pkuwm commented on a change in pull request #1066:
URL: https://github.com/apache/helix/pull/1066#discussion_r436228222
##########
File path:
helix-core/src/main/java/org/apache/helix/controller/stages/MessageDispatchStage.java
##########
@@ -78,7 +78,13 @@ protected void processEvent(ClusterEvent event,
MessageOutput messageOutput) thr
batchMessage(dataAccessor.keyBuilder(), messagesToSend, resourceMap,
liveInstanceMap,
manager.getProperties());
- List<Message> messagesSent = sendMessages(dataAccessor, outputMessages);
+ String expectedSession =
event.getAttribute(AttributeName.CONTROLLER_LEADER_SESSION.name());
+ // An early check for expected leader session. If the sessions don't
match, it means the
+ // controller lost leadership, then messages should not be sent. This
potentially avoid
+ // double masters for a single partition.
+ List<Message> messagesSent = manager.getSessionId().equals(expectedSession)
Review comment:
I proposed to do some checks at stages like `ReadClusterDataStage` and
terminate the remaining pipeline if controller loses leadership. In our issue,
I see that leader lost leadership at `ReadClusterDataStage`. So there is no
need to continue the pipeline and it would help save some computation resource
or write traffic to EV/persist assignment. But in the discussion we agreed on
not doing that and just blocking messages. We could definitely pick it up and
discuss again. @jiajunwang
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
