[
https://issues.apache.org/jira/browse/FLINK-7804?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16400613#comment-16400613
]
ASF GitHub Bot commented on FLINK-7804:
---------------------------------------
Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/5675#discussion_r174834772
--- Diff:
flink-yarn/src/main/java/org/apache/flink/yarn/YarnResourceManager.java ---
@@ -395,10 +405,13 @@ public void onNodesUpdated(List<NodeReport> list) {
@Override
public void onError(Throwable error) {
- onFatalError(error);
+ runAsync(() -> onFatalError(error));
--- End diff --
I think it would be good to let the error propagate directly. In case of an
OOM exception we want to quickly shut down the JVM.
> YarnResourceManager does not execute AMRMClientAsync callbacks in main thread
> -----------------------------------------------------------------------------
>
> Key: FLINK-7804
> URL: https://issues.apache.org/jira/browse/FLINK-7804
> Project: Flink
> Issue Type: Bug
> Components: Distributed Coordination, YARN
> Affects Versions: 1.4.0, 1.5.0
> Reporter: Till Rohrmann
> Assignee: Gary Yao
> Priority: Blocker
> Labels: flip-6
> Fix For: 1.5.0
>
>
> The {{YarnResourceManager}} registers callbacks at a {{AMRMClientAsync}}
> which it uses to react to Yarn container allocations. These callbacks (e.g.
> {{onContainersAllocated}} modify the internal state of the
> {{YarnResourceManager}}. This can lead to race conditions with the
> {{requestYarnContainer}} method.
> In order to solve this problem we have to execute the state changing
> operations in the main thread of the {{YarnResourceManager}}.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
