This can be seen consistently when sending messages over a federated link with acking turned on (not yet sure f acking is required). I also observed it during run_failover_soak but suspect that may be a separate case.
To reproduce, setup an acked federation link and send lots of messages over it. (More frequent acking will generally show the issue earlier). E.g. with two brokers on ports 5555 and 6666 set up the link between two test queues: qpid-config -a localhost:5555 --max-queue-count 5000 add queue test-queue qpid-config -a localhost:6666 add queue test-queue qpid-config -a localhost:6666 bind amq.direct test-queue test-queue qpid-config -a localhost:6666 bind amq.direct test-queue test-queue qpid-route --ack 1 queue add localhost:6666 localhost:5555 amq.direct test-queue Then start up a receiver on the destination broker: receiver --port 666 --ack 500 (ack frequency of receiver just improves throughput and makes test a little faster, it has no impact on issue) Then send lots of messages to the source broker: (for (( i=5000000; i--;)) do echo $i; done) | ./src/tests/sender --send-eos 1 --port 5555
Fixed by r750002.
Fix also includes correction for the 'push' case (i.e. where --src-is-local option is specified to qpid-route above).