novox/mesh-controller
8
0
Fork 0
Code Issues Pull Requests 5 Releases Wiki Activity

Keep a passed gate's verdict when the first machine's later reports go quiet (hq issue 335)

Browse Source 
A build that passed on its first machine was judged again from that
machine's next reports while its send to the rest waited; another walk's
unreported send there then failed the passed build at the wait's bound
and put it back.
This commit is contained in:
jochen
2026-10-08 21:08:12 +02:00
parent ca09a07fdf
commit fb74e24c9e
2 changed files with 92 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
cmd/mesh-controller/release_plan.go
+21 -1
View File
@@ -701,7 +701,6 @@ func advanceOnce(ctx context.Context, open *stores, p *inventory.Plan,
}
}
now := time.Now().UTC()
step := nextRollout(*state, running, policy.Together, reports, now, planWaitBound)
// **Sent with others, judged with them** (issue 281): the gate of the send that carried it is its
// verdict on its first machine. A failure there stopped the plan already.
if state.GatedBy != "" {
@@ -715,6 +714,9 @@ func advanceOnce(ctx context.Context, open *stores, p *inventory.Plan,
}
passedWith(m, state, state.GatedBy, lead.Gate)
}
// Read after its pass is taken over from the send that carried it, so a passed gate is never judged
// again from the first machine's later reports (novox/hq issue 335).
step := nextRollout(*state, running, policy.Together, reports, now, planWaitBound)
switch {
case step.failed != "":
// The first machine refused or failed what it was sent, or never said: the gate failed, and
@@ -974,6 +976,15 @@ func failFirstSend(ctx context.Context, open *stores, p *inventory.Plan, module
fmt.Printf("%s: %s\n", p.ID, p.Note)
}()
g := state.Gate
if g != nil && g.Verdict == inventory.GatePassed {
// **A build that passed its gate is never put back for what came after** (novox/hq issue 335): its
// verdict stands; the plan stops, and says why, but the build is not marked failed.
state.Why = "passed its gate; its walk then stopped: " + why
p.State = inventory.PlanFailed
p.Note = fmt.Sprintf("%s passed its gate on %s in tier %d (%s) and is kept; its walk stopped after: %s",
module, strings.Join(g.Machines, ", "), p.Tier, g.Why, why)
return
}
if g != nil && slices.Contains(g.Returned, module) {
// Put back at once when it broke: its rollback was made then, and is not made again.
state.Why = "put back when it broke; its send failed: " + g.Why
@@ -1061,6 +1072,15 @@ func nextRollout(s inventory.PlanModule, running []string, together bool, report
rest = append(rest, n)
}
}
// **A passed gate is the first machine's verdict, and its later reports are not** (novox/hq issue 335).
// Once the build passed there, what that machine reports next is about whatever it was sent after —
// another walk's send, a push — and says nothing of this build. On 2026-10-08 a build passed on the
// laptop, its send to the rest waited on another walk, that walk sent the laptop a new declaration it did
// not report for half an hour, and the plan read the silence as the first machine never applying the
// passed build: it marked it failed at its gate and put it back. The rest are sent, as the pass said.
if s.Gate != nil && s.Gate.Verdict == inventory.GatePassed {
return rolloutStep{send: rest}
}
var waiting, failed []string
for _, n := range s.First {
r, said := byNode[n]