Github user andrewor14 commented on the pull request:
https://github.com/apache/spark/pull/2798#issuecomment-59450574
Hey @PraveenSeluka I took a closer look at this. I think functionality-wise
this basically does exactly what we need. I have tested this on a cluster and
it seems to work (though there were some thrashing when I removed executors,
but I think there's a quick fix for that). However, I think we should move the
functionality into the scheduler, because the "autoscaling server" should
behave differently for each cluster manager, whereas right now it's specific to
Yarn. Also, for Yarn client mode, there is already an existing actor on the AM
side (`MonitorActor`) that we can probably reuse.
I don't know what your timeline looks like, but I would like to get a
first-cut working prototype for SPARK-3174 as soon as possible before the 1.2
release (~end of month). Is it OK if I build my prototype on top of the changes
in this PR? We'll be sure to include you on the list of contributors for the
release.
---
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]
