[marcosqlbi]

When 1.7.6 was released, the redeploy job asked for a deployment from main and GitHub answered HTTP 502: Server Error. The job failed after one attempt, nothing deployed, and the site kept offering 1.7.5 until Publish site was run by hand (run 277).

The 502 had still created a run, 276, with no jobs. It stays queued, and GitHub refuses both cancel and force-cancel on it. It did not block run 277.

What changes

The step that runs gh workflow run publish-site.yml now tries up to five times when the error is an HTTP 5xx, waiting 15, 30, 45, and 60 seconds between attempts, with a warning on each retry. Any other error fails at once, because asking again would get the same answer. After five server errors the job fails with a message saying to run Publish site on main with the tag by hand.

A retry after a 5xx can start a second deployment if the first request did go through. That is harmless, because the pages concurrency group runs deployments one at a time, and both deploy main for the same tag.

Testing

The step's script was extracted from the YAML and run under bash -e -o pipefail, the shell GitHub uses, with a fake gh:

• Success on the first call: exits 0 after one call.
• 502, 502, then success: exits 0 after three calls.
• 422: exits 1 after one call, with no retry.
• 502 five times: exits 1 after five calls, with the error annotation.

The real path only runs on the next release.

🤖 Generated with Claude Code