[ejabberd] Message delays in ejabberd

Santiago Ignacio Poli santiago at etermax.com
Sun Mar 27 22:42:54 MSK 2016


Our ejabberd cluster is having very serious problems in production: All the
messages are taking an eternity to be processed. The IQs seem to be queued
and the responses are sometimes received 5 minutes after. Even simple
messages take minutes to reach the other end.
We have approximately 100k concurrent connections across three nodes (33k
connections per node). Each of them is running inside an Amazon EC2
r3,xlarge (4 cores, 30GB ram, 80GB SSD). We think hardware isn't the
problem because after monitoring the nodes, we haven't seen any CPU or
memory bottlenecks. Sometimes a few cores go to 100% but only for a second.

We have some custom modules (which are likely the problem). My guess is
that the IQ handlers have very small queues but I haven't seen any
configuration regarding that. We've changed the queue type from 'one_queue'
to 10 queues (the documentation states that you have to put {queues, 10}
but that didn't work. What worked was put only '10') and the problem still
persists. We are using ejabberd 15.04

What configurations should we check? Is there any specific information  we
can provide here to be more clear?

Thanks in advance

*Santiago Ignacio Poli*
*Software Developer & Scrum Master **@etermax*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.jabber.ru/pipermail/ejabberd/attachments/20160327/92b340b4/attachment.html>

More information about the ejabberd mailing list