The `get-merged-pr` workflow sometimes gets empty output

[qinsoon]

get-merged-pr is triggered when a PR is merged to get its pull request number, and set the pull request number as the output for binding-refs to read and check what binding PRs are associated with it.

mmtk-core/.github/workflows/auto-merge.yml

Lines 12 to 25 in 76ef95a

	get-merged-pr:
	runs-on: ubuntu-latest
	outputs:
	commit: ${{ steps.get-commit-hash.outputs.commit }}
	pr: ${{ steps.get-pr.outputs.pr }}
	steps:
	- id: get-commit-hash
	run: echo "commit=${GITHUB_SHA}" >> "$GITHUB_OUTPUT"
	- id: get-pr
	run: |
	PR_NUMBER=$(gh pr list --search "${GITHUB_SHA}" --state merged --repo mmtk/mmtk-core --json number --jq '.[0].number')
	echo "pr=$PR_NUMBER" >> "$GITHUB_OUTPUT"
	env:
	GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

However, we saw this workflow failed to get the pull request number in https://github.com/mmtk/mmtk-core/actions/runs/19812966874/job/56860107289. I attempted a few reruns and it still failed.

I also saw this happened before once, but it worked after a rerun.

Will need to investigate the issue.

[qinsoon]

The workflow also worked after another rerun https://github.com/mmtk/mmtk-core/actions/runs/19812966874/job/56860107289 (need to rerun all the jobs, rather than failed jobs). I can't reproduce the issue locally. The gh pr list worked fine locally.

Possibly there was a delay for the gh pr list query. We could add retry for this workflow if the PR number is empty.

[wks]

Is the commit hash stable w.r.t. squash merging and multiple jobs in the merge queue?

[qinsoon]

Is the commit hash stable w.r.t. squash merging and multiple jobs in the merge queue?

I think so.

1. We only use squash merging in mmtk-core so it works fine for squash merging.
2. The workflow is triggered when a commit is pushed to master. So it doesn't matter how many jobs there are in the merge queue, this workflow is triggered for each commit in master.

[wks]

I think we can do the following things to make it easier to debug.

1. Echo ${GITHUB_SHA} to stdout so that we see the value in the CI log.
2. Execute gh pr list --search "${GITHUB_SHA}" --state merged --repo mmtk/mmtk-core, without using JSON, and without assigning the output to any variable, so that we can see human-readable output from the CI log.
3. Also execute gh pr list --search "${GITHUB_SHA}" --state mmtk/mmtk-core so we know whether the error was because the PR was not merged.
4. Echo the value of ${PR_NUMBER} to stdout

[qinsoon]

The issue showed up again in https://github.com/mmtk/mmtk-core/actions/runs/19851343850/job/56878651255. With the extra output we added in #1428, we got:

Commit hash: 3b3a2ed3ab4fb3e20735262bdb91384c838614ab
Pull request for the commit is below:
(empty)
Output pr=

The commit hash was correct, but the query with gh pr list returned empty for the hash. I queried it again locally, and got the right pull request:

$ gh pr list --search 3b3a2ed3ab4fb3e20735262bdb91384c838614ab --state merged --repo mmtk/mmtk-core

Showing 1 of 1 pull request in mmtk/mmtk-core that matches your search

ID     TITLE                                           BRANCH                       CREATED AT        
#1428  Print more info for debugging in get-merged-pr  qinsoon:debug-get-merged-pr  about 17 hours ago

It should be what I guessed: there was a delay between a hash was pushed and a hash could be queried with gh. If the workflow queries the hash immediately after it is pushed, the query may not find the hash. It worked fine before. Possibly Github made some changes that made that delay much longer than before. Adding a retry and wait like #1427 should fix the problem.