GitHub user rmetzger opened a pull request:
https://github.com/apache/flink/pull/717
Improve Flink on YARN stability
This issue is addressing three issues:
- NPE in the history view of the JobManager
- Change the Akka configuration for YARN TaskManagers so that they kill
themselves on fatal akka errors (this will make sure that YARN notices when a
TM died through an OOM error)
- I've changed the default configuration values for the amount of heapspace
we are removing from the requested TM size.
For small containers, it seems that we need at least 600 MB of security
margin. I know that this is far from perfect, but it will make things stable in
the meantime.
I would like to merge this soon because the change is improving the YARN
experience and we have users affected by this.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/rmetzger/flink flink2079-complete
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/flink/pull/717.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #717
----
commit 484696cce2d2bb67741b64d00d151b8fe9275df1
Author: Robert Metzger <[email protected]>
Date: 2015-05-19T17:38:02Z
[jobmanager] Fix potential null pointer exception in jobmanager webfrontend
commit 617a6b1cc49b4d7142ddf61de63ffed4886b2bb0
Author: Robert Metzger <[email protected]>
Date: 2015-05-22T11:51:02Z
[FLINK-2079] Add TaskManager deathwatch thread for YARN case
commit 93f0a05281a5de36909c5238b8a23a56ea991aaf
Author: Robert Metzger <[email protected]>
Date: 2015-05-22T16:10:15Z
[yarn] Adjust default values for YARN heap memory cutoff
----
---
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.
---
