tillrohrmann commented on a change in pull request #7880: [FLINK-11336][zk]
Delete ZNodes when ZooKeeperHaServices#closeAndCleanupAllData
URL: https://github.com/apache/flink/pull/7880#discussion_r262018551
##########
File path:
flink-runtime/src/main/java/org/apache/flink/runtime/highavailability/zookeeper/ZooKeeperHaServices.java
##########
@@ -225,6 +227,12 @@ public void closeAndCleanupAllData() throws Exception {
exception = t;
}
+ try {
+ cleanupZooKeeperPaths();
Review comment:
True, `ZKPaths#getPathAndNode` and `ZkPaths#makePath` look like good helper
functions. I'll update the PR accordingly.
At the moment, I think the problem with stand-by `JobMasters` can occur. The
`RunningJobsRegistry` was intended to solve this problem. However, at the
moment, we clear everything once the `ExecutionGraph` reaches a globally
terminal state. Instead what one could do is the following:
- Only delete the `zNodes` not belonging to the running jobs registry if you
are the leader transitioning the `ExecutionGraph` into a terminal state
- If you are the last `JobMaster` candidate which gains leadership, delete
also the running jobs registry zNode. For that to work one needs to check the
number of leader contenders.
I think this is actually a follow up issue because it is a bit bigger
refactoring.
----------------------------------------------------------------
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]
With regards,
Apache Git Services
