fix(ci): count the acknowledgement timeout from the start of the server when it is later

A runner that could not reach a server that was down is not punished for it,
as for the heartbeat deadline.

commit: 5df470be37beaf70d3561ccd6bf0a1b5b5f2f5a8
author: brice · 9d ago · 2026-10-12 18:40 UTC
parent: 1769e4f
signature: GPG VerifiedCD88F9E564E0102646FB6077A1A07C7C7C3E349D · arch-gpg Verified for the committer email

1 file · +13 -2

M internal/usecase/ci_watch.go +13 -2


@@ -130,13 +130,24 @@ func (c *CI) lastBeat(j domain.CIJob) time.Time {
130	130		func (c *CI) due(j domain.CIJob, now time.Time) bool {
131	131		switch {
132	132		case j.Phase == domain.CIPhaseAssigned:
133		-	return now.Sub(j.AssignedAt) > c.cfg.AckTimeout
	133	+	return now.Sub(c.assignedSince(j)) > c.cfg.AckTimeout
134	134		case j.CancelReason != domain.CICancelNone:
135	135		return now.Sub(j.CancelRequestedAt) > c.cfg.CancelGrace
136	136		}
137	137		return now.Sub(c.lastBeat(j)) > c.cfg.HeartbeatDeadline \|\| now.After(c.deadline(j))
138	138		}
139	139
	140	+	// assignedSince is when the acknowledgement of an assignment starts to count:
	141	+	// its assignment, or the start of this process when that is later, so that a
	142	+	// runner that could not reach a server that was down is not punished for it
	143	+	// (the same grace as the heartbeat deadline).
	144	+	func (c *CI) assignedSince(j domain.CIJob) time.Time {
	145	+	if c.boot.After(j.AssignedAt) {
	146	+	return c.boot
	147	+	}
	148	+	return j.AssignedAt
	149	+	}
	150	+
140	151		// deadline is when the job's time budget ends for the server: timeout-minutes
141	152		// from the assignment, plus the slack that lets the runner's own timeout win.
142	153		func (c *CI) deadline(j domain.CIJob) time.Time {

@@ -147,7 +158,7 @@ func (c *CI) deadline(j domain.CIJob) time.Time {
147	158		// reports whether it changed.
148	159		func (c CI) enforce(ctx context.Context, rs runState, j *domain.CIJob, now time.Time) bool {
149	160		switch {
150		-	case j.Phase == domain.CIPhaseAssigned && now.Sub(j.AssignedAt) > c.cfg.AckTimeout:
	161	+	case j.Phase == domain.CIPhaseAssigned && now.Sub(c.assignedSince(*j)) > c.cfg.AckTimeout:
151	162		c.cancelAssignment(ctx, rs, j)
152	163		return true
153	164		case j.CancelReason != domain.CICancelNone && now.Sub(j.CancelRequestedAt) > c.cfg.CancelGrace:

 func (c *CI) due(j domain.CIJob, now time.Time) bool {
 	switch {
 	case j.Phase == domain.CIPhaseAssigned:
-		return now.Sub(j.AssignedAt) > c.cfg.AckTimeout
+		return now.Sub(c.assignedSince(j)) > c.cfg.AckTimeout
 	case j.CancelReason != domain.CICancelNone:
 		return now.Sub(j.CancelRequestedAt) > c.cfg.CancelGrace
 	}
 	return now.Sub(c.lastBeat(j)) > c.cfg.HeartbeatDeadline || now.After(c.deadline(j))
 }
+// assignedSince is when the acknowledgement of an assignment starts to count:
+// its assignment, or the start of this process when that is later, so that a
+// runner that could not reach a server that was down is not punished for it
+// (the same grace as the heartbeat deadline).
+func (c *CI) assignedSince(j domain.CIJob) time.Time {
+	if c.boot.After(j.AssignedAt) {
+		return c.boot
+	}
+	return j.AssignedAt
+}
 // deadline is when the job's time budget ends for the server: timeout-minutes
 // from the assignment, plus the slack that lets the runner's own timeout win.
 func (c *CI) deadline(j domain.CIJob) time.Time {
 // reports whether it changed.
 func (c *CI) enforce(ctx context.Context, rs *runState, j *domain.CIJob, now time.Time) bool {
 	switch {
-	case j.Phase == domain.CIPhaseAssigned && now.Sub(j.AssignedAt) > c.cfg.AckTimeout:
+	case j.Phase == domain.CIPhaseAssigned && now.Sub(c.assignedSince(*j)) > c.cfg.AckTimeout:
 		c.cancelAssignment(ctx, rs, j)
 		return true
 	case j.CancelReason != domain.CICancelNone && now.Sub(j.CancelRequestedAt) > c.cfg.CancelGrace: