[rabbitmq-discuss] high number of unack'ed messages when queue grows large

[mmenti at gmail.com]

On Fri, May 7, 2010 at 2:49 PM, Matthew Sackman <matthew at lshift.net> wrote:

> Hi Mario,
>
> On Fri, May 07, 2010 at 02:43:42PM +0100, mmenti at gmail.com wrote:
> > We've been seeing an issue where the number of unack'ed messages suddenly
> > starts growing much larger than the number of connected consumers,
> whenever
> > the queue gets to a certain size  (somewhere around 50k messages in the
> > queue, with 250 consumers connected). We've noticed that by reducing the
> > number of consumers, the issue becomes less pronounced, but was wondering
> if
> > anyone else is seeing this, and if there's any ways to avoid this? We're
> > using rabbitmq 1.7.0 on an 64-bit Amazon EC2 Ubuntu 8.04 hardy AMI.
>
> There are two parts to this issue. Firstly, some priorities of messages
> internally were changed either in 1.7.1 or 1.7.2 so that acks would have
> higher priority internally than publishes. This helps solve this
> problem, and so I would recommend you upgrade.
>

Thanks, good to know - we'll try and do that as soon as is feasible.


>
> Secondly, you can help by setting a qos prefetch count on your consumers
> - if, say, you set it to 10, then you're ensuring that each consumer
> will only receive 10 messages before it must a) ack a message and b)
> that ack is received *and*processed* by Rabbit. This will ensure the
> unacked message count stays low.
>


Strangely we already have prefetch set to 1 on the consumers, and still see
this issue.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20100507/5133c523/attachment.htm