[
https://issues.apache.org/jira/browse/AMQ-9747?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18009107#comment-18009107
]
Christopher L. Shannon commented on AMQ-9747:
---------------------------------------------
Thanks for the detailed bug report. I agree that if the broker continues to
run, the checkpoint runner should not stop. Most likely this issue hasn't come
up before because most users just use standard KahaDB and are not using
LeaseLockerIOExceptionHandler so the startStopConnectors flag is false. That
means if there is an IOException that is thrown by the runner, the entire
broker will just shut down.
I think the fix is probably to do some variation of what you already did. I
still think the DefaultIOExceptionHandler should be used because it will shut
the broker down on IOException under the standard use case (and it implements
the IOExceptionHandler so it's easy for a user to replace the behavior). But
instead of letting unchecked exceptions propagate and kill the thread, we
should catch and log a warning or error so that the user can be alerted but let
the cleanup thread keep running.
> Kahadb checkpoint runner thread dies without catching exception
> ---------------------------------------------------------------
>
> Key: AMQ-9747
> URL: https://issues.apache.org/jira/browse/AMQ-9747
> Project: ActiveMQ Classic
> Issue Type: Bug
> Components: KahaDB
> Affects Versions: 6.1.4, 6.1.6, 5.18.7, 6.1.7
> Reporter: ritesh adval
> Assignee: Christopher L. Shannon
> Priority: Major
> Attachments: image-2025-07-21-11-14-20-510.png
>
>
> This bug affects all version of activemq, I have added a few ones starting
> 5.18.4. This is a really critical bug, which causes checkpoint runner thread
> to SILENTLY GET KILLED (exception eaten away), causing kahadb journal files
> to KEEP GROWING, till you restart activemq. There is a bug in handling of io
> exception... see screen shot below. In the catch block its calls
> brokerService.handleIOException().
> !image-2025-07-21-11-14-20-510.png! iif you take a look at the default io
> exception handler which is used, it will throw this SuppressReplyException at
> [https://github.com/apache/activemq/blob/main/activemq-broker/src/main/java/org/apache/activemq/util/DefaultIOExceptionHandler.java#L165]
> and if stopStartConnectors is true (which is if you use
> LeaseLockerIOExceptionHandler) then also it throws this
> SuppressReplyException at
> [https://github.com/apache/activemq/blob/main/activemq-broker/src/main/java/org/apache/activemq/util/DefaultIOExceptionHandler.java#L155]
> and because of this, the CheckPoint runner thread as shown in above screen
> shot would silently die, even though broker is still running...
> it seems the checkpoint runner should not be dying.... we had a situation
> where we were using EFS as our storage for kahadb... and due to a blip in
> connection EFS, an io exception in page.flush in MessageDatabase was thrown
> (we were using LeaseLockerIOExceptionHandler)... that caused
> DefaultIOExceptionHandler logic to start and stop connectors and return
> SuppressReplyException as i mentioned above, causing CheckPoint runner to
> silient get killed... while broker was still running....
>
> we had this in production and the fix we did is to extend
> LeaseLockerIOExceptionHandler and catch exception throw from
> handle(IOException ex) method and log it as warn and not propogate it up to
> checkpoint runner thread... but this is temporary fix.. i am not even sure if
> CheckpointRunner needs to use DefaultIOExceptionHandler.... it shouldn't die
> silently...
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
For further information, visit: https://activemq.apache.org/contact
