[rabbitmq-discuss] hanging on message sending with py-amqplib

[tsuraan]

> For the archives, setting a prefetch message count limit of 1,000 is
> allowing my other agents to proceed.  I'm concerned that there is
> might be a starvation condition inside of rabbit, but the throttling
> does seem to keep it from manifesting.

Nevermind that, it solved one case of hanging, but I have another on a
totally idle machine where all my tx_commit calls are hanging.  There
is no disk activity at all; the system is just stuck with a small
handful of processes (consumers/producers) stuck at tx_commit.  I can
restart the processes, and they run until their first tx_commit, and
then hang.  top shows no CPU activity, and iotop shows no disk
activity.  This is also with rabbit 2.1.0 and erlang 13B4.

A sample python script that's hanging for me is:

>>> c=connect()
>>> from amqplib.client_0_8 import Message
>>> ch=c.channel()
>>> ch.tx_select()
>>> ch.basic_publish(m, 'amq.direct', 'foo')
>>> ch.tx_commit()

The tx_commit call just never returns.  There isn't anything
apparently consuming from any queue, as the disks are entirely idle,
so I don't know what's going on here.  I have full shell access to
this machine, so any commands that you can think of that could be done
through remsh, I'd be happy to try.  I don't know if I can get you
guys direct access to the machine, but I can try if you need me to.

I have a total of 13 queues, most of which are pretty close to empty.
One of them has ~50,000 and another has ~600,000 messages.  All are
durable, and all messages are persistent.  The huge queue's consumer
is completely disabled right now, so its consumer isn't generating any
traffic of any kind.