Github user andrewor14 commented on the pull request:
https://github.com/apache/spark/pull/4841#issuecomment-77228784
So my proposal that we revert all of these related patches and submit the
right fix, which is something like the following: Before we decrement the
`coresGranted`, we check whether the application has exited. If it has, then we
don't decrement it.
Note that this is not trivially easy, however, because the `Master`
currently detects application completion through `DisassociatedEvent`, so there
could be a race condition where the `Master` is notified of the executor's exit
before the application's completion. To work around this, we need to send an
explicit "unregister" message form the driver to the `Master`. I tested a rough
implementation of it and verified that it works as expected.
What do others think?
---
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]
