Github user jerryshao commented on the pull request:
https://github.com/apache/spark/pull/9684#issuecomment-156600658
Hi @vanzin , I have no strong inclination on either solution, I can change
to the first solution.
The problem here with first solution is that the executor loss reason is
`SlaveLost`, not actually `ExecutorExited`, so it is not well informed to the
user, of course I could change the loss reason to make it more reasonable.
Besides, still I think there's a race condition between querying loss
reason and process completed containers, since they're in two threads, here we
assume processing completed containers should be later than querying loss
reason, but I don't think we could guarantee this. Maybe we could receive
`onDisconnected` later than processing completed containers. I think it is
possible, and in such case we're failed to get loss reason, so here in this PR
I changed to store loss reason if processing completed containers is ahead of
querying, unless we could accept failing to get loss reason. What's your
opinion?
Thanks a lot.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
