mustafa kamal created CAMEL-24545:
-------------------------------------
Summary: Camel Shutdown Locked State With ZookeeprCluster Service
Key: CAMEL-24545
URL: https://issues.apache.org/jira/browse/CAMEL-24545
Project: Camel
Issue Type: Bug
Components: camel-zookeeper, camel-zookeeper-master
Affects Versions: 4.22.1, 4.18.5
Environment: One ZK instance and one camel instance running
ZooKeeperClusterService you need to use the main branch or any fix branch as
this PR didn't span into a released version.
Reporter: mustafa kamal
Attachments: dump.txt
During shutdown, Camel's ClusteredRoutePolicy.onRemove and the Curator leader
selector thread both run concurrently because some routes keep the shutdown
window open long enough. These two threads acquire the same two locks in
opposite order, causing a classic deadlock that permanently hangs the JVM.
You can check the attached Thread Dump to see the technical level of the issue
more.
Possible fix is to surround `fireLeadershipChangedEvent((CamelClusterMember)
null);` line 154 in `ZooKeeperClusterView` class, with `if
(!isStoppingOrStopped())` that will solve the dead lock issue
Another issue it solved the unnecessary firing of the Leadership Change event
because on stopping/shutting down the view all the registered listeners will be
removed so any reaction they need to do should be before stopping the view
because stopping the view is not as the same as Losing the Leadership Lock so
that's more of an enhancement than breaking functionality.
To reproduce the issue run the application wait until it's the Leader by
acquiring the lock in ZK then shutting down the application, not all routes
will produce the issue some routes have a long enough shutdown windows that
allows the Dead Lock to happen
This issue is related to the PR [https://github.com/apache/camel/pull/25611]
for ticket https://issues.apache.org/jira/projects/CAMEL/issues/CAMEL-24457
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
