Hello Mat ! I think I've not received or seen any reply to my ticket. It doesn't really matter, and thanks anyway.
I know that for a given queue if multiple producers concurrently send messages, only messages from each one is supposed to be ordered. That's what we expect. I know I've documented the ticket with too many details... Both our business case and the last little test I've added later consider the order of messages from a given producer. In the first case, messages are in fact produced by different producers, but what's failing (with some messages delivered twice), is for a given producer. In the little test, the failing queue is filled by a single producer (a single consumer seda route). What I'm able to say today about the context of our issue : - messages must be sent fast, since with a slow producer it always works - filling queues with multiple producers helps to produce the issue - the failing queue consumer is slower than producers, and it processes late incoming messages that increase the queue size. Stopping the consumer for a while is the worst situation. Operation downtimes almost always lead to few errors. - only small unread messages in the queue seem not to cause it. In our real business case, messages payloads are between 1 KB to 120 MB, and the consumer is always slower, leading to many unread messages for a while. We have history data that seems to say that the problem exists with older versions of ActiveMQ (Our old implementation runs on Red Hat Fuse 6.3 with a packaged 5.11 AMQ), and has the same behavior as the standalone 5.18 we use now. I hope all that will help you to find an explanation. Thanks in advance. Regards. Le mar. 11 févr. 2025 à 14:56, Matt Pavlovich <mattr...@gmail.com> a écrit : > Hi Emphereis- > > I replied to the ticket. I have not had a time to deep-dive into al your > notes, but provided some guidance. > > We would need a concise unit test that demonstrates an ordering bug to > confirm any type of fix is needed. > > Keep in mind that the message ordering is guaranteed by an instance of a > single _producer_ and not any wider scope. Queues can be multi-threaded, so > there is no order guaranteed across multiple producers. > > Thanks, > Matt Pavlovich > > > On Feb 11, 2025, at 6:31 AM, Ephemeris Lappis < > ephemeris.lap...@gmail.com> wrote: > > > > Hello ! > > > > I'm a bit embarrassed: I reported a problem on ActiveMQ (Classic), > > regarding the disordered distribution of messages, both on this list and > by > > creating a JIRA ticket (https://issues.apache.org/jira/browse/AMQ-9653), > > but I have no feedback. I'm a bit surprised. We would really need help to > > check this point, possibly challenge our analysis, identify a > configuration > > fault, etc. > > > > I tried to gradually document the ticket with all the elements noted > during > > our tests. > > > > Could an MQ expert take a look, please? > > > > Thanks in advance. > > > > Regards. > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: users-unsubscr...@activemq.apache.org > For additional commands, e-mail: users-h...@activemq.apache.org > For further information, visit: https://activemq.apache.org/contact > > >
