network", Scope: ScopeNode, Decision: "novox/hq ADR 0110"},
@@ -374,3 +414,36 @@ func SeatsWithAProtocol() []Seat {
}
return out
}
// serviceManagerVerbs is the contract every holder of node-service-manager serves (novox/hq ADR
// 0177): the units on the machine in both scopes, read and acted on by name. Every verb takes an
// optional scope — "system" when absent, "user" for the operator account's own manager — so a
// caller asks for a user unit the way it asks for a system one.
func serviceManagerVerbs() []Verb {
scoped := func(more map[string]string, required []string) map[string]any {
props := map[string]string{"scope": "\"system\" (the default) or \"user\": the operator account's own manager"}
for k, v := range more {
props[k] = v
}
return schema(props, required)
}
unit := map[string]string{"unit": "the unit's name, as the service manager knows it"}
return []Verb{
{Name: "units", Description: "The units the service manager knows in a scope, each with its load, active and sub state; narrowed to a pattern when asked.",
Input: scoped(map[string]string{"pattern": "a glob the unit's name must match (optional)"}, nil)},
{Name: "status", Description: "One unit as the service manager sees it now: its states, whether it starts at boot, its main process, and whether the mesh declares it.",
Input: scoped(unit, []string{"unit"})},
{Name: "start", Description: "Start one unit. For a unit the mesh declares, the answer says the host will restore what its declaration says at the next apply.",
Input: scoped(unit, []string{"unit"})},
{Name: "stop", Description: "Stop one unit; for a mesh-declared unit the answer says the host will restore its declared state.",
Input: scoped(unit, []string{"unit"})},
{Name: "restart", Description: "Restart one unit.",
Input: scoped(unit, []string{"unit"})},
{Name: "enable", Description: "Make one unit start at boot (or at the account's login, in user scope).",
Input: scoped(unit, []string{"unit"})},
{Name: "disable", Description: "Stop one unit starting at boot (or at login, in user scope).",
Input: scoped(unit, []string{"unit"})},
{Name: "journal", Description: "The last lines of one unit's journal.",
Input: scoped(map[string]string{"unit": unit["unit"], "lines": "how many lines from the end (default 100)"}, []string{"unit"})},
}
}
+3 -2
View File
@@ -44,8 +44,9 @@ func TestTheSeatsAreAClosedSetAndEachNamesItsDecision(t *testing.T) {
delivered[s.Delivers] = s.Name
}
}
if len(Seats()) != 15 {
t.Errorf("the mesh defines %d seats rather than 14; the set is closed, so a change here is "+
// Sixteen since node-service-manager (novox/hq ADR 0177).
if len(Seats()) != 16 {
t.Errorf("the mesh defines %d seats rather than 16; the set is closed, so a change here is "+
"a decision (novox/hq ADR 0110): %s", len(Seats()), seatNames())
}
}
+128
View File
@@ -3,6 +3,7 @@ package catalogue
import (
"encoding/json"
"fmt"
"regexp"
"sort"
"strings"
)
@@ -275,6 +276,11 @@ func UnusedSettings(m Manifest, layers []Layer) []string {
if key == AccessesSetting && len(m.Accesses) > 0 {
continue
}
// `networks` keeps a found network for a taken container on one adopted machine
// (novox/hq ADR 0163). Validated in KeptNetworks, so not stray.
if key == NetworksSetting {
continue
}
unused = append(unused, fmt.Sprintf(
"%s sets %q, and %s has no file that merges it, asks for no ${setting:%s}, and "+
"declares no %q in what it contributes or serves",
@@ -298,3 +304,125 @@ func stringsOf(v any) []string {
}
return out
}
// NetworksSetting is the settings key that keeps a found network for a taken container, on one
// adopted machine (novox/hq ADR 0163, rule 4):
//
// {"networks": {"server": ["predecessor_default"]}}
//
// has the module's container `server` also join `predecessor_default` once taken, so a neighbour
// that resolves it by name on that network keeps resolving it. Migration scaffolding in the sense
// of ADR 0104: assigned only on an adopted machine, reported while it stands, removed when the
// neighbours are taken. Keyed by the container's resource id; the value is the networks it keeps.
const NetworksSetting = "networks"
// KeptNetworks reads which found networks each of a module's containers keeps, by container id.
//
// Refused from a mesh-wide layer — a found network is a fact about one machine — for an id the
// module declares no container under, for a name that is not a network's, and on a machine that
// is not adopted: the setting exists so neighbours the mesh has not taken yet keep reaching the
// container, and a converged machine has no such neighbours.
func KeptNetworks(m Manifest, layers []Layer, adopted bool) (map[string][]string, error) {
containers := map[string]bool{}
for _, r := range m.Resources {
if fmt.Sprint(r["type"]) == "container" {
containers[fmt.Sprint(r["id"])] = true
}
}
out := map[string][]string{}
for _, layer := range layers {
raw, ok := layer.Values[NetworksSetting]
if !ok {
continue
}
if layer.From == MeshWideLayer {
return nil, fmt.Errorf("%s: %s is given per node — a found network is a fact about one "+
"machine; set it with --node", m.Module, NetworksSetting)
}
if !adopted {
return nil, fmt.Errorf("%s: %s keeps a found network for neighbours the mesh has not taken "+
"yet, and %s is converged — nothing on it is found; clear the setting", m.Module,
NetworksSetting, layer.From)
}
blocks, ok := raw.(map[string]any)
if !ok {
return nil, fmt.Errorf("%s: %s is a { container: [network, …] } map, and %q set it to "+
"something else", m.Module, NetworksSetting, layer.From)
}
for id, body := range blocks {
if !containers[id] {
return nil, fmt.Errorf("%s: %s names the container %q, which it does not declare — "+
"the setting reaches nothing; it declares %s", m.Module, NetworksSetting, id,
orNothing(sortedKeys(containers)))
}
names := stringsOf(body)
if len(names) == 0 {
return nil, fmt.Errorf("%s: %s for %q is a list of network names, and %q set it to %v",
m.Module, NetworksSetting, id, layer.From, body)
}
for _, n := range names {
if !networkName.MatchString(n) {
return nil, fmt.Errorf("%s: %s for %q names %q, which is not a network name",
m.Module, NetworksSetting, id, n)
}
}
sort.Strings(names)
out[id] = names
}
}
if len(out) == 0 {
return nil, nil
}
return out, nil
}
// networkName is what a container runtime accepts as a network's name.
var networkName = regexp.MustCompile(`^[A-Za-z0-9][A-Za-z0-9_.-]*$`)
// JudgeSettings composes a module's settings against its definition and refuses the first thing
// that cannot work, naming the module, the layer and the key (novox/hq ADR 0163, rule 6).
//
// **The same judgement where a setting is stored and where a machine is declared.** Stored, a
// setting that cannot compose is refused before it is kept; composed later, a definition that has
// moved under a stored setting leaves that module out of the machine's declaration rather than
// the machine without one. Every reader of settings runs here: a port given, an exposure, a reach,
// an endpoint, a placement, an access, a kept network, a mergeable file's keys and a file's
// `${setting:…}`. A key that reaches nothing is not here: it cannot break a composition, so it is
// refused where it is stored (SetSettings, with UnusedSettings) and said where a plan is read,
// and never costs a module its place.
func JudgeSettings(m Manifest, layers []Layer, adopted bool) error {
// With no layers too: a definition may ask for a setting nobody made — an access placed by
// nobody, a file's ${setting:…} nothing sets — and that is the same statement, missing.
if _, err := GivenPorts(m, layers); err != nil {
return err
}
if _, err := Reaches(m, layers); err != nil {
return err
}
if _, err := Endpoints(m, layers); err != nil {
return err
}
if _, err := Places(m, layers); err != nil {
return err
}
if _, _, err := accessesFor(m, layers); err != nil {
return err
}
if _, err := KeptNetworks(m, layers, adopted); err != nil {
return err
}
for _, r := range m.Resources {
settled, err := ApplySettings(r, layers)
if err != nil {
return err
}
copied := map[string]any{}
for k, v := range settled {
copied[k] = v
}
if err := settingInto(copied, layers, m.Module); err != nil {
return err
}
}
return nil
}
+127
View File
@@ -0,0 +1,127 @@
package catalogue
import (
"strings"
"testing"
)
// A setting is judged where it is stored, and an impossible one costs a module, not a machine
// (novox/hq ADR 0163, rule 6): the one judgement, used by SetSettings before storing and by
// Compose when a definition has moved under a stored setting.
func TestASettingThatCannotComposeIsRefusedByNameAndLeavesOnlyItsModuleOut(t *testing.T) {
web := Manifest{Module: "hello-web",
Listens: []Listening{{Port: 8080, From: FromEverywhere}},
Resources: []map[string]any{{"id": "server", "type": "container", "name": "hello-web",
"ports": []any{"8080"}}}}
for _, c := range []struct {
name string
layer map[string]any
refuse string
}{
{"a port the module does not publish", map[string]any{PortsSetting: map[string]any{"9999": 10000}},
"hello-web gives port 9999 a machine port, and no container of its publishes 9999"},
{"a mesh-wide port", map[string]any{PortsSetting: map[string]any{"8080": 10000}},
"a port is a fact about one machine"},
} {
from := "anchor"
if c.name == "a mesh-wide port" {
from = MeshWideLayer
}
err := JudgeSettings(web, []Layer{{From: from, Values: c.layer}}, true)
if err == nil || !strings.Contains(err.Error(), c.refuse) {
t.Errorf("%s: judged %v, want %q", c.name, err, c.refuse)
}
}
if err := JudgeSettings(web, []Layer{{From: "anchor", Values: map[string]any{PortsSetting: map[string]any{"8080": 10000}}}}, true); err != nil {
t.Fatalf("a port the module publishes was refused: %v", err)
}
// Composed, a module whose stored setting no longer works is left out by name; the rest of
// the machine is declared.
r := anAdoptedAnchor()
with := anchorRendering(false)
with.Settings = SettingsBy{"hello-web": {{From: "anchor", Values: map[string]any{PortsSetting: map[string]any{"9999": 10000}}}}}
composed, err := r.Compose(with)
if err != nil {
t.Fatal(err)
}
why, left := composed.LeftOut["hello-web"]
if !left || !strings.Contains(why, "no container of its publishes 9999") {
t.Fatalf("hello-web is not left out by name: %v", composed.LeftOut)
}
if len(composed.LeftOut) != 1 {
t.Fatalf("more than hello-web is left out: %v", composed.LeftOut)
}
got := byID(composed.Resources)
if _, declared := got["hello-web.server"]; declared {
t.Fatal("the left-out module's container is still declared")
}
if _, declared := got["distribution.store"]; !declared {
t.Fatal("the rest of the machine was not declared")
}
if left := r.LeftOut(with.Settings, false); len(left) != 1 || left["hello-web"] == "" {
t.Fatalf("the judgement a plan reads differs from what compose did: %v", left)
}
}
// A taken container keeps a found network by a per-machine setting (novox/hq ADR 0163, rule 4):
// on an adopted machine only, for a container the module declares, and it reaches the container's
// declaration as the networks it also joins.
func TestAKeptNetworkReachesTheContainerOnAnAdoptedMachineOnly(t *testing.T) {
r := anAdoptedAnchor()
keep := SettingsBy{"hello-web": {{From: "anchor",
Values: map[string]any{NetworksSetting: map[string]any{"server": []any{"predecessor_default"}}}}}}
with := anchorRendering(true)
with.Settings = keep
composed, err := r.Compose(with)
if err != nil {
t.Fatal(err)
}
if len(composed.LeftOut) != 0 {
t.Fatalf("a kept network left a module out: %v", composed.LeftOut)
}
server := byID(composed.Resources)["hello-web.server"]
networks, _ := server["networks"].([]any)
if len(networks) != 1 || networks[0] != "predecessor_default" {
t.Fatalf("the container does not join the kept network: %v", server)
}
if _, has := byID(composed.Resources)["distribution.store"]["networks"]; has {
t.Fatal("another container joins a network nobody kept for it")
}
// Converged, the setting reaches nothing it was for, and the module is left out saying so.
with = anchorRendering(false)
with.Settings = keep
composed, err = r.Compose(with)
if err != nil {
t.Fatal(err)
}
if why := composed.LeftOut["hello-web"]; !strings.Contains(why, "anchor is converged") {
t.Fatalf("a kept network on a converged machine: %v", composed.LeftOut)
}
web := r.Modules[4]
for _, c := range []struct {
name string
layer Layer
want string
}{
{"mesh-wide", Layer{From: MeshWideLayer, Values: map[string]any{NetworksSetting: map[string]any{"server": []any{"x"}}}},
"a found network is a fact about one machine"},
{"an unknown container", Layer{From: "anchor", Values: map[string]any{NetworksSetting: map[string]any{"db": []any{"x"}}}},
`names the container "db", which it does not declare`},
{"not a list", Layer{From: "anchor", Values: map[string]any{NetworksSetting: map[string]any{"server": "x"}}},
"is a list of network names"},
{"not a network name", Layer{From: "anchor", Values: map[string]any{NetworksSetting: map[string]any{"server": []any{"a/b"}}}},
"which is not a network name"},
} {
_, err := KeptNetworks(web, []Layer{c.layer}, true)
if err == nil || !strings.Contains(err.Error(), c.want) {
t.Errorf("%s: %v, want %q", c.name, err, c.want)
}
}
if kept, err := KeptNetworks(web, nil, false); err != nil || kept != nil {
t.Fatalf("no setting: %v %v", kept, err)
}
}
+23
View File
@@ -129,6 +129,29 @@ var ControllerVerbs = []Verb{
"module": "an own secret: the module",
"secret": "an own secret: its name in the module's definition",
}, nil)},
{Name: "issue", Description: "Give a module on a machine its account on the bus: minted, and sealed to the " +
"machine as the module's own secret named broker, read at the next push of that machine. For a module " +
"whose definition declares that secret; refused with the reason otherwise. Issued again, it replaces the account.",
Input: schema(map[string]string{
"node": "the machine that runs the module",
"module": "the module's name",
}, []string{"node", "module"})},
{Name: "settings", Description: "Set what an assignment is configured with: a module's settings for the whole mesh, " +
"or for one machine. Replaces that layer whole — what it does not name, it no longer sets — and takes effect " +
"at the next push. With clear, removes the layer and the module is back to what its definition says.",
Input: schema(map[string]string{
"module": "the module's name",
"values": "the settings as a JSON object, for set",
"node": "one machine; the whole mesh when absent",
"clear": "\"true\" to remove the layer instead of setting it",
}, []string{"module"})},
{Name: "command", Description: "Run one command line of the controller's own, as you would type it at its " +
"shell — `node account g14 jochen`, `node show ace`, `module list` — and answer what it printed. The " +
"generic verb beside the named ones (novox/hq ADR 0154): everything the binary can do, without a verb " +
"per command. Any node may call any tool (ADR 0175), so nothing is held back here.",
Input: schema(map[string]string{
"command": "the command line, as the controller's binary takes it; quotes group a word with spaces",
}, []string{"command"})},
{Name: "build", Description: "Have the build machine build a repository. Answers at once with the build's id: " +
"`builds` with that id follows it line by line, and the module is registered when the outcome comes.",
Input: schema(map[string]string{
+89
View File
@@ -0,0 +1,89 @@
package catalogue
import (
"encoding/json"
"strings"
"testing"
)
// A scheduled step may hold its module's own containers still while it runs (novox/hq ADR 0189).
//
// The host judges what it receives — whether each id is a container on that machine. What the
// definition is the only place to see is judged here, near whoever wrote it.
func aStoreManifest(step map[string]any) []byte {
m := map[string]any{
"module": "distribution", "version": "1",
"resources": []any{
map[string]any{"id": "store", "type": "container", "name": "mesh-registry",
"image": "registry@sha256:" + strings.Repeat("a", 64)},
step,
},
}
raw, _ := json.Marshal(m)
return raw
}
func TestAMaintenanceWindowOnItsOwnModulesContainerIsAccepted(t *testing.T) {
raw := aStoreManifest(map[string]any{
"id": "collect", "type": "container", "name": "mesh-registry-collect",
"image": "registry@sha256:" + strings.Repeat("a", 64),
"schedule": "30 3 * * *", "while-stopped": []any{"store"},
})
if _, err := ParseManifest(raw); err != nil {
t.Fatalf("a step holding its own module's container still was refused: %v", err)
}
}
func TestAMaintenanceWindowIsRefusedWhereTheDefinitionShowsItCannotMean(t *testing.T) {
for _, c := range []struct {
name string
step map[string]any
says string
}{
{
"on a step with no schedule",
map[string]any{"id": "collect", "type": "container", "name": "c",
"image": "registry@sha256:" + strings.Repeat("a", 64),
"while-stopped": []any{"store"}},
"gates what is declared after it",
},
{
"on a run-once step, which already has order",
map[string]any{"id": "collect", "type": "container", "name": "c",
"image": "registry@sha256:" + strings.Repeat("a", 64),
"run-once": true, "while-stopped": []any{"store"}},
"A maintenance window is for a recurring step",
},
{
"naming a container this module does not declare",
map[string]any{"id": "collect", "type": "container", "name": "c",
"image": "registry@sha256:" + strings.Repeat("a", 64),
"schedule": "30 3 * * *", "while-stopped": []any{"the-broker"}},
"could quiesce a neighbour could stop the mesh",
},
{
"naming itself",
map[string]any{"id": "collect", "type": "container", "name": "c",
"image": "registry@sha256:" + strings.Repeat("a", 64),
"schedule": "30 3 * * *", "while-stopped": []any{"collect"}},
"naming itself",
},
{
"written as something that is not a list",
map[string]any{"id": "collect", "type": "container", "name": "c",
"image": "registry@sha256:" + strings.Repeat("a", 64),
"schedule": "30 3 * * *", "while-stopped": "store"},
"a list of this module's container ids",
},
} {
_, err := ParseManifest(aStoreManifest(c.step))
if err == nil {
t.Errorf("%s was accepted", c.name)
continue
}
if !strings.Contains(err.Error(), c.says) {
t.Errorf("%s: the refusal does not say %q:\n%v", c.name, c.says, err)
}
}
}
+97
View File
@@ -178,6 +178,103 @@ type Stray struct {
Detail string `json:"detail,omitempty"`
}
// A Filter is one place on a machine that refuses traffic, with its owner (novox/hq ADR 0168).
type Filter struct {
Where string `json:"where"`
Owner string `json:"owner"`
Refuses string `json:"refuses"`
}
// Owners of a filter, as the host names them (ADR 0168).
const (
FilterMesh = "mesh"
FilterFoundFirewall = "found-firewall"
FilterRuntime = "runtime"
FilterBan = "ban"
FilterOther = "other"
)
// FoundFirewall is the state of a converged machine's found firewall (ADR 0168): in force now or
// not, and how it came to be inactive.
type FoundFirewall struct {
Kind string `json:"kind"`
Active bool `json:"active"`
RetiredBy string `json:"retired_by,omitempty"`
}
// Filtering is what a machine last said filters it (ADR 0168).
type Filtering struct {
Filters []Filter
FoundFirewall *FoundFirewall
}
// Alone is whether the machine is filtered by the mesh alone: nothing in its list but the mesh's
// own, the runtime's plumbing and bans, and no found firewall in force.
func (f Filtering) Alone() bool {
for _, x := range f.Filters {
if x.Owner == FilterOther || x.Owner == FilterFoundFirewall {
return false
}
}
return f.FoundFirewall == nil || !f.FoundFirewall.Active
}
// Others is every filter that is neither the mesh's, the runtime's nor a ban.
func (f Filtering) Others() []Filter {
var out []Filter
for _, x := range f.Filters {
if x.Owner == FilterOther || x.Owner == FilterFoundFirewall {
out = append(out, x)
}
}
return out
}
// RecordFiltering keeps what a machine last said filters it, replacing what was there (ADR 0168).
func (i *Inventory) RecordFiltering(ctx context.Context, nodeID string, filters []Filter, found *FoundFirewall) error {
raw, err := json.Marshal(nonNil(filters))
if err != nil {
return err
}
var foundRaw any
if found != nil {
b, err := json.Marshal(found)
if err != nil {
return err
}
foundRaw = string(b)
}
_, err = i.store.Pool().Exec(ctx,
`update node set filters = $2, found_firewall = $3 where id = $1`, nodeID, raw, foundRaw)
return err
}
// FilteringOf is what a machine last said filters it; empty for a machine that never said.
func (i *Inventory) FilteringOf(ctx context.Context, name string) (Filtering, error) {
var filtersRaw, foundRaw []byte
err := i.store.Pool().QueryRow(ctx,
`select filters, found_firewall from node where name = $1`, name).Scan(&filtersRaw, &foundRaw)
if errors.Is(err, pgx.ErrNoRows) {
return Filtering{}, fmt.Errorf("%w: %s", ErrNoSuchNode, name)
}
if err != nil {
return Filtering{}, err
}
var out Filtering
if len(filtersRaw) > 0 {
if err := json.Unmarshal(filtersRaw, &out.Filters); err != nil {
return Filtering{}, err
}
}
if len(foundRaw) > 0 {
out.FoundFirewall = &FoundFirewall{}
if err := json.Unmarshal(foundRaw, out.FoundFirewall); err != nil {
return Filtering{}, err
}
}
return out, nil
}
// Reach is one thing reachable on an adopted node: a listening socket or a published port.
type Reach struct {
Protocol string `json:"protocol"`
+50
View File
@@ -605,6 +605,12 @@ func (i *Inventory) SetSettings(ctx context.Context, nodeName, module string, va
if err != nil {
return err
}
// Judged here, against the module's current definition, before it is kept (novox/hq ADR 0163,
// rule 6): a setting that cannot compose is refused where it is set, naming the node, the
// module, the layer and the key — never stored to refuse the whole machine where it is read.
if err := i.judgeSettings(ctx, nodeName, module, values); err != nil {
return err
}
if nodeName == "" {
// A port is a fact about one machine (novox/hq ADR 0100). Refused here, in composition's
// words: stored, it refuses every node running the module at composition, and the mesh
@@ -661,6 +667,50 @@ func (i *Inventory) SetSettings(ctx context.Context, nodeName, module string, va
return tx.Commit(ctx)
}
// judgeSettings composes a layer somebody is about to store against the module's definition, with
// the mesh-wide layer under it when the layer is one node's, and refuses the first thing that
// cannot work (ADR 0163, rule 6). The same judgement composition makes; what passes here composes.
func (i *Inventory) judgeSettings(ctx context.Context, nodeName, module string, values map[string]any) error {
m, err := i.declared(ctx, module)
if err != nil {
return fmt.Errorf("%w: %s", ErrNoSuchModule, module)
}
where, from := "the mesh", catalogue.MeshWideLayer
adopted := false
var layers []catalogue.Layer
if nodeName != "" {
node, err := i.NodeByName(ctx, nodeName)
if err != nil {
return err
}
where, from, adopted = nodeName, nodeName, node.Adopted
var meshWide []byte
err = i.store.Pool().QueryRow(ctx,
`select values from settings where module = $1 and node is null`, module).Scan(&meshWide)
if err != nil && !errors.Is(err, pgx.ErrNoRows) {
return err
}
if len(meshWide) > 0 {
var under map[string]any
if err := json.Unmarshal(meshWide, &under); err != nil {
return err
}
layers = append(layers, catalogue.Layer{From: catalogue.MeshWideLayer, Values: under})
}
}
layers = append(layers, catalogue.Layer{From: from, Values: values})
if err := catalogue.JudgeSettings(m, layers, adopted); err != nil {
return fmt.Errorf("refused: %s on %s cannot compose with the layer %q — %w", module, where, from, err)
}
// And a key that reaches nothing, refused here where somebody can still fix the spelling:
// stored, it would be a setting somebody believes they made.
if stray := catalogue.UnusedSettings(m, layers[len(layers)-1:]); len(stray) > 0 {
return fmt.Errorf("refused: %s on %s — these settings reach nothing:\n - %s", module, where,
strings.Join(stray, "\n - "))
}
return nil
}
// givenIn is the machine ports a node-level settings layer gives a module, software port →
// machine port (novox/hq ADR 0100). Nothing when the layer gives none; what is not a port is left
// for composition to refuse in its own words.
+242
View File
@@ -0,0 +1,242 @@
package inventory
import (
"context"
"encoding/json"
"strings"
)
// What the artifact store keeps, and what it may let go (novox/hq ADR 0189, issue 108).
//
// The store has never collected anything: every build pushes another layer set and nothing has
// ever removed one. The registry's own answer — collect what no tag names — is wrong here, because
// the mesh pushes each artifact under one moving tag and pins machines by digest, so every build
// but the newest is untagged and some machine may still be running it.
//
// **So the mesh decides, from its own records, and it never has to look in the store to do it.**
// It has never put anything there it did not record, which means every digest it could remove is
// already in a build row. A digest the mesh did not record making is therefore never named here —
// not as a safety margin but as the rule restated, and it is what keeps the sweep away from the
// images genesis pushed before any record existed (04-ISSUES/102, F4).
// KeptBuilds is how many successful builds of each module keep their artifacts, counting the
// newest. The newest is what the mesh hands a machine now; the four behind it are how far back a
// release that turns out wrong can be taken.
const KeptBuilds = 5
// ToCollect is every artifact the mesh made, no longer keeps, and has not already collected.
//
// Three reasons an artifact stays, and nothing else is a reason:
//
// - **a definition names it** — the reference appears in a module's recorded manifest, which is
// what the mesh would hand a machine now. No age limit: this is the floor;
// - **the mesh can still go back to it** — it is an artifact of one of the KeptBuilds most
// recent successful builds of its module;
// - it was already collected, in which case there is nothing left to do.
//
// Returned in a stated order so two runs over the same records ask for the same things in the
// same sequence, which is what makes a failed sweep safe to simply run again.
func (i *Inventory) ToCollect(ctx context.Context) ([]string, error) {
keep, err := i.keptReferences(ctx)
if err != nil {
return nil, err
}
rows, err := i.store.Pool().Query(ctx,
// Every artifact of every successful build, oldest first, minus what has already been
// collected. A failed build published nothing, so it names nothing to remove.
`select b.made
from build b
where b.failed = '' and b.module is not null and b.module <> ''
order by b.at asc, b.id asc`)
if err != nil {
return nil, err
}
defer rows.Close()
collected, err := i.alreadyCollected(ctx)
if err != nil {
return nil, err
}
seen := map[string]bool{}
var out []string
for rows.Next() {
var raw []byte
if err := rows.Scan(&raw); err != nil {
return nil, err
}
var made []Artifact
if err := json.Unmarshal(raw, &made); err != nil {
// One unreadable record must not stop the rest being collected — and an artifact this
// row named is simply not offered, which errs toward keeping.
continue
}
for _, a := range made {
if a.Reference == "" || keep[a.Reference] || collected[a.Reference] || seen[a.Reference] {
continue
}
seen[a.Reference] = true
out = append(out, a.Reference)
}
}
return out, rows.Err()
}
// keptReferences is every artifact reference the mesh still keeps, for either of the two reasons.
func (i *Inventory) keptReferences(ctx context.Context) (map[string]bool, error) {
keep := map[string]bool{}
// **Whatever a definition the mesh holds names.** Read as text rather than by walking the
// resource shapes: a reference may be a container's image, a bundle's source, or a field some
// later kind of resource grows, and what matters is only whether the mesh could hand this
// string to a machine. A manifest that mentions it is a manifest that might.
manifests, err := i.store.Pool().Query(ctx, `select manifest::text from module where manifest is not null`)
if err != nil {
return nil, err
}
defer manifests.Close()
var named []string
for manifests.Next() {
var text string
if err := manifests.Scan(&text); err != nil {
return nil, err
}
named = append(named, text)
}
if err := manifests.Err(); err != nil {
return nil, err
}
// The KeptBuilds most recent successful builds of each module, whole.
recent, err := i.store.Pool().Query(ctx,
`select made from (
select made, row_number() over (partition by module order by at desc, id desc) as back
from build
where failed = '' and module is not null and module <> ''
) ranked where back <= $1`, KeptBuilds)
if err != nil {
return nil, err
}
defer recent.Close()
for recent.Next() {
var raw []byte
if err := recent.Scan(&raw); err != nil {
return nil, err
}
var made []Artifact
if err := json.Unmarshal(raw, &made); err != nil {
continue
}
for _, a := range made {
if a.Reference != "" {
keep[a.Reference] = true
}
}
}
if err := recent.Err(); err != nil {
return nil, err
}
// And anything a manifest mentions. Done after the recent set so the scan runs over the
// candidates rather than over every reference ever recorded: a manifest holds a reference
// composed with the store's address or kept bare, so the search is for the digest within it.
if len(named) > 0 {
all, err := i.everyReferenceMade(ctx)
if err != nil {
return nil, err
}
for _, reference := range all {
if keep[reference] {
continue
}
digest := digestIn(reference)
if digest == "" {
// Not something the store holds by digest; nothing here can speak for it, so it
// is kept rather than guessed about.
keep[reference] = true
continue
}
for _, text := range named {
if strings.Contains(text, digest) {
keep[reference] = true
break
}
}
}
}
return keep, nil
}
// everyReferenceMade is every artifact reference any successful build recorded.
func (i *Inventory) everyReferenceMade(ctx context.Context) ([]string, error) {
rows, err := i.store.Pool().Query(ctx,
`select made from build where failed = '' and module is not null and module <> ''`)
if err != nil {
return nil, err
}
defer rows.Close()
seen := map[string]bool{}
var out []string
for rows.Next() {
var raw []byte
if err := rows.Scan(&raw); err != nil {
return nil, err
}
var made []Artifact
if err := json.Unmarshal(raw, &made); err != nil {
continue
}
for _, a := range made {
if a.Reference == "" || seen[a.Reference] {
continue
}
seen[a.Reference] = true
out = append(out, a.Reference)
}
}
return out, rows.Err()
}
// digestIn is the `sha256:<hex>` a reference names, empty when it names none.
func digestIn(reference string) string {
for _, marker := range []string{"@sha256:", "/sha256:"} {
if _, after, ok := strings.Cut(reference, marker); ok {
return "sha256:" + after
}
}
return ""
}
// alreadyCollected is what the store has already been asked to let go.
func (i *Inventory) alreadyCollected(ctx context.Context) (map[string]bool, error) {
rows, err := i.store.Pool().Query(ctx, `select reference from artifact_collected`)
if err != nil {
return nil, err
}
defer rows.Close()
out := map[string]bool{}
for rows.Next() {
var reference string
if err := rows.Scan(&reference); err != nil {
return nil, err
}
out[reference] = true
}
return out, rows.Err()
}
// MarkCollected records that the store no longer holds these.
//
// **A store that answered "not found" is recorded too.** The outcome wanted is that the artifact
// is gone, and it is; retrying it every sweep for ever is the failure this table exists to
// prevent. Only a store that could not be reached, or refused, leaves a reference unmarked — and
// then the next sweep asks again, which is what should happen.
func (i *Inventory) MarkCollected(ctx context.Context, references []string) error {
for _, reference := range references {
if _, err := i.store.Pool().Exec(ctx,
`insert into artifact_collected (reference) values ($1) on conflict (reference) do nothing`,
reference); err != nil {
return err
}
}
return nil
}
+147
View File
@@ -0,0 +1,147 @@
package inventory
import (
"context"
"fmt"
"testing"
"github.com/novox/mesh-controller/internal/catalogue"
)
// What the store keeps, and what it may let go (novox/hq ADR 0189, issue 108).
//
// The store has collected nothing since it was raised, and the registry's own answer — collect
// what no tag names — would delete images machines are running, because the mesh pushes under one
// moving tag and pins by digest. So the rule is the mesh's, read from its own records, and these
// are the three reasons an artifact stays and the one reason it goes.
// ref is an artifact reference as the mesh records one.
func ref(module, artifact string, n int) string {
return fmt.Sprintf("%s%s/%s@sha256:%064x", catalogue.ArtifactStoreScheme, module, artifact, n)
}
// built records one successful build of a module publishing one image.
func built(t *testing.T, inv *Inventory, id, module string, n int) string {
t.Helper()
reference := ref(module, "app", n)
b := aBuild(id, module, "")
b.Made = []Artifact{{Name: "app", Kind: "image", Reference: reference}}
if err := inv.RecordBuild(context.Background(), b); err != nil {
t.Fatal(err)
}
return reference
}
func TestTheStoreKeepsTheRecentBuildsAndLetsGoOfTheRest(t *testing.T) {
inv := fresh(t)
ctx := context.Background()
// Eight builds of one module, oldest first. Five are kept — the newest, and the four a
// release that turns out wrong can be taken back to.
var made []string
for i := 1; i <= 8; i++ {
made = append(made, built(t, inv, fmt.Sprintf("b%02d", i), "web", i))
}
go_, err := inv.ToCollect(ctx)
if err != nil {
t.Fatal(err)
}
want := made[:3] // the three oldest
if len(go_) != len(want) {
t.Fatalf("offered %v to collect; want the %d oldest of %d", go_, len(want), len(made))
}
for i := range want {
if go_[i] != want[i] {
t.Fatalf("offered %v; want %v — and in that order, so a failed sweep is safe to run again",
go_, want)
}
}
}
func TestADefinitionNamingAnArtifactKeepsItHoweverOldItIs(t *testing.T) {
// The floor: no age limit. A module recorded at an older commit still names what the mesh
// would hand a machine now, and that is what must not be collected out from under it.
inv := fresh(t)
ctx := context.Background()
var made []string
for i := 1; i <= 8; i++ {
made = append(made, built(t, inv, fmt.Sprintf("b%02d", i), "web", i))
}
oldest := made[0]
// A definition the mesh holds, whose container runs that oldest image.
m := catalogue.Manifest{Module: "web", Version: "1", Resources: []map[string]any{{
"id": "app", "type": "container", "name": "web", "image": oldest,
}}}
if err := inv.RegisterModule(ctx, m, Source{Repository: "https://forge.invalid/web.git"}); err != nil {
t.Fatal(err)
}
go_, err := inv.ToCollect(ctx)
if err != nil {
t.Fatal(err)
}
for _, reference := range go_ {
if reference == oldest {
t.Fatalf("the mesh offered to collect %s, which a definition it holds names", oldest)
}
}
if len(go_) != 2 {
t.Fatalf("offered %v; want the two oldest that nothing names", go_)
}
}
func TestWhatHasBeenCollectedIsNotOfferedAgain(t *testing.T) {
// Without this the sweep reissues a delete for every artifact it has ever collected, every
// time it runs, for ever — a number of requests that grows with the mesh's whole history.
inv := fresh(t)
ctx := context.Background()
for i := 1; i <= 7; i++ {
built(t, inv, fmt.Sprintf("b%02d", i), "web", i)
}
first, err := inv.ToCollect(ctx)
if err != nil {
t.Fatal(err)
}
if len(first) != 2 {
t.Fatalf("offered %v, want two", first)
}
if err := inv.MarkCollected(ctx, first); err != nil {
t.Fatal(err)
}
again, err := inv.ToCollect(ctx)
if err != nil {
t.Fatal(err)
}
if len(again) != 0 {
t.Fatalf("offered %v again after collecting it", again)
}
}
func TestAFailedBuildNamesNothingToCollectAndEachModuleIsCountedOnItsOwn(t *testing.T) {
inv := fresh(t)
ctx := context.Background()
// A failed build published nothing, so it is neither kept nor collected — and it must not
// count against the module's five.
for i := 1; i <= 6; i++ {
built(t, inv, fmt.Sprintf("w%02d", i), "web", i)
}
if err := inv.RecordBuild(ctx, aBuild("w99", "web", "the recipe would not build")); err != nil {
t.Fatal(err)
}
// And a second module with three builds keeps all three: five each, not five between them.
for i := 1; i <= 3; i++ {
built(t, inv, fmt.Sprintf("d%02d", i), "db", 100+i)
}
go_, err := inv.ToCollect(ctx)
if err != nil {
t.Fatal(err)
}
if len(go_) != 1 || go_[0] != ref("web", "app", 1) {
t.Fatalf("offered %v; want only web's oldest — db's three are all within its five", go_)
}
}
+8 -1
View File
@@ -31,8 +31,11 @@ func TestRegisteringAModuleAgainKeepsWhatTheMeshHoldsForIt(t *testing.T) {
if err := inv.RecordSealingKey(ctx, node.ID, key); err != nil {
t.Fatal(err)
}
// A mergeable file, so any setting composes (novox/hq ADR 0163, rule 6: a setting is judged
// where it is stored).
m := catalogue.Manifest{Module: "step-ca", Version: "1",
Provides: catalogue.Offers("acme-ca"), OwnSecrets: catalogue.OwnSecrets{"password": {Path: "/run/password"}}}
Provides: catalogue.Offers("acme-ca"), OwnSecrets: catalogue.OwnSecrets{"password": {Path: "/run/password"}},
Resources: []map[string]any{{"id": "conf", "type": "file", "path": "/etc/step-ca.json", "content": "{}", "merge": "json"}}}
if err := inv.RegisterModule(ctx, m, Source{}); err != nil {
t.Fatal(err)
}
@@ -229,5 +232,9 @@ func TestAModuleStillAssignedRefusesBeforeAnythingAboutWhatItHolds(t *testing.T)
// declares (novox/hq 04-ISSUES/078).
func withOwnSecret(m catalogue.Manifest, name string) catalogue.Manifest {
m.OwnSecrets = catalogue.OwnSecrets{name: {Path: "/run/" + name}}
// And a mergeable file, so any setting these tests store composes (novox/hq ADR 0163, rule 6:
// a setting is judged where it is stored).
m.Resources = append(m.Resources, map[string]any{"id": "conf", "type": "file",
"path": "/etc/" + m.Module + ".json", "content": "{}", "merge": "json"})
return m
}
@@ -0,0 +1,7 @@
-- What filters a machine, with owners, as the host reports it with every apply (novox/hq ADR 0168):
-- every table and chain that refuses traffic — the mesh's, the found firewall's, the runtime's own,
-- a ban, or other — so the mesh says truthfully what filters a converged machine and names what it
-- did not write. And the state of the firewall a converged machine was found with: in force now or
-- not, and who retired it.
alter table node add column filters jsonb;
alter table node add column found_firewall jsonb;
@@ -0,0 +1,23 @@
-- What the artifact store no longer keeps (novox/hq ADR 0189, issue 108).
--
-- The mesh removes from its store only what it put there and can account for: every digest it
-- could remove is already in a build record, so the sweep reads its own records rather than
-- enumerating the store. What it does not get from those records is whether it has already
-- removed something -- `build.made` says what that build published, for ever, which is history
-- and not an index of what is on disk.
--
-- Without this the sweep would reissue a delete for every artifact it has ever collected, every
-- time it runs, and each one would answer 404 -- a number of requests that grows with the mesh's
-- whole history and never shrinks.
--
-- Keyed by the reference as the mesh records it (`artifact-store://<module>/<artifact>@sha256:…`),
-- because that is the identity the record uses everywhere else. Not a foreign key to build: two
-- builds can publish the same digest (the same source built twice produces the same bytes), and
-- what is collected is the artifact, not the attempt that made it.
create table artifact_collected (
reference text primary key,
-- When the store answered. Kept so a reader of an old build record can tell "this artifact is
-- gone" from "this artifact was never there", which are different kinds of surprise.
at timestamptz not null default now()
);
+9 -2
View File
@@ -15,9 +15,16 @@ func aNodeWithModules(t *testing.T, modules ...string) (*Inventory, string) {
if _, err := inv.AddNode(ctx, "anchor"); err != nil {
t.Fatal(err)
}
// Each publishes the port these tests give it a machine port for: a port given for one the
// module does not publish is refused where it is stored (novox/hq ADR 0163, rule 6).
publishes := map[string]string{"postgres": "5432", "another-database": "5432", "cache": "6379", "web": "8080"}
for _, m := range modules {
if err := inv.RegisterModule(ctx,
catalogue.Manifest{Module: m, Version: "1"}, Source{}); err != nil {
manifest := catalogue.Manifest{Module: m, Version: "1"}
if port, known := publishes[m]; known {
manifest.Resources = []map[string]any{{"id": "server", "type": "container", "name": m,
"image": "x", "ports": []any{port}}}
}
if err := inv.RegisterModule(ctx, manifest, Source{}); err != nil {
t.Fatal(err)
}
}
+49
View File
@@ -820,3 +820,52 @@ func (i *Inventory) SharedHolders(ctx context.Context, provider, providerModule,
sort.Strings(out)
return out, nil
}
// SecretState is one secret a module holds on a machine, as a take compares it (novox/hq ADR
// 0163): its name, where it came from — made by the mesh or accepted from a person — and, for a
// credential the module requires from a provider, which node provides it and the local name it
// goes by where the module keeps several.
type SecretState struct {
Name string
// Local is the credential's name inside the module (ADR 0094); empty for an own secret or the
// ordinary one.
Local string
// Origin is OriginMade or OriginAccepted.
Origin string
// Provider is the node providing a required secret; empty for the module's own.
Provider string
}
// Own says the secret is the module's own rather than one it requires from a provider.
func (s SecretState) Own() bool { return s.Provider == "" }
// SecretsOf is every secret a module holds on a machine: its own, and each credential it requires
// from a provider — with where each value came from. What a take reads to refuse minting over a
// service that already has one (ADR 0163, rule 2).
func (i *Inventory) SecretsOf(ctx context.Context, node, module string) ([]SecretState, error) {
record, err := i.NodeByName(ctx, node)
if err != nil {
return nil, err
}
rows, err := i.store.Pool().Query(ctx,
`select name, '' as local, origin, '' as provider from module_secret
where node = $1 and module = $2
union all
select s.name, s.local, s.origin, p.name from secret s
join node p on p.id = s.provider
where s.consumer = $1 and s.consumer_module = $2
order by 4, 1, 2`, record.ID, module)
if err != nil {
return nil, err
}
defer rows.Close()
var out []SecretState
for rows.Next() {
var s SecretState
if err := rows.Scan(&s.Name, &s.Local, &s.Origin, &s.Provider); err != nil {
return nil, err
}
out = append(out, s)
}
return out, rows.Err()
}
+44
View File
@@ -934,3 +934,47 @@ func TestASharedCredentialIsOneValueSealedToEveryHolder(t *testing.T) {
t.Fatalf("a consumer binding after an acceptance must be refused with the way out: %v", err)
}
}
// SecretsOf is every secret a module holds on a machine with where each came from — what a take
// reads to refuse minting over a service that already has a value (novox/hq ADR 0163, rule 2).
func TestSecretsOfSaysEachSecretsOriginAndProvider(t *testing.T) {
inv, ctx := twoNodesWithKeys(t)
if err := inv.RegisterModule(ctx, catalogue.Manifest{Module: "forge", Version: "1",
Requires: []string{"secret", "postgres-database"},
Secrets: map[string]string{"secret": "/run/secret", "postgres-database": "/run/pg"},
OwnSecrets: catalogue.OwnSecrets{"admin": {Path: "/run/admin"}}}, Source{}); err != nil {
t.Fatal(err)
}
if _, err := inv.SecretForModule(ctx, "consumer", "forge", "admin"); err != nil {
t.Fatal(err)
}
if _, err := inv.SecretFor(ctx, "postgres-database", "consumer", "forge", "provider", ""); err != nil {
t.Fatal(err)
}
if err := inv.AcceptSecretForPair(ctx, "secret", "consumer", "forge", "provider", "", "hunter2"); err != nil {
t.Fatal(err)
}
got, err := inv.SecretsOf(ctx, "consumer", "forge")
if err != nil {
t.Fatal(err)
}
want := []SecretState{
{Name: "admin", Origin: OriginMade},
{Name: "postgres-database", Origin: OriginMade, Provider: "provider"},
{Name: "secret", Origin: OriginAccepted, Provider: "provider"},
}
if len(got) != len(want) {
t.Fatalf("got %+v", got)
}
for i := range want {
if got[i] != want[i] {
t.Errorf("secret %d: got %+v, want %+v", i, got[i], want[i])
}
}
if !got[0].Own() || got[1].Own() {
t.Error("own and required are not told apart")
}
if other, _ := inv.SecretsOf(ctx, "consumer", "gitea"); len(other) != 0 {
t.Fatalf("another module's secrets: %+v", other)
}
}
@@ -0,0 +1,69 @@
package inventory
import (
"strings"
"testing"
"github.com/novox/mesh-controller/internal/catalogue"
)
// A setting is judged where it is stored (novox/hq ADR 0163, rule 6): one that cannot compose with
// the module's definition is refused naming the node, the module, the layer and the key, and is not
// kept; one that reaches nothing is refused the same way.
func TestASettingIsJudgedWhereItIsStored(t *testing.T) {
inv := fresh(t)
ctx := t.Context()
if _, err := inv.AddNode(ctx, "anchor"); err != nil {
t.Fatal(err)
}
web := catalogue.Manifest{Module: "web", Version: "1",
Resources: []map[string]any{
{"id": "server", "type": "container", "name": "web", "image": "x", "ports": []any{"8080"}},
{"id": "conf", "type": "file", "path": "/etc/web.json", "content": "{}", "merge": "json"},
}}
plain := catalogue.Manifest{Module: "plain", Version: "1",
Resources: []map[string]any{{"id": "server", "type": "container", "name": "plain", "image": "x"}}}
for _, m := range []catalogue.Manifest{web, plain} {
if err := inv.RegisterModule(ctx, m, Source{}); err != nil {
t.Fatal(err)
}
}
err := inv.SetSettings(ctx, "anchor", "web", map[string]any{catalogue.PortsSetting: map[string]any{"9999": 10000}})
for _, want := range []string{"refused: web on anchor", `layer "anchor"`, "9999"} {
if err == nil || !strings.Contains(err.Error(), want) {
t.Errorf("a port the module does not publish: %v, want %q", err, want)
}
}
if layers, _ := inv.SettingsFor(ctx, "anchor", "web"); len(layers) != 0 {
t.Fatalf("the refused layer was stored: %v", layers)
}
// A key that reaches nothing is refused too, where the spelling can still be fixed; a module
// with a mergeable file takes any key.
err = inv.SetSettings(ctx, "", "plain", map[string]any{"colour": "blue"})
if err == nil || !strings.Contains(err.Error(), "reach nothing") || !strings.Contains(err.Error(), `"colour"`) {
t.Fatalf("a stray key was stored: %v", err)
}
if err := inv.SetSettings(ctx, "", "web", map[string]any{"colour": "blue"}); err != nil {
t.Fatal(err)
}
// The mesh-wide layer is under the node's when the node's is judged.
if err := inv.SetSettings(ctx, "anchor", "web", map[string]any{catalogue.PortsSetting: map[string]any{"8080": 10000}}); err != nil {
t.Fatal(err)
}
// A kept network is for an adopted machine only (rule 4).
keep := map[string]any{catalogue.NetworksSetting: map[string]any{"server": []any{"predecessor_default"}}}
err = inv.SetSettings(ctx, "anchor", "plain", keep)
if err == nil || !strings.Contains(err.Error(), "anchor is converged") {
t.Fatalf("a kept network on a converged machine was stored: %v", err)
}
if err := inv.SetAdopted(ctx, "anchor", true); err != nil {
t.Fatal(err)
}
if err := inv.SetSettings(ctx, "anchor", "plain", keep); err != nil {
t.Fatal(err)
}
if err := inv.SetSettings(ctx, "anchor", "nothing", keep); err == nil {
t.Fatal("a setting for a module the mesh does not know was stored")
}
}
+17
View File
@@ -305,6 +305,23 @@ func (e Enrolment) Heard(ctx context.Context, report Report) (news bool, err err
return false, err
}
}
// What filters the machine, and the state of its found firewall (novox/hq ADR 0168), whenever
// it says — every apply of a host that knows how, adopted or converged; never cleared by a
// report that carries none, which is every bare word that the node is there.
if len(report.Filters) > 0 || report.FoundFirewall != nil {
filters := make([]inventory.Filter, 0, len(report.Filters))
for _, f := range report.Filters {
filters = append(filters, inventory.Filter{Where: f.Where, Owner: f.Owner, Refuses: f.Refuses})
}
var found *inventory.FoundFirewall
if report.FoundFirewall != nil {
found = &inventory.FoundFirewall{Kind: report.FoundFirewall.Kind, Active: report.FoundFirewall.Active,
RetiredBy: report.FoundFirewall.RetiredBy}
}
if err := e.Inventory.RecordFiltering(ctx, node.ID, filters, found); err != nil {
return false, err
}
}
// Which of its links face outside (novox/hq ADR 0140), whenever it says so. Recorded on every
// report that carries it, adopted or converged, because the filter the mesh composes is written
// around it — and never cleared by a report that carries none, which is every bare word that the
+42
View File
@@ -218,3 +218,45 @@ func TestWhatAnAdoptedNodeHoldsIsKeptAndAnAliveWordDoesNotWipeIt(t *testing.T) {
t.Fatalf("a report from an adopted node holding nothing did not empty held: %+v", got)
}
}
// What filters a machine, and the state of its found firewall, are kept from every report that
// carries them and never cleared by one that does not (novox/hq ADR 0168).
func TestWhatFiltersAMachineIsKeptFromItsReport(t *testing.T) {
inv, _, _ := heardFrom(t, link.Report{
Node: "home-server", Applied: []string{"a"},
Filters: []link.Filter{
{Where: "table inet mesh, chain forward", Owner: "mesh", Refuses: "policy drop"},
{Where: "chain HAL-MESH-ONLY (iptables-legacy)", Owner: "other", Refuses: "-j DROP"},
},
FoundFirewall: &link.FoundFirewall{Kind: "ufw", Active: false, RetiredBy: "found-inactive"},
})
ctx := context.Background()
f, err := inv.FilteringOf(ctx, "home-server")
if err != nil {
t.Fatal(err)
}
if len(f.Filters) != 2 || f.Filters[1].Owner != inventory.FilterOther || f.Alone() {
t.Fatalf("recorded %+v", f)
}
if f.FoundFirewall == nil || f.FoundFirewall.RetiredBy != "found-inactive" || f.FoundFirewall.Active {
t.Fatalf("the found firewall's state: %+v", f.FoundFirewall)
}
if len(f.Others()) != 1 || f.Others()[0].Where != "chain HAL-MESH-ONLY (iptables-legacy)" {
t.Fatalf("others: %+v", f.Others())
}
// A bare word that the node is there clears nothing.
if _, err := (link.Enrolment{Inventory: inv}).Heard(ctx, link.Report{Node: "home-server"}); err != nil {
t.Fatal(err)
}
if again, _ := inv.FilteringOf(ctx, "home-server"); len(again.Filters) != 2 {
t.Fatalf("a bare report cleared what filters the machine: %+v", again)
}
// The next full report replaces it: the chain removed by hand is gone from the record.
if _, err := (link.Enrolment{Inventory: inv}).Heard(ctx, link.Report{Node: "home-server", Applied: []string{"a"},
Filters: []link.Filter{{Where: "table inet mesh, chain forward", Owner: "mesh", Refuses: "policy drop"}}}); err != nil {
t.Fatal(err)
}
if again, _ := inv.FilteringOf(ctx, "home-server"); len(again.Filters) != 1 || !again.Alone() {
t.Fatalf("the next report did not replace what filters the machine: %+v", again)
}
}
+24
View File
@@ -197,6 +197,15 @@ type Report struct {
// Strays is what runs on the machine that the mesh neither wrote nor holds (ADR 0163).
Strays []Stray `json:"strays,omitempty"`
// Filters is what filters the machine now: every table and chain that refuses traffic, with
// its owner — the mesh's, the found firewall's, the container runtime's own, a ban, or other
// (novox/hq ADR 0168). Every machine reports it, adopted or converged; absent from a host older
// than this.
Filters []Filter `json:"filters,omitempty"`
// FoundFirewall is the state of the firewall a converged machine was found with: in force now
// or not, and how it came to be inactive — the mesh disabled it, or it was found so (ADR 0168).
FoundFirewall *FoundFirewall `json:"found_firewall,omitempty"`
// Profile is what the machine can do, detected again by this apply (novox/hq ADR 0161): the
// same shape enrolment sends, so a machine that gained or lost a capability — switched its
// network manager — is known at its next push and not at its next enrolment. Absent from a host
@@ -278,6 +287,21 @@ type Held struct {
Facts map[string]any `json:"facts,omitempty"`
}
// A Filter is one place on a machine that refuses traffic, with its owner (novox/hq ADR 0168):
// the host's own shape, carried as data.
type Filter struct {
Where string `json:"where"`
Owner string `json:"owner"`
Refuses string `json:"refuses"`
}
// FoundFirewall is the state of a converged machine's found firewall (ADR 0168).
type FoundFirewall struct {
Kind string `json:"kind"`
Active bool `json:"active"`
RetiredBy string `json:"retired_by,omitempty"`
}
// A Stray is a container a machine runs that the mesh neither wrote nor holds (ADR 0163).
type Stray struct {
Kind string `json:"kind"`