aos publish-dev-base exceeds the shared runner 15m cap - release run 1102 cancelled by 'maximum run time' #551

Open
opened 2026-07-12 22:01:48 +00:00 by coilyco-ops · 0 comments
Member

Fresh evidence from tonight's two-stage rollout (aos run 1102, release branch, 2026-07-12 21:40Z): the run's test + plan jobs went green, then publish-dev-base died at the runner wall - the job log ends with 'The runner cancelled the job because it exceeds the maximum run time' and Forgejo records the run as cancelled. No runner pod restarts this time (all 2d4h stable, node at 47% memory after the comfyui scale-down per #547), so this is the OTHER phantom-cancellation class behind the aos 'publish-dev-base red on main' history (aos#452): the shared docker runner's runner.timeout: 15m (deploy/forgejo-runner.yml, docs/forgejo-runner-timeouts.md) vs a multi-arch dev-base family build that needs well beyond 15m when caches are cold.

docs/forgejo-runner-timeouts.md already names the remedy: a job family that regularly needs more than the runner cap gets a dedicated label/runner, not a fatter shared cap. tap-writer is host-executor and DinD-free, so it does not fit. Proposal: a 'heavy' docker runner class (capacity 1, runner.timeout ~90m) that aos release.yml's publish-dev-base targets via runs-on, leaving the shared 15m envelope intact.

Until then aos stage 2 only completes when the buildx/registry cache is warm. ward (v0.631.0) and cli-guard (v0.96.0) already released end to end tonight - their release jobs fit the cap.

Refs #547, aos#452, aos#477, ward#1117.

  • Claude (she/her), claude-windows-kai-desktop-tower-bc51
Fresh evidence from tonight's two-stage rollout (aos run 1102, release branch, 2026-07-12 21:40Z): the run's test + plan jobs went green, then publish-dev-base died at the runner wall - the job log ends with 'The runner cancelled the job because it exceeds the maximum run time' and Forgejo records the run as cancelled. No runner pod restarts this time (all 2d4h stable, node at 47% memory after the comfyui scale-down per #547), so this is the OTHER phantom-cancellation class behind the aos 'publish-dev-base red on main' history (aos#452): the shared docker runner's runner.timeout: 15m (deploy/forgejo-runner.yml, docs/forgejo-runner-timeouts.md) vs a multi-arch dev-base family build that needs well beyond 15m when caches are cold. docs/forgejo-runner-timeouts.md already names the remedy: a job family that regularly needs more than the runner cap gets a dedicated label/runner, not a fatter shared cap. tap-writer is host-executor and DinD-free, so it does not fit. Proposal: a 'heavy' docker runner class (capacity 1, runner.timeout ~90m) that aos release.yml's publish-dev-base targets via runs-on, leaving the shared 15m envelope intact. Until then aos stage 2 only completes when the buildx/registry cache is warm. ward (v0.631.0) and cli-guard (v0.96.0) already released end to end tonight - their release jobs fit the cap. Refs #547, aos#452, aos#477, ward#1117. - Claude (she/her), claude-windows-kai-desktop-tower-bc51
Sign in to join this conversation.
No description provided.