Woodpecker: a manual-trigger pipeline errors with zero steps while the identical config loads on push #1468

Closed
opened 2026-10-27 13:40:31 +00:00 by sam · 0 comments
Owner

Both occurrences are in the sam/catlow-tech repo on the Woodpecker server at ci.catlow.cloud (repo id 8).

EVIDENCE

  • Pipelines #25 and #26: created with woodpecker-cli pipeline create --branch main sam/catlow-tech (the manual trigger). Both came back status=error with ZERO steps: woodpecker-cli pipeline ps sam/catlow-tech 25 prints nothing. #26 was a retry after #24 had completed successfully, so it is not a race with a concurrent run.
  • The SAME .woodpecker.yml loads fine on the push trigger in the same window: #24 and #27 through #29 all cloned and ran their steps (clone/check-secrets/publish/deploy-up). So this is not a config error that lint or the push path can see: woodpecker-cli lint .woodpecker.yml reports '✅ Config is valid'.
  • The pipeline declares manual in its top-level when (event: [push, pull_request, tag, manual, cron]), so the event should be eligible.
  • An earlier attempt to restart a pipeline in place (woodpecker-cli pipeline start sam/catlow-tech) failed differently: 'missing step number'.

WHY IT MATTERS / WHAT IS BLOCKED BY IT
The on-demand path is the only way for a lane to RUN a check and watch it work without waiting for the schedule or pushing a commit. Without it, proving a cron-only step fires requires temporarily changing its when to include push and pushing (which is what we had to do for the artifact-drift step, runs #27/#28). That is a workaround with a real cost: it puts a failing step on the push path of main for the duration.

THE DIAGNOSTIC IS STRANDED
The reason is only in the database (pipelines.errors for repo 8), and the app lane cannot read it: woodpecker-cli authenticates as sam but stores no token that can be extracted (contexts.json holds only server_url/log_level, and no Woodpecker env vars are set), so pipeline log show is empty for a zero-step run and there is no API route available. Whoever has DB access can settle this in one row.

POSSIBLE CONNECTION TO THE EARLIER ZERO-STEP INCIDENTS — worth checking whether one fix closes several
This session saw zero-step errors three times before this one, and each had a KNOWN cause found later:

  1. #1 (manual): the config had both environment and settings on one step — a schema error the server rejected at load.
  2. #2/#3 (push): unable to parse variable name, from a bash substring form in the config.
  3. The wedge where a force-recreated Woodpecker server left an agent unable to reconnect; pipelines were numbered with correct metadata and zero steps because no agent took the work.
    So zero steps has at least three distinct causes, only one of which is 'config rejected'. #25/#26 do not match any of them: the config is valid and passes on push. If #25/#26 turn out to share a cause with #1 (both manual) then that earlier diagnosis was incomplete, and if they do not, the useful output is a distinguishing rule for the next person: a zero-step error is NOT automatically a config-load failure, and the message lives in pipelines.errors.

Close reason

Duplicate of archie-core-hn7u: same repository, manual runs #25/#26 and missing pre-step diagnostic. Canonical issue updated with current live checks and acceptance criteria.


Imported from beads archie-core-nam3 · GitHub #1097 · type bug · priority P2 · created 2026-09-22 · closed 2026-10-04

Both occurrences are in the sam/catlow-tech repo on the Woodpecker server at ci.catlow.cloud (repo id 8). EVIDENCE - Pipelines #25 and #26: created with `woodpecker-cli pipeline create --branch main sam/catlow-tech` (the manual trigger). Both came back status=error with ZERO steps: `woodpecker-cli pipeline ps sam/catlow-tech 25` prints nothing. #26 was a retry after #24 had completed successfully, so it is not a race with a concurrent run. - The SAME .woodpecker.yml loads fine on the push trigger in the same window: #24 and #27 through #29 all cloned and ran their steps (clone/check-secrets/publish/deploy-up). So this is not a config error that lint or the push path can see: `woodpecker-cli lint .woodpecker.yml` reports '✅ Config is valid'. - The pipeline declares `manual` in its top-level `when` (`event: [push, pull_request, tag, manual, cron]`), so the event should be eligible. - An earlier attempt to restart a pipeline in place (`woodpecker-cli pipeline start sam/catlow-tech`) failed differently: 'missing step number'. WHY IT MATTERS / WHAT IS BLOCKED BY IT The on-demand path is the only way for a lane to RUN a check and watch it work without waiting for the schedule or pushing a commit. Without it, proving a cron-only step fires requires temporarily changing its `when` to include push and pushing (which is what we had to do for the artifact-drift step, runs #27/#28). That is a workaround with a real cost: it puts a failing step on the push path of main for the duration. THE DIAGNOSTIC IS STRANDED The reason is only in the database (`pipelines.errors` for repo 8), and the app lane cannot read it: `woodpecker-cli` authenticates as sam but stores no token that can be extracted (contexts.json holds only server_url/log_level, and no Woodpecker env vars are set), so `pipeline log show` is empty for a zero-step run and there is no API route available. Whoever has DB access can settle this in one row. POSSIBLE CONNECTION TO THE EARLIER ZERO-STEP INCIDENTS — worth checking whether one fix closes several This session saw zero-step errors three times before this one, and each had a KNOWN cause found later: 1. #1 (manual): the config had both `environment` and `settings` on one step — a schema error the server rejected at load. 2. #2/#3 (push): `unable to parse variable name`, from a bash substring form in the config. 3. The wedge where a force-recreated Woodpecker server left an agent unable to reconnect; pipelines were numbered with correct metadata and zero steps because no agent took the work. So zero steps has at least three distinct causes, only one of which is 'config rejected'. #25/#26 do not match any of them: the config is valid and passes on push. If #25/#26 turn out to share a cause with #1 (both manual) then that earlier diagnosis was incomplete, and if they do not, the useful output is a distinguishing rule for the next person: a zero-step error is NOT automatically a config-load failure, and the message lives in `pipelines.errors`. ### Close reason Duplicate of archie-core-hn7u: same repository, manual runs #25/#26 and missing pre-step diagnostic. Canonical issue updated with current live checks and acceptance criteria. --- _Imported from beads `archie-core-nam3` · GitHub #1097 · type bug · priority P2 · created 2026-09-22 · closed 2026-10-04_
sam added the mediumbug labels 2026-10-27 13:40:31 +00:00
sam self-assigned this 2026-10-27 13:40:31 +00:00
sam closed this issue 2026-10-27 13:40:31 +00:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: sam/archie#1468