Thanks for following up, Fredrik. I believe the changes from ARTEMIS-4065 [1] will resolve your issue in 2.28.0. You can test your reproducer with a snapshot [2] in the mean-time if you like.
Justin [1] https://issues.apache.org/jira/browse/ARTEMIS-4065 [2] https://repository.apache.org/content/repositories/snapshots/org/apache/activemq/apache-artemis/2.28.0-SNAPSHOT/ On Tue, Jan 10, 2023 at 6:16 AM Fredrik Jonson <fred...@jonson.org> wrote: > In Justin Bertram wrote: > > > I see you're using a metrics plugin. Is this what you're using to > > determine the message count? > > Indeed. Though I believe I have confirmed that artemis post office api > report the same value as the micrometer metrics plugin. See test case > below. > > My current hunch is that Artemis somehow confuses the queue page count and > the message count given the journal files from our deployment. > > Just to ensure that I have not misunderstood, am I correct to assume that > the message count should not include the page count? I.e regardless if > messages are paged or not, the page count should not be the baseline for > the > message count, right? > > When I run 'atermis data print' I can see that the section Page Counters > matches the baseline for artemis.message.count from the broker metrics. > > ### Page Counters > Queue 52 value=146 > Queue 42 value=1 > > I have created a small reproducible test case, with journal files, > published > at https://github.com/fredrikjonson/apache-artemis-metrics-message-count > if > anyone is curious. > > I have found a workaround. After running 'artemis data recover' and then > removing the old journal files and replacing them with the recovered > journal, the metric artemis.message.count again as expected reports a > baseline of 0.0 on the affected queues. > > Unfortunately as mentioned we are uncertain when and how the issue > manifested itself, and do not know which version between 2.24 and 2.27.1 > was > used at that time. Unclean jvm shutdowns occurred too, unrelated to > Artemis. > > Anyway, given that 'data recover' appears to resolve the issue, we consider > the issue settled from our perspective. Though of course I'm still a bit > curious how we manged to confuse Artemis in this way. :) > > Thanks again for all your input Justin! > > -- > Fredrik Jonson > >