Github user mxm commented on a diff in the pull request:
https://github.com/apache/flink/pull/1741#discussion_r56860123
--- Diff:
flink-runtime/src/main/scala/org/apache/flink/runtime/jobmanager/JobManager.scala
---
@@ -312,59 +323,125 @@ class JobManager(
leaderSessionID = None
- case RegisterTaskManager(
- connectionInfo,
- hardwareInformation,
- numberOfSlots) =>
+ case msg: RegisterResourceManager =>
+ log.debug(s"Resource manager registration: $msg")
+
+ // ditch current resource manager (if any)
+ currentResourceManager = Option(msg.resourceManager())
+
+ val taskManagerResources =
instanceManager.getAllRegisteredInstances.asScala.map(
+ instance => instance.getResourceId).toList.asJava
+
+ // confirm registration and send known task managers with their
resource ids
+ sender ! decorateMessage(new RegisterResourceManagerSuccessful(self,
taskManagerResources))
+
+ case msg: DisconnectResourceManager =>
+ log.debug(s"Resource manager disconnect: $msg")
+
+ currentResourceManager match {
+ case Some(rm) if rm.equals(msg.resourceManager()) =>
+ // we should ditch the current resource manager
+ log.debug(s"Disconnecting resource manager $rm.")
+ // send the old one a disconnect message
+ rm ! decorateMessage(new TriggerRegistrationAtJobManager(self))
--- End diff --
You're right that the message can be lost. Apart from the leader election
service there is currently no mechanism to detect a lost connection from RM to
the JM.
The above code is part of a special case where the RM doesn't reply to a TM
registration and the JM decides to disconnect the RM. I suppose we should keep
retrying to send the TriggerRegistrationAtJobManager until we receive a
registration of a RM.
---
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.
---
