[rabbitmq-discuss] RabbitMQ hangs, does not accept connections

[Dmitri Minaev]

On 7 March 2012 19:22, Simon MacMullen <simon at rabbitmq.com> wrote:
> On 07/03/12 08:57, Dmitri Minaev wrote:
>>
>> Yes, there's quite a lot of those 'failed to accept' messages.
>> Actually, this is the second time when Rabbit stops since installation
>> of that tcp_acceptor patch, and every time those messages appear in
>> large number.
>
>
> So there's few things going on here. The primary issue is that RabbitMQ is
> running out of file descriptors due to too many connections being opened,
> and for various reasons the internal accounting that is supposed to prevent
> this from causing harm is getting out of sync with reality. There are
> already improvements that will be coming in the next release that will
> improve this situation and we have some ideas for how to fix it altogether.
>
> But in the meantime you should look at increasing the number of file
> descriptors that are available to RabbitMQ.

Oh... Thanks. I thought, that if TCP connection is accepted, number of
file descriptors should have no effect on the further events, since
the socket already exists. Even now I can connect to port 5672 on that
server, but AMQP does not respond.

If I understand correctly, the number of file descriptors used by
RabbitMQ in normal situation is roughly equal to the number of
`rabbitmqctl list_connections` + some constant (~30)? In case of that
hanging server, the number of AMQP connections never was close to the
FD limit (ulimit -n is 1024, fs.file-max = 1605698). The graph
reflecting  the number of open AMQP connections is attached to this
message.

-- 
With best regards,
Dmitri Minaev
-------------- next part --------------
A non-text attachment was scrubbed...
Name: rabbitmq_connections-month.png
Type: image/png
Size: 21291 bytes
Desc: not available
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20120309/9b2b9016/attachment.png>