Github user mccheah commented on the pull request:
https://github.com/apache/spark/pull/8007#issuecomment-132400369
So my next steps: I'll go with @vanzin 's suggestion for using
RpcEndpointRef.ask instead and I think that doing that the right way
invalidates a lot of the multiple-get-executor-loss-reason race condition
nastiness @markgrover is seeing. As for making sure tasks aren't temporarily
being allocated on dead executors, I'll possibly tackle that separately to keep
commits smaller in scope.
The latter problem seems harder and would involve a concerning amount of
code churn, and I'm not sure if I want to expand this commit any more than I
have to... but making more jobs fail because tasks are allocated to bad
executors and hence contribute to spark.task.maxFailures (in the case that the
executor failure is fatal, so to speak) is a non-trivial issue.
---
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]
