Make pending assignments safe to race, settle them on a tick, and say only what was checked
Review of #150: a withdrawal could land between the look and the act, a failed ask read as a build in flight, a request kept the wrong asker, a build being registered read as not built, build "true" could ask a build nothing waited on, and a status read changed state. Claim a row under the machine's hold before making it, keep a request only once asked, settle on the controller's own tick, raise an assignment not made as a condition until it is answered, and tie each row to its machine.
This commit is contained in:
+346
-93
@@ -9,11 +9,12 @@ import (
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
"github.com/novox/mesh-controller/internal/conditions"
|
||||
"github.com/novox/mesh-controller/internal/inventory"
|
||||
"github.com/novox/mesh-controller/internal/link"
|
||||
)
|
||||
|
||||
// A module not registered yet, and the assignment that waits for it (novox/hq issue 325).
|
||||
// A module not registered yet, and the assignment that waits for it (novox/hq issue 325, ADR 0261).
|
||||
//
|
||||
// On 2026-10-08 the catalogue's pull request adding `sensors` merged; a minute later `assign g14 sensors`
|
||||
// answered "no module of that name: sensors", and a few minutes later the same call worked. The merge had
|
||||
@@ -23,34 +24,55 @@ import (
|
||||
// So an assignment of a module the catalogue does not hold looks further before it refuses, and says which
|
||||
// of three cases it is in:
|
||||
//
|
||||
// - **a build is in flight**: a build request for the module's directory has no outcome yet. The
|
||||
// assignment is kept as a **pending assignment** and made when the build registers the module.
|
||||
// - **known, not built**: the mesh asked for the directory before, and the build failed, was not
|
||||
// - **a build is in flight**: a build request for the module's directory has no outcome yet, or its build
|
||||
// succeeded a moment ago and is being registered. The assignment is kept as a **pending assignment**,
|
||||
// which the controller makes when the build registers the module (ADR 0261).
|
||||
// - **known, not built**: the controller asked for the directory before, and the build failed, was not
|
||||
// registered, said nothing within its bound, or could not be asked. Said, with `build` and assign's
|
||||
// own build argument, which asks for the build and keeps the assignment pending on it.
|
||||
// - **unknown**: no module, build request or merge by that name. Refused as before, with the closest names.
|
||||
// - **unknown**: no module registered and no build request kept by that name. Refused as before, with the
|
||||
// closest names.
|
||||
//
|
||||
// **Pending by default while a build is in flight**, without an argument to ask for it: an assignment is
|
||||
// what a person meant and is kept even when its machine does not resolve (acts.go), nothing is sent until a
|
||||
// push, and `unassign` withdraws it. Asking for a second act once the build lands is exactly the two acts by
|
||||
// hand issue 300 removed. A build is never asked for by default: it runs on another machine, and a directory
|
||||
// whose last build failed is a fault to look at before it is a thing to retry.
|
||||
// **Pending by default while a build is in flight** (ADR 0261), without an argument to ask for it: an
|
||||
// assignment is what a person meant and is kept even when its machine does not resolve (acts.go), and
|
||||
// `unassign` withdraws it. Asking for a second act once the build lands is the two acts by hand issue 300
|
||||
// removed. A build is never asked for by default: it runs on another machine, and a directory whose last
|
||||
// build failed is a fault to look at before it is a thing to retry.
|
||||
//
|
||||
// **Settling is the controller's tick, never a read** (settlingPending): `status`, the board and the summary
|
||||
// only read the rows.
|
||||
|
||||
// buildRequestBound is how long a build request may go without an outcome before it is no longer read as in
|
||||
// flight, and a pending assignment waiting for it expires. Generous: a build waits behind others on the
|
||||
// build seat, and a pending assignment that waits a little longer costs nothing.
|
||||
const buildRequestBound = 2 * time.Hour
|
||||
|
||||
// pendingShownFor is how long an ended pending assignment is said in `status`.
|
||||
// registerGrace is how long a build heard as built is read as still in flight while it is not registered:
|
||||
// the outcome is recorded first and registered after it, in the same take-in. Past it, the build was
|
||||
// recorded and not registered, and says so.
|
||||
const registerGrace = 5 * time.Minute
|
||||
|
||||
// pendingShownFor is how long an ended pending assignment is listed in `status`.
|
||||
const pendingShownFor = 24 * time.Hour
|
||||
|
||||
// settleEvery is how often the controller settles the pending assignments.
|
||||
const settleEvery = time.Minute
|
||||
|
||||
// claimStaleAfter is how long a pending assignment may be held as applying before the tick reads the claim
|
||||
// as left by a controller that stopped, and settles it from what the mesh holds.
|
||||
const claimStaleAfter = 5 * time.Minute
|
||||
|
||||
// kindPendingEnded is a pending assignment that ended without being made: expired or refused.
|
||||
const kindPendingEnded = "assignment-not-made"
|
||||
|
||||
// assignOptions are what an assignment was asked with beside its machine and modules.
|
||||
type assignOptions struct {
|
||||
// Build asks for the build of a module known and not built, and keeps the assignment pending on it.
|
||||
Build bool
|
||||
}
|
||||
|
||||
// recordBuildRequest keeps a build request so `assign` can find it. Said, never fatal: the build was asked.
|
||||
// recordBuildRequest keeps a build request so `assign` can find it, once it is asked. Said, never fatal: the
|
||||
// build was asked.
|
||||
func recordBuildRequest(ctx context.Context, inv *inventory.Inventory, r inventory.BuildRequest) {
|
||||
if err := inv.RecordBuildRequest(ctx, r); err != nil {
|
||||
fmt.Fprintf(os.Stderr, "build %s was asked, and could not be kept as a build request, so `assign` "+
|
||||
@@ -73,25 +95,30 @@ type notHeld struct {
|
||||
said string
|
||||
}
|
||||
|
||||
// whereIs looks for a module the catalogue does not hold among the build requests the mesh keeps.
|
||||
// whereIs looks for a module the catalogue does not hold among the build requests the controller keeps.
|
||||
func whereIs(ctx context.Context, inv *inventory.Inventory, module string, now time.Time) (notHeld, error) {
|
||||
requests, err := inv.RequestsNamed(ctx, module)
|
||||
if err != nil {
|
||||
return notHeld{}, err
|
||||
}
|
||||
if r, ok := inFlight(requests, now); ok {
|
||||
return notHeld{kind: buildInFlight, request: r, said: fmt.Sprintf("%s is not registered yet: it is being "+
|
||||
"built from %s since %s, as build %s; it can be assigned once that build registers it", module,
|
||||
being := "being built"
|
||||
if r.Heard {
|
||||
being = "built and being registered"
|
||||
}
|
||||
return notHeld{kind: buildInFlight, request: r, said: fmt.Sprintf("%s is not registered yet: it is %s "+
|
||||
"from %s since %s, as build %s; it can be assigned once that build registers it", module, being,
|
||||
requestSource(r), clock(r.At), r.ID)}, nil
|
||||
}
|
||||
if len(requests) > 0 {
|
||||
return notHeld{kind: knownNotBuilt, request: requests[0], said: fmt.Sprintf("%s is known and not "+
|
||||
"registered: %s", module, whyNotBuilt(requests[0], now))}, nil
|
||||
}
|
||||
said := fmt.Sprintf("%v: %s — the mesh holds no module, build request or merge by that name",
|
||||
inventory.ErrNoSuchModule, module)
|
||||
// Only what was looked at is claimed: the catalogue, and the build requests the controller keeps.
|
||||
said := fmt.Sprintf("%v: %s — the catalogue holds no module of that name, and no build request the "+
|
||||
"controller kept (the last 30 days) is for a directory of that name", inventory.ErrNoSuchModule, module)
|
||||
if near := closestNames(ctx, inv, module); len(near) > 0 {
|
||||
said += "; the closest it holds: " + strings.Join(near, ", ")
|
||||
said += "; the closest names it holds: " + strings.Join(near, ", ")
|
||||
}
|
||||
return notHeld{kind: unknownModule, said: said}, nil
|
||||
}
|
||||
@@ -111,6 +138,20 @@ func notInCatalogue(ctx context.Context, open *stores, modules []string) ([]stri
|
||||
return missing, nil
|
||||
}
|
||||
|
||||
// openPending is the open pending assignment of a module to a machine, if there is one.
|
||||
func openPending(ctx context.Context, inv *inventory.Inventory, node, module string) (*inventory.PendingAssignment, error) {
|
||||
rows, err := inv.PendingFor(ctx, node, module)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
for i := range rows {
|
||||
if rows[i].Open() {
|
||||
return &rows[i], nil
|
||||
}
|
||||
}
|
||||
return nil, nil
|
||||
}
|
||||
|
||||
// notRegistered answers an assignment of a module the catalogue does not hold: pending on a build in flight,
|
||||
// known and not built, or unknown (novox/hq issue 325). An answer with no error is a pending assignment kept.
|
||||
func notRegistered(ctx context.Context, open *stores, node, module string, opts assignOptions) (string, error) {
|
||||
@@ -123,19 +164,31 @@ func notRegistered(ctx context.Context, open *stores, node, module string, opts
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
waiting, err := openPending(ctx, inv, node, module)
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
switch where.kind {
|
||||
case buildInFlight:
|
||||
lead := where.said
|
||||
if opts.Build {
|
||||
lead += "\n no second build was asked: this one is running"
|
||||
}
|
||||
return keepPending(ctx, open, node, module, where.request, lead)
|
||||
return keepPending(ctx, open, node, module, where.request, waiting, lead)
|
||||
case knownNotBuilt:
|
||||
last := where.request
|
||||
if !opts.Build {
|
||||
return "", fmt.Errorf("%w: %s\n assign it again with build \"true\" to ask for its build and keep this "+
|
||||
also := ""
|
||||
if waiting != nil {
|
||||
also = fmt.Sprintf("\n %s already has a pending assignment of %s, waiting for build %s; build "+
|
||||
"\"true\" asks for a new build and makes it wait for that one", node, module, waiting.Build)
|
||||
}
|
||||
return "", fmt.Errorf("%w: %s%s\n assign it again with build \"true\" to ask for its build and keep this "+
|
||||
"assignment until the build registers it; or `build` it (repository %s, path %s) and assign it "+
|
||||
"once it is registered", inventory.ErrNoSuchModule, where.said, last.Repository, orRoot(last.Path))
|
||||
"once it is registered", inventory.ErrNoSuchModule, where.said, also, last.Repository, orRoot(last.Path))
|
||||
}
|
||||
if waiting != nil && waiting.State == inventory.PendingApplying {
|
||||
return "", fmt.Errorf("%s is being assigned %s now; nothing was asked", node, module)
|
||||
}
|
||||
source := buildSource{Repository: last.Repository, Seat: last.Seat}
|
||||
id, err := askABuild(ctx, source, last.Path, last.Ref)
|
||||
@@ -147,20 +200,38 @@ func notRegistered(ctx context.Context, open *stores, node, module string, opts
|
||||
recordBuildRequest(ctx, inv, asked.BuildRequest)
|
||||
lead := fmt.Sprintf("%s\n build %s of %s is asked for now; %s can be assigned once it registers it",
|
||||
where.said, id, requestSource(asked), module)
|
||||
return keepPending(ctx, open, node, module, asked, lead)
|
||||
return keepPending(ctx, open, node, module, asked, waiting, lead)
|
||||
}
|
||||
answer := strings.TrimPrefix(where.said, inventory.ErrNoSuchModule.Error()) +
|
||||
"\n a module in a repository is built with `build` (its repository and path), then assigned"
|
||||
if opts.Build {
|
||||
answer += "\n build \"true\" asks for nothing here: the mesh does not know where a module of that name is"
|
||||
answer += "\n build \"true\" asks for nothing here: the controller has no build request saying where a " +
|
||||
"module of that name is"
|
||||
}
|
||||
return "", fmt.Errorf("%w%s", inventory.ErrNoSuchModule, answer)
|
||||
}
|
||||
|
||||
// keepPending records the assignment as pending on a build request, and says so.
|
||||
func keepPending(ctx context.Context, open *stores, node, module string, r inventory.RequestOutcome, lead string) (string, error) {
|
||||
// keepPending records the assignment as pending on a build request — or, where one is already open, makes it
|
||||
// wait for that build — and says so.
|
||||
func keepPending(ctx context.Context, open *stores, node, module string, r inventory.RequestOutcome,
|
||||
waiting *inventory.PendingAssignment, lead string) (string, error) {
|
||||
inv := open.inventory
|
||||
p, err := inv.RecordPending(ctx, inventory.PendingAssignment{Node: node, Module: module, Build: r.ID,
|
||||
if waiting != nil {
|
||||
if waiting.Build == r.ID {
|
||||
return lead + fmt.Sprintf("\n %s already waits for %s on this build: nothing changed", node, module), nil
|
||||
}
|
||||
moved, err := inv.RepointPending(ctx, waiting.ID, r.ID, r.Repository, r.Path)
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
if !moved {
|
||||
return lead + fmt.Sprintf("\n the pending assignment of %s to %s ended while this was asked: `status` "+
|
||||
"says how", module, node), nil
|
||||
}
|
||||
return lead + fmt.Sprintf("\n the pending assignment of %s to %s, which waited for build %s, now waits "+
|
||||
"for build %s", module, node, waiting.Build, r.ID), nil
|
||||
}
|
||||
_, err := inv.RecordPending(ctx, inventory.PendingAssignment{Node: node, Module: module, Build: r.ID,
|
||||
Repository: r.Repository, Path: r.Path})
|
||||
if errors.Is(err, inventory.ErrAlreadyPending) {
|
||||
return lead + fmt.Sprintf("\n %s already waits for %s: nothing changed", node, module), nil
|
||||
@@ -168,51 +239,63 @@ func keepPending(ctx context.Context, open *stores, node, module string, r inven
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
answer := lead + fmt.Sprintf("\n the assignment is kept as pending: %s is assigned %s when the build "+
|
||||
"registers it, and `status` says it until then. If the build fails it expires and says why; "+
|
||||
"`unassign %s %s` withdraws it", node, module, node, module)
|
||||
// The build may have registered the module between the look and the record: made at once, then.
|
||||
if shelf, err := inv.Catalogue(ctx); err == nil {
|
||||
if _, now := shelf[module]; now {
|
||||
if line := applyPending(ctx, open, p, r.ID); line != "" {
|
||||
answer += "\n " + line
|
||||
}
|
||||
}
|
||||
}
|
||||
return answer, nil
|
||||
// The controller makes it, not this act (ADR 0261): when the build registers the module, or at its next
|
||||
// tick if the module was registered between the look and the record.
|
||||
return lead + fmt.Sprintf("\n the assignment is kept as pending: the controller assigns %s to %s when the "+
|
||||
"build registers it, and `status` lists it until then. If the build fails it expires, says why and "+
|
||||
"raises a condition; `unassign %s %s` withdraws it", module, node, node, module), nil
|
||||
}
|
||||
|
||||
// inFlight is the newest build request for the module with no outcome yet, asked within the bound.
|
||||
// inFlight is the newest build request for the module still to register it: no outcome yet within the
|
||||
// bound, or a successful outcome heard a moment ago and being registered.
|
||||
func inFlight(requests []inventory.RequestOutcome, now time.Time) (inventory.RequestOutcome, bool) {
|
||||
for _, r := range requests {
|
||||
if r.NotAsked == "" && !r.Heard && now.Sub(r.At) < buildRequestBound {
|
||||
if stillComing(r, now) {
|
||||
return r, true
|
||||
}
|
||||
}
|
||||
return inventory.RequestOutcome{}, false
|
||||
}
|
||||
|
||||
// whyNotBuilt says what came of a module's last build request.
|
||||
func stillComing(r inventory.RequestOutcome, now time.Time) bool {
|
||||
if r.Heard {
|
||||
return r.Failed == "" && now.Sub(r.HeardAt) < registerGrace
|
||||
}
|
||||
return r.NotAsked == "" && now.Sub(r.At) < buildRequestBound
|
||||
}
|
||||
|
||||
// whyNotBuilt says what came of a module's last build request. What was heard comes first: an outcome is the
|
||||
// last word whatever its asker said.
|
||||
func whyNotBuilt(r inventory.RequestOutcome, now time.Time) string {
|
||||
where := fmt.Sprintf("%s in %s", orRoot(r.Path), r.Repository)
|
||||
switch {
|
||||
case r.NotAsked != "":
|
||||
return fmt.Sprintf("%s; the merge that added it (%s) could not ask for its build: %s", where,
|
||||
short(r.Commit), firstLine(r.NotAsked))
|
||||
case !r.Heard:
|
||||
return fmt.Sprintf("%s; build %s was asked at %s, and no outcome has been heard in %s", where, r.ID,
|
||||
clock(r.At), now.Sub(r.At).Round(time.Minute))
|
||||
case r.Failed != "":
|
||||
case r.Heard && r.Failed != "":
|
||||
return fmt.Sprintf("%s; its last build, %s at %s, failed: %s", where, r.ID, clock(r.HeardAt),
|
||||
firstLine(r.Failed))
|
||||
case r.Module != "" && r.Module != r.Name():
|
||||
return fmt.Sprintf("%s; its last build, %s, registered it as %s", where, r.ID, r.Module)
|
||||
default:
|
||||
case r.Heard && r.Module != "" && r.Module != r.Name():
|
||||
return fmt.Sprintf("%s; its last build, %s, built it as %s", where, r.ID, r.Module)
|
||||
case r.Heard:
|
||||
return fmt.Sprintf("%s; its last build, %s at %s, was recorded and not registered — `builds` says why",
|
||||
where, r.ID, clock(r.HeardAt))
|
||||
case r.NotAsked != "" && r.For == "merge":
|
||||
return fmt.Sprintf("%s; the merge that added it (%s) could not ask for its build: %s", where,
|
||||
short(r.Commit), firstLine(r.NotAsked))
|
||||
case r.NotAsked != "":
|
||||
return fmt.Sprintf("%s; build %s, asked by %s, was not handed over or not waited for: %s", where, r.ID,
|
||||
askerOr(r.For), firstLine(r.NotAsked))
|
||||
default:
|
||||
return fmt.Sprintf("%s; build %s was asked at %s, and no outcome has been heard in %s", where, r.ID,
|
||||
clock(r.At), now.Sub(r.At).Round(time.Minute))
|
||||
}
|
||||
}
|
||||
|
||||
func askerOr(who string) string {
|
||||
if who == "" {
|
||||
return "the controller"
|
||||
}
|
||||
return who
|
||||
}
|
||||
|
||||
// requestSource is a build request's source as a person reads it: repository@commit (directory).
|
||||
func requestSource(r inventory.RequestOutcome) string {
|
||||
at := short(r.Commit)
|
||||
@@ -228,8 +311,8 @@ func requestSource(r inventory.RequestOutcome) string {
|
||||
// clock is a moment as a person reads it, in the controller's local time.
|
||||
func clock(t time.Time) string { return t.Local().Format("2006-01-02 15:04:05 MST") }
|
||||
|
||||
// closestNames is up to three names the mesh holds that are near the one asked: registered modules and
|
||||
// directories it asked to build.
|
||||
// closestNames is up to three names near the one asked: registered modules and directories the controller
|
||||
// asked to build.
|
||||
func closestNames(ctx context.Context, inv *inventory.Inventory, name string) []string {
|
||||
var candidates []string
|
||||
if shelf, err := inv.Catalogue(ctx); err == nil {
|
||||
@@ -292,26 +375,41 @@ func editDistance(a, b string) int {
|
||||
return prev[len(rb)]
|
||||
}
|
||||
|
||||
// applyPending makes a pending assignment now that its module is registered, and settles it: applied, or
|
||||
// refused with the refusal. Returns what it said, or nothing when another process settled it first.
|
||||
// applyPending makes a pending assignment now that its module is registered (ADR 0261). Under the machine's
|
||||
// hold it claims the row (waiting → applying), so a withdrawal cannot land between the look and the act,
|
||||
// then assigns by the same act a person's assign is, then says what came of it: applied, or refused with the
|
||||
// refusal. Returns what it said, or nothing when the row no longer waited.
|
||||
func applyPending(ctx context.Context, open *stores, p inventory.PendingAssignment, by string) string {
|
||||
inv := open.inventory
|
||||
ctx, release, err := holdNodes(ctx, open, []string{p.Node})
|
||||
if err != nil {
|
||||
// Left waiting: the next tick tries again.
|
||||
return fmt.Sprintf("the pending assignment of %s to %s waits: %s could not be held: %v", p.Module, p.Node,
|
||||
p.Node, err)
|
||||
}
|
||||
defer release()
|
||||
claimed, err := inv.ClaimPending(ctx, p.ID)
|
||||
if err != nil {
|
||||
return fmt.Sprintf("the pending assignment of %s to %s could not be taken for making: %v", p.Module, p.Node, err)
|
||||
}
|
||||
if !claimed {
|
||||
return ""
|
||||
}
|
||||
answer, err := assign(ctx, open, p.Node, p.Module)
|
||||
state, note := inventory.PendingApplied, ""
|
||||
if err != nil {
|
||||
switch {
|
||||
case err != nil:
|
||||
state = inventory.PendingRefused
|
||||
note = fmt.Sprintf("the pending assignment of %s to %s was refused when build %s registered it: %s",
|
||||
p.Module, p.Node, by, firstLine(err.Error()))
|
||||
} else {
|
||||
// Not promised as sent: a walk sends it if the module's upgrade announcement finds it
|
||||
// assigned, and nothing does if the announcement came first. `status` says which, as for any machine.
|
||||
note = fmt.Sprintf("%s is assigned %s: build %s registered it (pending since %s); `status` says whether "+
|
||||
"%s has been sent it, and `push %s` sends it if not", p.Node, p.Module, by, clock(p.Since), p.Node, p.Node)
|
||||
if strings.Contains(answer, "already runs") {
|
||||
note = fmt.Sprintf("%s already runs %s: the pending assignment is met", p.Node, p.Module)
|
||||
}
|
||||
case strings.Contains(answer, "already runs"):
|
||||
note = fmt.Sprintf("%s already runs %s: the pending assignment is met", p.Node, p.Module)
|
||||
default:
|
||||
// The same as a person's assign: nothing is sent by this act.
|
||||
note = fmt.Sprintf("%s is assigned %s: build %s registered it (pending since %s); `push %s` sends it",
|
||||
p.Node, p.Module, by, clock(p.Since), p.Node)
|
||||
}
|
||||
settled, serr := inv.SettlePending(ctx, p.ID, state, note)
|
||||
settled, serr := inv.SettlePending(ctx, p.ID, inventory.PendingApplying, state, note)
|
||||
if serr != nil {
|
||||
return fmt.Sprintf("%s — and it could not be recorded as settled: %v", note, serr)
|
||||
}
|
||||
@@ -321,11 +419,11 @@ func applyPending(ctx context.Context, open *stores, p inventory.PendingAssignme
|
||||
return note
|
||||
}
|
||||
|
||||
// expirePending ends a pending assignment whose build will not register its module, with why.
|
||||
// expirePending ends a waiting pending assignment whose build will not register its module, with why.
|
||||
func expirePending(ctx context.Context, inv *inventory.Inventory, p inventory.PendingAssignment, why string) string {
|
||||
note := fmt.Sprintf("the pending assignment of %s to %s expired: %s; nothing was assigned. Assign it again with "+
|
||||
"build \"true\" to ask for the build again", p.Module, p.Node, why)
|
||||
settled, err := inv.SettlePending(ctx, p.ID, inventory.PendingExpired, note)
|
||||
settled, err := inv.SettlePending(ctx, p.ID, inventory.PendingWaiting, inventory.PendingExpired, note)
|
||||
if err != nil {
|
||||
return fmt.Sprintf("%s — and it could not be recorded: %v", note, err)
|
||||
}
|
||||
@@ -369,21 +467,62 @@ func settlePendingOnBuild(ctx context.Context, open *stores, result link.BuildRe
|
||||
return said
|
||||
}
|
||||
|
||||
// settleStalePending settles what a missed outcome left waiting: a module registered meanwhile is assigned,
|
||||
// a build heard and not registered ends its assignment, and one with no outcome past the bound expires.
|
||||
// Asked where the mesh is read for status, so a pending assignment never waits silently for ever.
|
||||
func settleStalePending(ctx context.Context, open *stores, now time.Time) []string {
|
||||
inv := open.inventory
|
||||
waiting, err := inv.Pending(ctx, time.Time{})
|
||||
if err != nil {
|
||||
// Said, not swallowed; status reads the pending assignments again and says the same failure there.
|
||||
return []string{fmt.Sprintf("the pending assignments could not be read to settle them: %v", err)}
|
||||
// settlingPending is the controller's tick over the pending assignments (ADR 0261), every settleEvery until
|
||||
// the context ends. What it does is printed to the controller's log, kept in each row's note (which `status`
|
||||
// lists for a day), and raised as a condition for an assignment that was not made.
|
||||
func settlingPending(ctx context.Context, open *stores) {
|
||||
for {
|
||||
for _, line := range settlePending(ctx, open, time.Now()) {
|
||||
fmt.Println(line)
|
||||
}
|
||||
select {
|
||||
case <-ctx.Done():
|
||||
return
|
||||
case <-time.After(settleEvery):
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// settlePending is one tick: claims left by a controller that stopped are settled from what the mesh holds;
|
||||
// a module registered meanwhile is assigned; a build heard and not registered, or silent past its bound,
|
||||
// ends its assignment; ended ones are raised as conditions and cleared once answered; ended rows older than
|
||||
// KeptFor are deleted. Returns what it did.
|
||||
func settlePending(ctx context.Context, open *stores, now time.Time) []string {
|
||||
inv := open.inventory
|
||||
var said []string
|
||||
say := func(format string, args ...any) { said = append(said, fmt.Sprintf(format, args...)) }
|
||||
|
||||
shelf, err := inv.Catalogue(ctx)
|
||||
if err != nil {
|
||||
return []string{fmt.Sprintf("the catalogue could not be read to settle the pending assignments: %v", err)}
|
||||
say("the pending assignments are not settled this time: the catalogue could not be read: %v", err)
|
||||
return said
|
||||
}
|
||||
stuck, err := inv.Stuck(ctx, now.Add(-claimStaleAfter))
|
||||
if err != nil {
|
||||
say("claims left by a stopped controller could not be read: %v", err)
|
||||
}
|
||||
for _, p := range stuck {
|
||||
assigned, err := inv.Assigned(ctx, p.Node)
|
||||
if err != nil {
|
||||
say("the claim on %s for %s could not be settled: %v", p.Node, p.Module, err)
|
||||
continue
|
||||
}
|
||||
if containsString(assigned, p.Module) {
|
||||
if _, err := inv.SettlePending(ctx, p.ID, inventory.PendingApplying, inventory.PendingApplied,
|
||||
fmt.Sprintf("%s is assigned %s (settled after the controller that made it stopped)", p.Node, p.Module)); err != nil {
|
||||
say("the claim on %s for %s could not be settled: %v", p.Node, p.Module, err)
|
||||
}
|
||||
continue
|
||||
}
|
||||
if err := inv.ReleaseClaim(ctx, p.ID); err != nil {
|
||||
say("the claim on %s for %s could not be released: %v", p.Node, p.Module, err)
|
||||
}
|
||||
}
|
||||
|
||||
waiting, err := inv.Pending(ctx, time.Time{})
|
||||
if err != nil {
|
||||
say("the pending assignments could not be read to settle them: %v", err)
|
||||
}
|
||||
var said []string
|
||||
for _, p := range waiting {
|
||||
if _, ok := shelf[p.Module]; ok {
|
||||
if line := applyPending(ctx, open, p, p.Build); line != "" {
|
||||
@@ -393,7 +532,7 @@ func settleStalePending(ctx context.Context, open *stores, now time.Time) []stri
|
||||
}
|
||||
requests, err := inv.RequestsNamed(ctx, p.Module)
|
||||
if err != nil {
|
||||
said = append(said, fmt.Sprintf("the build requests for %s could not be read: %v", p.Module, err))
|
||||
say("the build requests for %s could not be read: %v", p.Module, err)
|
||||
continue
|
||||
}
|
||||
var r *inventory.RequestOutcome
|
||||
@@ -406,9 +545,9 @@ func settleStalePending(ctx context.Context, open *stores, now time.Time) []stri
|
||||
switch {
|
||||
case r == nil && now.Sub(p.Since) > buildRequestBound:
|
||||
why = fmt.Sprintf("build %s is no longer on record", p.Build)
|
||||
case r != nil && r.Heard:
|
||||
case r != nil && !stillComing(*r, now) && (r.Heard || r.NotAsked != ""):
|
||||
why = whyNotBuilt(*r, now)
|
||||
case r != nil && now.Sub(r.At) > buildRequestBound:
|
||||
case r != nil && !stillComing(*r, now):
|
||||
why = fmt.Sprintf("no outcome of build %s was heard within %s of asking", p.Build, buildRequestBound)
|
||||
}
|
||||
if why == "" {
|
||||
@@ -418,25 +557,139 @@ func settleStalePending(ctx context.Context, open *stores, now time.Time) []stri
|
||||
said = append(said, line)
|
||||
}
|
||||
}
|
||||
|
||||
said = append(said, raisePendingEnded(ctx, inv)...)
|
||||
if n, err := inv.ForgetEndedPending(ctx, now); err != nil {
|
||||
say("ended pending assignments could not be deleted: %v", err)
|
||||
} else if n > 0 {
|
||||
say("deleted %d pending assignment(s) ended more than %s ago", n, inventory.KeptFor)
|
||||
}
|
||||
return said
|
||||
}
|
||||
|
||||
// withdrawPending is `unassign` of a module only pending on a machine: the pending assignment withdrawn.
|
||||
func withdrawPending(ctx context.Context, inv *inventory.Inventory, node, module string) (string, bool, error) {
|
||||
waiting, err := inv.Pending(ctx, time.Time{})
|
||||
func containsString(xs []string, x string) bool {
|
||||
for _, y := range xs {
|
||||
if y == x {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
// pendingObservation is the condition for a pending assignment that was not made.
|
||||
func pendingObservation(p inventory.PendingAssignment) conditions.Observation {
|
||||
return conditions.Observation{Scope: conditions.ScopeMachine, ID: p.Node + "." + p.Module,
|
||||
Token: kindPendingEnded, Kind: kindPendingEnded, Machine: p.Node, Severity: conditions.Warning,
|
||||
Resolver: conditions.ResolverOperator, Source: "pending assignments",
|
||||
Summary: p.Note}
|
||||
}
|
||||
|
||||
// raisePendingEnded raises a condition for each pending assignment that expired or was refused, once, and
|
||||
// clears it when it is answered: the module was assigned to that machine since, a newer pending assignment of
|
||||
// it is open, or a person took it back with `unassign`. Where this process keeps no conditions, nothing is
|
||||
// stamped, and the serving controller's tick raises it.
|
||||
func raisePendingEnded(ctx context.Context, inv *inventory.Inventory) []string {
|
||||
keeper := conditionsFrom
|
||||
if keeper == nil {
|
||||
return nil
|
||||
}
|
||||
var said []string
|
||||
toRaise, err := inv.ToRaise(ctx)
|
||||
if err != nil {
|
||||
return []string{fmt.Sprintf("pending assignments not made could not be read to raise them: %v", err)}
|
||||
}
|
||||
for _, p := range toRaise {
|
||||
if _, err := keeper.Observe(ctx, pendingObservation(p)); err != nil {
|
||||
said = append(said, fmt.Sprintf("the condition for %s on %s could not be raised: %v", p.Module, p.Node, err))
|
||||
continue
|
||||
}
|
||||
if err := inv.MarkPending(ctx, p.ID, "raised"); err != nil {
|
||||
said = append(said, fmt.Sprintf("the condition for %s on %s was raised and not recorded: %v", p.Module, p.Node, err))
|
||||
}
|
||||
}
|
||||
toClear, err := inv.ToClear(ctx)
|
||||
if err != nil {
|
||||
return append(said, fmt.Sprintf("pending assignments raised could not be read to clear them: %v", err))
|
||||
}
|
||||
for _, p := range toClear {
|
||||
why, answered, err := pendingAnswered(ctx, inv, p)
|
||||
if err != nil {
|
||||
said = append(said, fmt.Sprintf("whether %s on %s is answered could not be read: %v", p.Module, p.Node, err))
|
||||
continue
|
||||
}
|
||||
if !answered {
|
||||
continue
|
||||
}
|
||||
if _, err := keeper.Clear(ctx, pendingObservation(p).Key(), why); err != nil {
|
||||
said = append(said, fmt.Sprintf("the condition for %s on %s could not be cleared: %v", p.Module, p.Node, err))
|
||||
continue
|
||||
}
|
||||
if err := inv.MarkPending(ctx, p.ID, "cleared"); err != nil {
|
||||
said = append(said, fmt.Sprintf("the condition for %s on %s was cleared and not recorded: %v", p.Module, p.Node, err))
|
||||
}
|
||||
}
|
||||
return said
|
||||
}
|
||||
|
||||
// pendingAnswered says whether a pending assignment that was not made has been answered since, and how.
|
||||
func pendingAnswered(ctx context.Context, inv *inventory.Inventory, p inventory.PendingAssignment) (string, bool, error) {
|
||||
if p.Acknowledged != nil {
|
||||
return "taken back with unassign", true, nil
|
||||
}
|
||||
assigned, err := inv.Assigned(ctx, p.Node)
|
||||
if err != nil {
|
||||
return "", false, err
|
||||
}
|
||||
for _, p := range waiting {
|
||||
if p.Node != node || p.Module != module {
|
||||
continue
|
||||
if containsString(assigned, p.Module) {
|
||||
return p.Module + " is assigned to " + p.Node + " since", true, nil
|
||||
}
|
||||
rows, err := inv.PendingFor(ctx, p.Node, p.Module)
|
||||
if err != nil {
|
||||
return "", false, err
|
||||
}
|
||||
for _, r := range rows {
|
||||
if r.ID != p.ID && r.Since.After(p.Since) {
|
||||
return "assigned again, pending on another build", true, nil
|
||||
}
|
||||
}
|
||||
return "", false, nil
|
||||
}
|
||||
|
||||
// withdrawPending is `unassign` of a module only pending on a machine, under the machine's hold: a waiting
|
||||
// pending assignment is withdrawn; one being made now is refused, never overridden; one that expired or was
|
||||
// refused is taken back, which answers its condition. False when there is none.
|
||||
func withdrawPending(ctx context.Context, inv *inventory.Inventory, node, module string) (string, bool, error) {
|
||||
rows, err := inv.PendingFor(ctx, node, module)
|
||||
if err != nil {
|
||||
return "", false, err
|
||||
}
|
||||
for _, p := range rows {
|
||||
switch p.State {
|
||||
case inventory.PendingApplying:
|
||||
return "", true, fmt.Errorf("%s is being assigned %s now, as build %s registered it: nothing was "+
|
||||
"withdrawn; `unassign %s %s` again once it is assigned takes it off", node, module, p.Build, node, module)
|
||||
case inventory.PendingWaiting:
|
||||
note := fmt.Sprintf("the pending assignment of %s to %s is withdrawn; build %s goes on, and registers "+
|
||||
"%s assigned nowhere", module, node, p.Build, module)
|
||||
settled, err := inv.SettlePending(ctx, p.ID, inventory.PendingWaiting, inventory.PendingWithdrawn, note)
|
||||
if err != nil {
|
||||
return "", false, err
|
||||
}
|
||||
if !settled {
|
||||
return "", true, fmt.Errorf("the pending assignment of %s to %s changed while it was withdrawn: "+
|
||||
"`status` says how; nothing was withdrawn", module, node)
|
||||
}
|
||||
return note, true, nil
|
||||
case inventory.PendingExpired, inventory.PendingRefused:
|
||||
if p.Cleared != nil || p.Acknowledged != nil {
|
||||
continue
|
||||
}
|
||||
if err := inv.MarkPending(ctx, p.ID, "acknowledged"); err != nil {
|
||||
return "", false, err
|
||||
}
|
||||
return fmt.Sprintf("the pending assignment of %s to %s, which %s, is taken back; its condition clears",
|
||||
module, node, p.State), true, nil
|
||||
}
|
||||
note := fmt.Sprintf("the pending assignment of %s to %s is withdrawn; build %s goes on, and registers %s "+
|
||||
"assigned nowhere", module, node, p.Build, module)
|
||||
if _, err := inv.SettlePending(ctx, p.ID, inventory.PendingWithdrawn, note); err != nil {
|
||||
return "", false, err
|
||||
}
|
||||
return note, true, nil
|
||||
}
|
||||
return "", false, nil
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user