The smoke check answered from the old task right after the stack PUT, so the deploy went green before anything rolled out:
birdnet-go run 58329 (2026-11-03): smoke 200 with uptime 1h38m, then ~1 min of 503 while the new task started;
revaj-crm-stage (2026-11-02): new image exited 1 twice, Swarm paused the update, CI green - still paused now.
Now: before the PUT remember each stack service's UpdateStatus.StartedAt; afterwards wait until every service whose value changed is completed, fail on paused / rollback_*, time out after rollout-timeout (600 s, image pulls included). Comparing with the previous value avoids runner/Swarm clock skew; a 20 s grace covers the PUT returning before Swarm registers the update (seen live: 25.908 vs 25.929). Services the deploy did not touch are ignored (a stale paused one does not fail other deploys). wait-rollout: "false" = old behaviour.
Tested against a fake Portainer API running the real step script: normal rollout (smoke only after completion), paused -> rc 1, no change -> passes after grace, stale paused untouched service -> rc 0, stuck -> timeout rc 1, wait-rollout=false. Response shape and label filter checked read-only against the live Portainer.
Release: hand-made change, tag v12 by hand after the merge (README note added).
The smoke check answered from the **old** task right after the stack PUT, so the deploy went green before anything rolled out:
- birdnet-go run 58329 (2026-11-03): smoke 200 with `uptime 1h38m`, then ~1 min of 503 while the new task started;
- revaj-crm-stage (2026-11-02): new image exited 1 twice, Swarm paused the update, CI green - still `paused` now.
Now: before the PUT remember each stack service's `UpdateStatus.StartedAt`; afterwards wait until every service whose value changed is `completed`, fail on `paused` / `rollback_*`, time out after `rollout-timeout` (600 s, image pulls included). Comparing with the previous value avoids runner/Swarm clock skew; a 20 s grace covers the PUT returning before Swarm registers the update (seen live: 25.908 vs 25.929). Services the deploy did not touch are ignored (a stale paused one does not fail other deploys). `wait-rollout: "false"` = old behaviour.
Tested against a fake Portainer API running the real step script: normal rollout (smoke only after completion), paused -> rc 1, no change -> passes after grace, stale paused untouched service -> rc 0, stuck -> timeout rc 1, wait-rollout=false. Response shape and label filter checked read-only against the live Portainer.
Release: hand-made change, tag `v12` by hand after the merge (README note added).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
The smoke URL answered from the old task right after the stack PUT, so CI
went green before anything was deployed (birdnet-go 2026-11-03: green, then
1 min of 503) and a paused update stayed unnoticed (revaj-crm-stage
2026-11-02, new image exited 1 twice, CI green). Now the step waits until
every service whose UpdateStatus.StartedAt changed is completed and fails
on paused / rollback_*; untouched services are ignored.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tth
merged commit 7c8359099c into main2026-11-03 01:55:02 +01:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
The smoke check answered from the old task right after the stack PUT, so the deploy went green before anything rolled out:
uptime 1h38m, then ~1 min of 503 while the new task started;pausednow.Now: before the PUT remember each stack service's
UpdateStatus.StartedAt; afterwards wait until every service whose value changed iscompleted, fail onpaused/rollback_*, time out afterrollout-timeout(600 s, image pulls included). Comparing with the previous value avoids runner/Swarm clock skew; a 20 s grace covers the PUT returning before Swarm registers the update (seen live: 25.908 vs 25.929). Services the deploy did not touch are ignored (a stale paused one does not fail other deploys).wait-rollout: "false"= old behaviour.Tested against a fake Portainer API running the real step script: normal rollout (smoke only after completion), paused -> rc 1, no change -> passes after grace, stale paused untouched service -> rc 0, stuck -> timeout rc 1, wait-rollout=false. Response shape and label filter checked read-only against the live Portainer.
Release: hand-made change, tag
v12by hand after the merge (README note added).🤖 Generated with Claude Code