Can you please submit a bug in JIRA for this behavior? If you're interested in digging in and doing the investigation and the fix, that's great, but let's get the problem captured, along with enough information to reproduce the problem and test a fix. If you're interested in doing that, you might want to put a comment on the JIRA issue to indicate that, and/or join the developer mailing list to ask questions as you get into it.
If you do want to do the investigation, once you have the problem reliably reproduced I'd encourage you to attach a debugger to the broker and use that to walk through the message cursor that pages messages in from the underlying message store and the code that accepts acknowledgements from the consumer. What you described sounds likely to be in one of those two areas, so I'd suggest you start there. Tim On Tue, Mar 2, 2021, 8:30 AM Gekeler Ralf (IOC/TBC1) <ralf.geke...@bosch.io.invalid> wrote: > I am currently testing an application which uses ActiveMQ 5.15.10 and > persistent messages. The messages come in from one producer and get > dispatched to 10 consumers. The dispatching is based on message groups to > ensure data consistency. > > I have configured a prefetching limit of 100 and a message group bucket > size of 1024. Maxpagesize is 200. > > Over time the inflight count starts to decrease and increase again in > waves (see attached screenshot<https://i.stack.imgur.com/suvc6.png>). > Looks like a sawtooth pattern. I can extend the time until it starts by > increasing the maxpagesize. When it starts, the consumers don't receive any > messages before they finished processing all of the prefetched ones - > instead of continuously receiving messages. > > I don't see any limits on infrastructure metrics. Has anyone seen such a > pattern before and has a hint how I can find the root cause and mitigate > that behavior? > > > Thx, > Ralf >
