Hello JB ! Hope you are well !
>1. The message goes in redelivery (because it expired or client >rollback transaction) and so it can be taken by another consumer. As >you use session_transacted, the "first" client has to deal with the >rollback For me that was not so likely because the message are read with a 1 second interval on two different service pods. How could i confirm that ? Also logging the messageId would help right ? >2. Do you see "poison ack" in the log ? No trace in logs but we will double check. >Oh by the way, what's your consumer prefetch ? I guess it's more than 1 ? Yes, we use prefetch to 100 (over tcp openwire). Thanks for support ! i am off this week but probably i will ping you directly in coming weeks since you proposed it. Best, Fred Le lun. 4 mars 2024 à 07:28, Jean-Baptiste Onofré <[email protected]> a écrit : > Oh by the way, what's your consumer prefetch ? I guess it's more than 1 ? > > Regards > JB > > On Fri, Mar 1, 2024 at 4:52 PM Frédéric Curvat <[email protected]> wrote: > > > > Hello ! > > > > At my company we are using Apache ActiveMQ 5.18.3. > > We suspect that in some rare cases, a queue message is read twice by > > different consumers. > > For more context : > > - broker is classic primary/secondary (secondary started but not active - > > not a network of brokers). > > - we are using persisted queues with PostgreSQL backend. > > - A single queue is being read by several consumers : 10 consumers for a > > single java app deployed in HA other several k8s pods. > > - We use SESSION_TRANSACTED session for either consumers and producers. > > - We use PooledConnectionFactory with 1 connection, > > maximumActiveSessionPerConnection 500, expiryTimeout 10000 > > We see no transaction or other error in logs, either service of activemq > > broker at the time of the "double read". > > > > Has something like this already been seen ? Can it be a bug or a > > misconfiguration somewhere ? > > > > Best, > > > > Fred >
