### What changes were proposed in this pull request?
This patch uses the "concurrency" syntax to replace the "cancel job" workflow:
- .github/workflows/benchmark.yml
- .github/workflows/labeler.yml
- .github/workflows/notify_test_workflow.yml
- .github/workflows/test_report.yml
Remove the .github/workflows/cancel_duplicate_workflow_runs.yml
Note that the push/schedule based job are not changed to keep the same config in a4b70758d3:
- .github/workflows/build_and_test.yml
- .github/workflows/publish_snapshot.yml
- .github/workflows/stale.yml
- .github/workflows/update_build_status.yml
### Why are the changes needed?
We are using [cancel_duplicate_workflow_runs](a70e66ecfa/.github/workflows/cancel_duplicate_workflow_runs.yml (L1)) job to cancel previous jobs when a new job is queued. Now, it has been supported by the github action by using ["concurrency"](https://docs.github.com/en/actions/reference/workflow-syntax-for-github-actions#concurrency) syntax to make sure only a single job or workflow using the same concurrency group.
Related: https://github.com/apache/arrow/pull/10416 and https://github.com/potiuk/cancel-workflow-runs
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
triger the PR manaully
Closes#32806 from Yikun/SPARK-X.
Authored-by: Yikun Jiang <yikunkero@gmail.com>
Signed-off-by: Hyukjin Kwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
TL;DR: now it shows green yellow read status of tests instead of relying on a comment in a PR, **see https://github.com/HyukjinKwon/spark/pull/41 for an example**.
This PR proposes the GitHub status checks instead of a comment that link to the build (from forked repository) in PRs.
This is how it works:
1. **forked repo**: "Build and test" workflow is triggered when you create a branch to create a PR which uses your resources in GitHub Actions.
1. **main repo**: "Notify test workflow" (previously created a comment) now creates a in-progress status (yellow status) as a GitHub Actions check to your current PR.
1. **main repo**: "Update build status workflow" regularly (every 15 mins) checks open PRs, and updates the status of GitHub Actions checks at PRs according to the status of workflows in the forked repositories (status sync).
**NOTE** that creating/updating statuses in the PRs is only allowed from the main repo. That's why the flow is as above.
### Why are the changes needed?
The GitHub status shows a green although the tests are running, which is confusing.
### Does this PR introduce _any_ user-facing change?
No, dev-only.
### How was this patch tested?
Manually tested at:
- https://github.com/HyukjinKwon/spark/pull/41
- HyukjinKwon#42
- HyukjinKwon#43
- https://github.com/HyukjinKwon/spark/pull/37
**queued**:
<img width="861" alt="Screen Shot 2021-04-16 at 10 56 03 AM" src="https://user-images.githubusercontent.com/6477701/114960831-c9a73080-9ea2-11eb-8442-ddf3f6008a45.png">
**in progress**:
<img width="871" alt="Screen Shot 2021-04-16 at 12 14 39 PM" src="https://user-images.githubusercontent.com/6477701/114966359-59ea7300-9ead-11eb-98cb-1e63323980ad.png">
**passed**:
**failure**:
Closes#32193 from HyukjinKwon/update-checks-pr-poc.
Lead-authored-by: HyukjinKwon <gurwls223@apache.org>
Co-authored-by: Hyukjin Kwon <gurwls223@apache.org>
Co-authored-by: Yikun Jiang <yikunkero@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR removes the old Probot Autolabeler labeling configuration, as the probot autolabeler has been deprecated. I've updated the configs in Iceberg and in Avro, and we also need to update here. This PR adds in an additional workflow for labeling PRs and migrates the old probot config to the new format. Unfortunately, because certain features have not been released upstream, we will not get the _exact_ behavior as before. I have documented where that is and what changes are neeeded, and in the associated ticket I've also discussed other options and why I think this is the best way to go. Definitely a follow up ticket is needed to get the original behavior back in these few cases, but PRs have not been labeled for almost a month and so it's probably best to get it right 95% of the time and occasionally have some UI related PRs labeled as `CORE` while the issue is resolved upstream and/or further investigated.
### Why are the changes needed?
The probot autolabeler is dead and will not be maintained going forward. This has been confirmed with github user [at]mithro in an issue in their repository.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
To test this PR, I first merged the config into my local fork. I then edited it several times and ran tests on that.
Unfortunately, I've overwritten my fork with the apache repo in order to create a proper PR. However, I've also added the config for the same thing in the Iceberg repo as well as the Avro repo.
I have now merged this PR into my local repo and will be running some tests on edge cases there and for validating in general:
- [Check that the SQL label is applied for changes directly below repo root's sql directory](https://github.com/kbendick/spark/pull/16) ✅
- [Check that the structured streaming label is applied](https://github.com/kbendick/spark/pull/20) ✅
- [Check that a wildcard at the end of a pattern will match nested files](https://github.com/kbendick/spark/pull/19) ✅
- [Check that the rule **/*pom.xml will match the root pom.xml file](https://github.com/kbendick/spark/pull/25) ✅
I've also discovered that we're likely not killing github actions that run (like large tests etc) when users push to their PR. In most cases, I see that a user has to mark something as "OK to test", but it still seems like we might want to discuss whether or not we should add a cancellation step In order to save time / capacity on the runners. If so desired, we would add an action in each workflow that cancels old runs when a `push` action occurs on a PR. This will likely make waiting for test runners much faster iff tests are automatically rerun on push by anybody (such as PMCs, PRs that have been marked OK to test, etc). We could free a large number of resources potentially if a cancellation step was added to all of the workflows in the Apache account (as github action API limits are set at the account level).
Admittedly, the fact that the "old" workflow runs weren't cancelled could admittedly be because of the fact that I was working in a fork, but given that there are explicit actions to be added to the start of workflows to cancel old PR workflows and given that we don't have them configured indicates to me that likely this is the case in this repo (and in most `apache` repos as well), at least under certain circumstances (e.g. repos that don't have "Ok to test"-like webhooks as one example).
This is a separate issue though, which I can bring up on the mailing list once I'm done with this PR. Unfortunately I've been very busy the past two weeks, but if somebody else wanted to work on that I would be happy to support with any knowledge I have.
The last Apache repo to still have the probot autolabeler in it is Beam, at which point we can have Gavin from ASF Infra remove the permissions for the probot autolabeler entirely. See the associated JIRA ticket for the links to other tickets, like the one for ASF Infra to remove the dead probot autolabeler's read and write permissions to our PRs in the Apache organization.
Closes#30244 from kbendick/begin-migration-to-github-labeler-action.
Authored-by: Kyle Bendickson <kjbendickson@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
Hyukjin Kwon
Yikun Jiang
Kyle Bendickson
