novox/mesh-controller
13
0
Fork 0
Code Issues Pull Requests 4 Releases Wiki Activity

issue 974: the builders say the checks they run in a bucket that ages, and a builder drops a waiting request for a head and trunk commit another is running

Browse Source
This commit is contained in:
jschoubben
2026-10-17 14:10:26 +02:00
parent 357e08a4f6
commit bd257afbd5
10 changed files with 380 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
internal/link/builds_nats.go
+27 -6
View File
@@ -367,7 +367,11 @@ func (m *natsMachine) Take(ctx context.Context, do func(context.Context, Build))
// the ask to a second machine nor counts the wait against its deliveries.
working := make(chan struct{})
go stillWorking(msg, working, heartbeatWhileWorking)
// And a check says it is running, to the other builders (ADR 0406 §3): its key put now and at each heartbeat,
// deleted when it ends, whatever the outcome.
stopRunning := keepRunning(ctx, m.js.Conn(), request, m.on, heartbeatWhileWorking)
do(ctx, build)
stopRunning()
close(working)
}
}
@@ -431,17 +435,34 @@ func (m *natsMachine) next(ctx context.Context) (*nats.Msg, *takeQueue, error) {
// mesh-delivery's, and a check wrongly dropped is worse than one run twice).
func (m *natsMachine) judgedAlready(ctx context.Context, r BuildRequest) string {
c := r.Check
if c == nil || c.Group != "" || c.Try || c.Number <= 0 {
if c == nil || c.Try || (c.Group == "" && c.Number <= 0) {
return ""
}
key, found, err := ReadPullKey(ctx, m.js.Conn(), c.Owner, c.Repo, c.Number)
if c.Group == "" {
key, found, err := ReadPullKey(ctx, m.js.Conn(), c.Owner, c.Repo, c.Number)
if err != nil {
fmt.Fprintf(os.Stderr, "could not read the key of %s/%s#%d, so its check runs: %v\n", c.Owner, c.Repo, c.Number, err)
} else if found && key.Judged(r.Ref, c.Trunk) {
return fmt.Sprintf("%s/%s#%d is judged already at %.12s on %.12s (%s)", c.Owner, c.Repo, c.Number, r.Ref, c.Trunk,
key.Verdict)
}
}
return m.beingChecked(ctx, r)
}
// beingChecked says why a request need not run, or nothing: another builder's check of the same head and trunk commit is
// running, and says so under its pull request's key in the bucket of the checks running (ADR 0406 §3). A key a builder
// stopped renewing is gone within the bucket's age, and the request is taken then. Never the ask's own id: a redelivered ask
// is its own, and a key an older builder wrote that names no head or trunk is no match.
func (m *natsMachine) beingChecked(ctx context.Context, r BuildRequest) string {
running, ok, err := ReadRunning(ctx, m.js.Conn(), r)
if err != nil {
fmt.Fprintf(os.Stderr, "could not read the key of %s/%s#%d, so its check runs: %v\n", c.Owner, c.Repo, c.Number, err)
fmt.Fprintf(os.Stderr, "could not read whether %s is being checked, so its check runs: %v\n", r.ID, err)
return ""
}
if found && key.Judged(r.Ref, c.Trunk) {
return fmt.Sprintf("%s/%s#%d is judged already at %.12s on %.12s (%s)", c.Owner, c.Repo, c.Number, r.Ref, c.Trunk,
key.Verdict)
if ok && running.Build != r.ID && running.Same(r) {
return fmt.Sprintf("%s is being checked at %.12s on %.12s by %s as %s", r.ID, r.Ref, r.Check.Trunk, running.Builder,
running.Build)
}
return ""
}