Github user JoshRosen commented on the pull request:
https://github.com/apache/spark/pull/3550#issuecomment-65194277
I considered something like this, but I think that this re-introduces cases
where a single bad host can cause the entire application to fail. Imagine that
I have a cluster where all but one of the hosts are functioning correctly; I'll
register executors on the good hosts once at the beginning of the app and can
then experience an infinite number of executor launch failures on the buggy
host since we don't have a blacklist. So, we might have a case where the
application is able to make progress with the executors that it has but is
killed due to failed attempts to acquire more executors, since all of the
resets/decrements to the "progress towards failure" counter will only occurred
at the beginning of the app, while the increments occurred continuously.
---
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]
