Github user harishreedharan commented on the pull request:
https://github.com/apache/spark/pull/5385#issuecomment-113336608
On that note, not a whole lot should change between spark streaming and
spark for dynamic allocation I think. The only thing of concern is the timeouts
- because we don't want to lose blocks that have not been processed yet or
ones that are required for windowing. I am wondering whether it makes sense to
add some way of "pinning" blocks so executors don't get removed.
I guess it is alleviated to a very large extent with the direct kafka
dstream which should work out of the box with dynamic allocation, since no data
is actually downloaded until it is ready to be processed, which happens within
tasks which means it won't be killed by `ExecutorAllocationManager` while
downloading data.
---
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]
