SEZ9 commented on PR #12156:
URL: https://github.com/apache/seatunnel/pull/12156#issuecomment-6889166966

   Thanks @CryoThrust for running the Build on the unchanged head 
`7938f302fd1f3ba212993f258bb65d4cc7f58f3b` 
(https://github.com/CryoThrust/seatunnel/actions/runs/43150977139) and for the 
diagnosis. I agree with the read of the two red jobs ([JDK 
11](https://github.com/CryoThrust/seatunnel/actions/runs/43150977139/job/130960712990),
 [JDK 
17](https://github.com/CryoThrust/seatunnel/actions/runs/43150977139/job/130960712862)):
 both die on the `404` while pulling `docker.io/atmoz/sftp:alpine-3.7`, so 
`SftpFileIT` never reaches its assertions. That is a fixture problem, not 
something this engine change introduced, and the replacement tag is already on 
`dev` via #13203 (`379292194cc9ee9ecba2f6adf07cefcc94a7db21`). Please don't add 
any SFTP or dependency patch here — just rebase/merge `dev` so the next run 
picks it up.
   
   On the actual review: the head hasn't moved, so my four earlier points are 
still open. To keep this concrete, here is what I'm looking for in the next 
push:
   
   1. **cancelJob() not-started check** — drop the second, unsynchronized IMap 
read and reuse the already-captured `jobStatus` local, mirroring what 
`stopJob()` does, so the null-state path isn't silently changed.
   2. **Check-then-act race in cancelJob()/stopJob()** — the not-started test 
and the `updateJobState(CANCELED)` currently run outside the `PhysicalPlan` 
monitor, so a concurrent `startJob()`/`stateProcess()` can move the job to 
SCHEDULED/RUNNING in between and a job with deploying pipelines ends up 
directly in CANCELED. Please move the check and the transition under the same 
lock (or otherwise make them atomic) and briefly explain the chosen approach in 
the PR description.
   3. **Tests** — `JobStatusTest` pins the enum table, but nothing exercises 
the new `NOT_STARTED_STATUSES` decision in `PhysicalPlan`. Please add a test 
that drives `cancelJob()`/`stopJob()` through a not-started state and a started 
state, and add a short comment on the guard test telling a future contributor 
what to do when it fails (i.e., whether to update the set or the test).
   4. **Javadoc / comments** — give `NOT_STARTED_STATUSES` a Javadoc that ties 
it to the job state machine, and trim the inline comments at the two call sites 
so they don't duplicate the set contents.
   
   Once `dev` is synced and the Build is green on the new head, I'll go through 
items 1–4 again. If you disagree with any of them, a reply here is fine and we 
can sort it out before you push.
   
   <!-- streview-comment:2368 -->


-- 
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]

Reply via email to