When this occurs, are some consumers "paused" while others merrily churn through their backlog, or are all consumers "paused"?
Is the rate of publishing messages to each message group relatively uniform, both across time and from group to group? JConsole u s an easy way to get a rough idea of whether expensive GCs are occurring. On Aug 21, 2015 5:00 AM, "Richard Sinek" <rsi...@intouchgps.com> wrote: > We have a finite set of message groups - 20 (0-19) > > We are monitoring the consumers via JMX and at one time were having issues > with memory leaks and GC operation that have since been solved. Now > everything looks good on the consumer side. > > There are variations in how long messages take to process - ranges in > roughly 50 to 150 ms but nothing that ever takes a really long time. I am > not seeing signs that they are not processing the messages - more that they > just stop receiving them. However, the logs do often indicate that a > message > is in transit... I will have to find one for exact wording. > > The consumers usually start when there are messages in the queue. In our > testing we have often started the consumer with well over 100k messages in > the queue and it will usually churn through them without issue - which can > sometimes take an hour or two. In ideal circumstances the consumers would > remain running indefinitely - we only start and stop them for updates and > because of this problem. > > This is what is perplexing, because the consumers are operating under > heaviest load at these times and we never saw an issue. If the normal > incoming message rate is low enough we do not usually see a pause, it is > when the message rate flowing into AMQ increases that we see the > problem.... > in other words it seems to be more related to the rate at which messages > flow into AMQ than the rate at which it dispatches to the consumer. > > As I said in the original post - when we have to completely separate > consumer applications processing from the same AMQ instance they pause and > resume together. It does not seem possible that the consumers could > simultaneously run into the same memory/resource issue. What is common > between them is the AMQ instance, database server and Redis but the latter > two are also shared by other consumers that are not affected at the same > time. > > > > -- > View this message in context: > http://activemq.2283324.n4.nabble.com/AMQ-pauses-sending-to-consumers-tp4701242p4701267.html > Sent from the ActiveMQ - User mailing list archive at Nabble.com. >
