Github user harishreedharan commented on the pull request:
https://github.com/apache/spark/pull/2024#issuecomment-53089849
Yes, it is expected that the driver would come up at a different address.
The driver "pings" the executors when it comes up (there is a doc on the jira +
some discussion regarding this) - which persist the driver failure. Once the
driver pings the executors, it gets the block info stored in executor memory to
reconstruct the batch of data which has not been persisted yet.
In client mode, it is tricky because the AM is unmanaged - so YARN will not
restart this - so we'd need some kind of watchdog/wrapper - for now, I think
Cluster mode is the only one we can think of.
---
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]
