Zombie consumers, how do I garbage collect them?

[Parnell Springmeyer]

I am using the heartbeat to keep track of the connection but what happens if an exception is thrown by a consumer when consuming a message that kills it and a new consumer is spawned right away but the connection and channel were never lost. We end up with a lot of zombie consumers and I don't know how to prune them.

I attempted issuing a basic.cancel before basic.consume as a catch-all cancel for that specific consumer tag since I'm using a specific (instead of random) consumer tag. But that doesn't appear to cancel any of the zombie consumers.

[Michael Klishin]

What client do you use? What RabbitMQ version? What does `rabbitmqctl report` output? How can we reproduce the issue?

MK

On 6/7/2015, at 21:07, Parnell Springmeyer <par...@plumlife.com> wrote:

I am using the heartbeat to keep track of the connection but what happens if an exception is thrown by a consumer when consuming a message that kills it and a new consumer is spawned right away but the connection and channel were never lost. We end up with a lot of zombie consumers and I don't know how to prune them.

I attempted issuing a basic.cancel before basic.consume as a catch-all cancel for that specific consumer tag since I'm using a specific (instead of random) consumer tag. But that doesn't appear to cancel any of the zombie consumers.

--
You received this message because you are subscribed to the Google Groups "rabbitmq-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rabbitmq-user...@googlegroups.com.
To post to this group, send email to rabbitm...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Parnell Springmeyer]

I'm using the Erlang client. I don't believe this is a problem with RabbitMQ per-se but it's reproducible by spawning a separate consumer process and then killing that process and letting the supervisor start it up again, I end up with multiple consumers with the same tag and only one of them is actually active.

I think I might have to propagate the consumer failure out to the process responsible for the AMQP connection so that it gets restarted too.

ᐧ

--

Parnell Springmeyer

Cloud & Mobile Software at Plum Inc.

702.445.3545

0xDCCF89258EAD874A

[Michael Klishin]

On 6 Jul 2015 at 21:43:41, Parnell Springmeyer (par...@plumlife.com) wrote:
> I'm using the Erlang client. I don't believe this is a problem
> with RabbitMQ per-se but it's reproducible by spawning a separate
> consumer process and then killing that process and letting the
> supervisor start it up again, I end up with multiple consumers
> with the same tag and only one of them is actually active.
>
> I think I might have to propagate the consumer failure out to the
> process responsible for the AMQP connection so that it gets restarted
> too.

There are two Erlang clients: network (works much like clients in other languages)
and direct. Are you using the direct one? Error handling in it can indeed be quite a bit
different.

The idea is that Erlang client processes need to be monitored in your code because
a dying consumer isn’t necessarily a reason to kill its channel, leave alone connection,
so the library lets the user decide. 
--
MK

Staff Software Engineer, Pivotal/RabbitMQ

[Parnell Springmeyer]

I'm definitely using the network driver. If I restart the process that owns the channel will that "clean up" the zombie consumers on the host?

ᐧ

[Michael Klishin]

On 6 Jul 2015 at 22:11:28, Parnell Springmeyer (par...@plumlife.com) wrote:
> If I restart the process that owns the channel will that "clean
> up" the zombie consumers on the host?

If you close the channel all consumers should be deleted. Otherwise RabbitMQ
server is unaware of your consumer process death.