HDDS-4017. Acceptance check may run against wrong commit

[adoroszlai]

What changes were proposed in this pull request?

Use commit SHA to checkout a specific commit on push and schedule events. Keep using name for pull_request events (eg. refs/pull/1234/merge), because merge commit may not be available by SHA.

Use github context instead of environment variables. Github substitutes these values before running the command, so we can see them in the log. Comparison:

git clone https://github.com/${GITHUB_REPOSITORY}.git /mnt/ozone
cd /mnt/ozone
git fetch origin "${GITHUB_REF}"
git checkout FETCH_HEAD

vs.

git clone https://github.com/adoroszlai/hadoop-ozone.git /mnt/ozone
cd /mnt/ozone
git fetch origin "refs/heads/HDDS-4017"
git checkout FETCH_HEAD

https://issues.apache.org/jira/browse/HDDS-4017

How was this patch tested?

push event:
https://github.com/adoroszlai/hadoop-ozone/runs/903007668#step:4:8

pull_request event:
https://github.com/apache/hadoop-ozone/pull/1249/checks?check_run_id=903076398#step:4:6

[elek]

Thanks the patch @adoroszlai

I am fine with it, but have a question (to improve my understanding):

Based on my understanding for pull requests the merge commit points to a potential merge. In case of using github checkout action, not the github.ref but the merge commit is used.

Do you have any example when it was wrong?

PS:

FYI: I started to experiment with a custom checkout script:

elek#13

My main motivation was to always merge to the latest master with coverage data to have better codecov data.

[adoroszlai]

Based on my understanding for pull requests the merge commit points to a potential merge. In case of using github checkout action, not the github.ref but the merge commit is used.

Both ${GITHUB_REF} and ${{ github.ref }} point to the merge commit for PRs, and that's fine.

The problem is with push builds, because in that case both refs are something like refs/heads/master. The commit this points to can change in the meantime, so we need to use the SHA. (Your original solution was fine for push commits, my fix for PRs introduced this problem.)

Do you have any example when it was wrong?

The Jira issue HDDS-4017 has an example.

[elek]

The commit this points to can change in the meantime, so we need to use the SHA.

Thanks to explain it. Now I got it. 💡 💡 💡

Let me merge it.

[adoroszlai]

Thanks @elek for reviewing and committing it.