Skip to content

ci(savings): a failed nightly run opens an issue, or comments on the open one - #293

Merged
angeloINTJ merged 1 commit into
mainfrom
ci/savings-failure-issue
Oct 25, 2026
Merged

angeloINTJ merged 1 commit into
mainfrom
ci/savings-failure-issue

Conversation

@angeloINTJ

Copy link
Copy Markdown
Owner

What

The savings watermark job runs only at night and on demand, never on a pull request. So its failures reach nobody: from 2026-10-02 to 2026-10-22 it failed every night without anyone seeing it. The maintainer chose an issue as the alarm (2026-10-24), over running the job on every pull request that touches src/ (+16 to 21 minutes of CI each).

  • A new last step runs when a scheduled run fails. It comments on the open issue titled Savings watermark failed on main, or opens one, with the run's link. GitHub mails the maintainer. One issue covers a streak of red nights, and closing it after the fix re-arms the alarm.
  • permissions is now contents: read and issues: write.
  • The manual dispatch gains simular_falha. It fails the first step before any build, so the expensive steps are skipped and the alarm runs as on a red night. This is how the alarm was proved.

Proof

Two dispatches of this branch's workflow, with simular_falha=true:

Run The first step The seven build steps The alarm
40148177331 failed (simulated) skipped opened #292
40148205543 failed (simulated) skipped commented on #292, no second issue

#292 was closed afterwards, with a comment saying both failures were simulated. The YAML parses, with the steps in that order. actionlint is not installed here, so the workflow was proved by those two runs.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LyLm7Q1o7kCHngE63BnoTx

…open one

The savings watermark job does not run on pull requests, so its failures
reach nobody: from 2026-10-02 to 2026-10-22 it failed every night without
anyone seeing it. The maintainer chose an issue as the alarm (2026-10-24),
over running the job on every pull request that touches src/, which would
add 16 to 21 minutes of CI to each.

- A new last step runs when a scheduled run fails. It comments on the open
  issue titled "Savings watermark failed on main", or opens one, with the
  run's link. GitHub then mails the maintainer. One issue covers a streak of
  red nights; closing it after the fix re-arms the alarm.
- `permissions` is now `contents: read` and `issues: write`. The token
  needs nothing else: the checkout reads, and the cache does not use it.
- The manual dispatch gains `simular_falha`, which fails the first step
  before any build. The expensive steps are then skipped, and the alarm runs
  as on a red night. This is how the alarm was proved.

Proved on this branch, with two dispatches at simular_falha=true:
- run 40148177331: the first step failed, the seven build steps were
  skipped, and the alarm opened issue #292;
- run 40148205543: the alarm commented on #292 and opened no second issue
  (one open issue with that title, one comment).
#292 was closed afterwards, with a comment saying both failures were
simulated. The YAML parses, with the steps in this order. actionlint is not
installed here, so the workflow was proved by those two runs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LyLm7Q1o7kCHngE63BnoTx
@github-actions github-actions Bot added the ci Continuous integration and automation label Oct 25, 2026
@angeloINTJ
angeloINTJ merged commit 6adf102 into main Oct 25, 2026
13 checks passed
@angeloINTJ angeloINTJ mentioned this pull request Oct 29, 2026
@angeloINTJ
angeloINTJ deleted the ci/savings-failure-issue branch October 31, 2026 13:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ci Continuous integration and automation

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant