Hello Steve. Your report is a little surprising because a 1-3 seconds pause is huge. I'm curious under what conditions you see that latency - is it only when running at full load?
Things that can be done to diagnose this down further: 1. View the broker's average enqueue time for messages in the queue - do the two version of the broker show significantly different values for this metric? (note this is less effective if a large backlog of messages ends up sitting in the broker for longer). 2. Add logging in the broker with timing information for the messages 3. Use a test broker plugin that records the time the message is received and the time it is dispatched to a consumer 4. Measure system performance (disk I/O, network I/O) and compare between the runs Questions that come to mind: - How quickly are the messages consumed from the queues? I.E. is there ever a backlog of messages in the queue, and if so, how big does it get for both versions of the broker? - How is the 1-3 second pause vs. instantaneous timing being meaused? It is possible that default settings have changed, or additional functionality was added that impacts performance. BTW, is the messaging use-case that sensitive to latency? Art On Tue, Jan 23, 2024 at 8:58 AM Steve Hill <sh...@futurelogix.co.uk> wrote: > Hi: > > We have successfully been using ActiveMQ Classic for many years. > Currently we are running ActiveMQ 5.15.4 and it successfully processes > millions of messages without issue. Our setup is fairly simple, we have > it operating in master/slave fashion utilizing shared storage in the > event ActiveMQ fails from one node to another. > > Several years ago we attempted to upgrade to 5.16.x but ended up having > to remove it from production as we were seeing performance issues. We > have now attempted to upgrade to 5.18.3 and again are experiencing > delays; fortunately this time we caught it in our test environment! > > This is consistent across applications that are consuming from the > queue's, we are seeing the initial time to retrieve the message from the > queue goes from near instantaneous on 5.15.4 to 1-3 second pause after > upgrading to 5.18.3. We make use of pooled connections to ActiveMQ, > depending on the queue there could be between 5 and 80 consumers. The > configuration is vanilla and the following policy is the largest change. > > > <policyEntry queue=">" enableAudit="false" > producerFlowControl="false" queuePrefetch="1" maxPageSize="10" > lazyDispatch="true" expireMessagesPeriod="600000"> > </policyEntry> > > Does anyone have any ideas about how we can diagnose the issue further; > we would really like to become current on the version of ActiveMQ we are > running. > > Thanks! > Steve. > >
