Github user mwws commented on the pull request:
https://github.com/apache/spark/pull/8760#issuecomment-160864884
@kayousterhout It's an interesting discussion on PR: #3541. As I
understand, this PR can satisfy both use case.
For @davies , you might want to use `ExecutorAndNodeStrategy` by change
`spark.scheduler.blacklist.strategy` configuration to enable host level
blacklist. By default if 3 executors on the same host are blacklisted, we think
the host should also be blacklisted. Also the threshold are configurable.
For @mridulm , I tried to keep the same behavior as before in this PR, the
default "SingleTaskStrategy" is what you want. It maintain an executor level
blacklist for every TaskSet, and the timeout mechanism is retained. (Although
`spark.scheduler.executorTaskBlacklistTime` is renamed to
`spark.scheduler.blacklist.timeout`, but the old configuration name is also
supported to keep backward compatibility. So nothing need to be changed from
your side.)
---
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]
