[rabbitmq-discuss] Survey: RPC volumentrics + Uptime stats

Thu Dec 10 16:48:59 GMT 2009

On Thu, Dec 10, 2009 at 8:17 AM, Ben Hood <0x6e6562 at gmail.com> wrote:
> Hi,
>
> I'm just wondering if anybody in the community has any statistics that
> they could about production deployments of Rabbit.
>
> Basically I'd like to get a feel for some volumetrics of real
> production instances using Rabbit as an RPC mechanism such as:
>
> - What is the peak and average of number of RPC rountrips per second?

Our volume is low relative to Rabbit's capacity. I might see as many
as 30 calls per second, but latency is high (our apps).

> - How many clients and brokers was this achieved with?
> - What is the Rabbit portion of the overall roundtrip latency?

Negligible a percent as our call roundtrip latency is typically quite high.

> - How many RPC timeouts were observed?
> - Is RPC over Rabbit a good idea anyway?

It's overkill if you can by with a simple client/server topology. If a
system is likely to grow in complexity, the investment in messaging
patterns may well pay for itself with the flexibility it gains.

> Also, on an unrelated note, can anybody share any experiences of their
> typical uptimes for Rabbit? Is there any feeling that a heavily used
> Rabbit needs to be rebooted periodically?

My only problem with Rabbit server has been when we're letting
messages pile up in queues. A couple issues here:

1. The boolean flags for exclusive, autodelete, etc. have, some subtle
(IMO) implications on some very important memory issues in the broker.
Once you get them sorted out, it's smooth sailing, but there is that
important learning curve.

2. Not having a message time to live or expiration means that you
*really* need to make sure that either your queues are auto deleted or
that you're keeping up with the inflow.

I understand that there are features planned (or have landed) that
help with these issues. But that's really the only set of problems
I've faced.

Garrett