[
https://issues.apache.org/jira/browse/FLINK-6341?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15976714#comment-15976714
]
ASF GitHub Bot commented on FLINK-6341:
---------------------------------------
Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/3745#discussion_r112456827
--- Diff:
flink-runtime/src/main/scala/org/apache/flink/runtime/jobmanager/JobManager.scala
---
@@ -356,7 +356,12 @@ class JobManager(
msg.resourceManager() ! decorateMessage(new
TriggerRegistrationAtJobManager(self))
// try again after some delay
context.system.scheduler.scheduleOnce(2 seconds) {
- self ! decorateMessage(msg)
+ currentResourceManager match {
--- End diff --
This access is problematic, because we're accessing actor state from within
another thread.
> JobManager can go to definite message sending loop when TaskManager registered
> ------------------------------------------------------------------------------
>
> Key: FLINK-6341
> URL: https://issues.apache.org/jira/browse/FLINK-6341
> Project: Flink
> Issue Type: Bug
> Components: JobManager
> Reporter: Tao Wang
> Assignee: Tao Wang
>
> When TaskManager register to JobManager, JM will send a
> "NotifyResourceStarted" message to kick off Resource Manager, then trigger a
> reconnection to resource manager through sending a
> "TriggerRegistrationAtJobManager".
> When the ref of resource manager in JobManager is not None and the
> reconnection is to same resource manager, JobManager will go to a infinite
> message sending loop which will always sending himself a
> "ReconnectResourceManager" every 2 seconds.
> We have already observed that phonomenon. More details, check how JobManager
> handles `ReconnectResourceManager`.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
