Closed woodsaj closed 8 years ago
Comment by Dieterbe Tuesday Oct 20, 2015 at 20:26 GMT
(this is from the alerting dashboard in dev stack)
actually it seems like, even though the rabbit queue is being fed constantly, the workers aren't really pulling many jobs out of it. (PS: it would be nice to have the queue graphs from rabbit also available in graphite, we're not using the internal jobqueue here, we're using rabbit) also this is with a lightly loaded system, no process doing more than 50% cpu.
Comment by ctdk Tuesday Oct 20, 2015 at 20:45 GMT
@Dieterbe No ideas at the moment, but I'm in there anyway so I'll take a look.
Comment by Dieterbe Tuesday Oct 20, 2015 at 22:50 GMT
https://github.com/raintank/raintank-docker/issues/41 might have something to do with this, though if that was only it, at least some messages should be acked to rabbit.
Comment by ctdk Tuesday Oct 20, 2015 at 23:47 GMT
There aren't any un-acked messages in a litmus stack brought up with chef and not using env-load, fwiw.
Comment by woodsaj Thursday Oct 22, 2015 at 14:05 GMT
looks like the problem could be due to https://github.com/raintank/grafana/blob/master/pkg/services/rabbitmq/consumer.go#L146
Msgs are being processes sequentially. We probably need to refactor the code so that each message is processed in its own goroutine.
Issue by Dieterbe Tuesday Oct 20, 2015 at 20:19 GMT Originally opened as https://github.com/raintank/grafana/issues/496
how to reproduce: 1) update & spin up dev stack 2)
env-load load
3) open http://localhost:15672 -> both in main overview of queued messages, as well as individual queues the number of unacked = total messages. although the exchanges do say msg in rate = msg out rate, and alerting seems to work