[
https://issues.apache.org/jira/browse/ARTEMIS-4089?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17631938#comment-17631938
]
john lilley commented on ARTEMIS-4089:
--------------------------------------
Hi Clebert,
We figured out that auto-delete is not good for these situations. Auto-create
is fine. We have some per-job-instance queues whose number would grow without
bound if they aren't reaped, but we are cleaning those up in an explicit ageout
process now. I think its good to keep this bug open, even if we don't fix it,
so that there is a clear record of what is happening.
Thanks
john
> Auto-deleted queue with active producer leaves producer disabled (or impotent)
> ------------------------------------------------------------------------------
>
> Key: ARTEMIS-4089
> URL: https://issues.apache.org/jira/browse/ARTEMIS-4089
> Project: ActiveMQ Artemis
> Issue Type: Bug
> Components: ActiveMQ-Artemis-Native
> Affects Versions: 2.26.0
> Reporter: john lilley
> Assignee: Clebert Suconic
> Priority: Major
> Attachments: image-2022-11-09-14-33-14-216.png,
> image-2022-11-09-14-33-54-881.png, image-2022-11-09-14-35-52-506.png,
> image-2022-11-09-14-36-13-274.png, image-2022-11-09-14-36-41-631.png,
> image-2022-11-09-14-40-30-793.png, scratch.zip
>
>
> -DISCLAIMER: this bug happens sometimes and then goes away, even using the
> simple test case I've attached. I would love to know why or how it comes and
> goes, but I just don't know. It has been happening for weeks, I was able to
> reproduce this morning using the attached project and procedure, and now ...
> everything works just fine. I'm pretty sure this isn't a false alarm,
> because we tracked down bugs in our software to this phenomenon, and
> everything got better when we disabled auto-delete of our queues. Anyway,
> here goes:-
> UPDATE: I figured out exactly what is the right sequence to reproduce.
> Configure Artemis to auto-delete queues.
> * there is a queue producer and queue consumer running in different processes
> * the producer sends a few messages, and the consumer drains the queue
> * quit the consumer process
> * enough time elapses that Artemis auto-deletes the queue
> * send a few messages from the producer
> * start the consumer process
> Expected: queue is re-created and consumer is delivered messages when it
> starts
> Actual: producer send() fails silently, queue is not re-created, consumer
> sees no messages
> To reproduce, build the attached maven project in scratch.zip. The version
> of java doesn't seem to matter - at least, Java 8 and 17 are the same.
> In separate consoles run
> {{java -cp target/scratch-10.0.1-SNAPSHOT-shaded.jar
> net.redpoint.test.TestSend}}
> {{java -cp target/scratch-10.0.1-SNAPSHOT-shaded.jar
> net.redpoint.test.TestConsume}}
> You will see something like
> !image-2022-11-09-14-33-14-216.png!
> Hit enter on the sending console, and it will send a message (or two, on
> Windows)
> !image-2022-11-09-14-33-54-881.png!
> Ctrl-C to kill the consumer but _leave the producer running_
> Bring up the Artemis console and browse queues
> !image-2022-11-09-14-36-41-631.png!
> Wait just long enough for the Artemis console to show that "myqueue" has been
> auto-deleted
> !image-2022-11-09-14-40-30-793.png!
> (or wait until the artemis command-line window logs that the queue was
> deleted)
> Hit Enter on the producer window. It will send a message.
> But.. the queue is not recreated, and the send() fails silently.
> Start the consumer again. It does not receive any messages.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
