SEZ9 commented on issue #13320:
URL: https://github.com/apache/seatunnel/issues/13320#issuecomment-6992096063

   Thanks for the detailed report, and thanks @DanielLeens for confirming the 
behavior on `89f91a89af`: when the probe invocation's `.get()` throws 
`ExecutionException`, both the vertex and the IMap state move to `FAILING`, 
exactly as for a definitive `false`. As noted in that comment, the remote task 
was a test double, so this has not yet been shown with a healthy real worker 
during a real partition or GC pause.
   
   @zhangshenghang, if you also observed this in a real failover, the master 
and worker logs around the switch (including the "checkTaskGroupIsExecuting 
return false" line) and the cluster setup would help validate the fix against a 
real case.
   
   @imRYiUK, thanks for volunteering. The plan sounds reasonable: keep the 
change inside `PhysicalVertex`, retry the probe a bounded number of times (3 
attempts, 1s apart) on `ExecutionException`, still fail the vertex immediately 
when the worker is no longer a cluster member or returns a definitive `false`, 
keep the result `false` if every attempt fails so the existing fallback is 
unchanged, and stop retrying on interrupt while keeping the interrupt flag. The 
unit tests you listed (transient failure retried, definitive `false` not 
retried, bounded attempts, interrupt) cover the cases discussed here. A PR is 
welcome.
   
   <!-- streview-comment:2434 -->


-- 
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