[rabbitmq-discuss] Flow control behavior

Mon Apr 21 17:43:32 BST 2014

I could have sworn at one point, flow control would kick in when consumers
couldn't keep up with publish rates.  I recall running tests about a year
ago where consumer slow downs would slow down publishes.  But I could also
be out of my mind with more than a few loose screws (actually I know I've
got a few loose).  But anyways, while testing publishing, it appears flow
control only kicks in when the queues can't keep up due to disk/io or ram
utilization. The rabbit docs seem to verify this. Did something change on
this or am I imaging how things used to be?

This is my test setup:
1 node, publishes via shovel to two other servers.
Publish at 800 msgs/sec to node 1, shovel at 800/sec, consume at 15
messages/sec from other two servers.
Publish at 800 msgs/sec to node 1, no shovel, consume at 15 messages/sec
from node 1

Observations:
  1)  It appears that as long as rabbit has ram and disk space, i can
continue to publish at 800/sec no problem.  Even if the remote side slows
down to 400/sec over the shovel, rabbit on node 1 continues to publish and
backlog locally as long as node 1 has ram/disk capacity.
  2)  Remote sides going up and down have no impact on publishing rates,
other than possibly slowing things down due to having to read disks to get
the messages backlogged to disk.

Was there ever a time where flow control was based upon consumer rate not
queue rates?
Jason

-- 
Jason McIntosh
https://github.com/jasonmcintosh/
573-424-7612
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20140421/3166435c/attachment.html>