NoahKusaba opened a new pull request, #2439:
URL: https://github.com/apache/datafusion-ballista/pull/2439

   # Which issue does this PR close?
   
   No issue filed. `main` is currently red and this unblocks it.
   
   # Rationale for this change
   
   `cargo test -p ballista-scheduler` has not compiled since #2016 merged.
   `deterministic_launch_rejection_fails_job` calls 
`SchedulerTest::await_completion`,
   which #2354 removed.
   
   Neither PR was wrong on its own:
   
   | Date | Event |
   | --- | --- |
   | 2026-08-18 | #2016's branch last merges `main` |
   | 2026-08-30 | #2354 removes `await_completion`, which had no callers |
   | 2026-09-09 | #2016 merges, 10 days stale, adding a call to it |
   
   The two edits touch different files, so git merged them cleanly. Nothing ever
   compiled the combination until the post-merge run on `main`.
   
   # What changes are included in this PR?
   
   Use the existing `SchedulerTest::run` instead of restoring the deleted 
helper.
   `run` is `submit` plus a poll to a terminal state, which is what the test was
   doing by hand, so this adds no test surface. It also drains the task-status
   channel, which `submit` alone does not.
   
   Verified: the test passes, inverting its assertion to expect a successful job
   makes it fail, and the full `ballista-scheduler` suite is green at 402 
passed.
   
   # Are there any user-facing changes?
   
   No. Test-only, three lines.
   
   # Preventing this class of failure
   
   `.asf.yaml` protects `main` with `required_approving_review_count: 1` and
   nothing else, so a two-month-old branch can merge without re-testing against
   current `main`. Review cannot catch this: there is no conflict to see, and 
both
   sides are individually correct.
   
   The fix is to require branches to be up to date before merging, either via
   strict status checks or a merge queue. A merge queue is the better fit here:
   strict checks force every contributor to rebase whenever `main` moves, which 
on
   a repo with this much CI is a real tax, whereas a queue tests the merge 
result
   once, at merge time, without pushing work onto contributors.
   
   Worth noting detection did work, just late. The post-merge run went red
   immediately. The gap is that it went red on `main` rather than on the PR.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to