[rabbitmq-discuss] rabbit disk_mode branch eating up all RAM, including swap, dying
brian at echonest.com
Fri Oct 9 21:07:51 BST 2009
Hi Matthew, all, I had a couple people here test out this new branch vs. our
old branch on a local VM. One thing we didn't know is that our messages
were relatively big -- 10KB each (I thought they were always much smaller.)
On a virtual machine with 1.5GB ram, we saw the following behaviour:
latest stable erlang + rabbitmq (20980) died once, and another time became
unresponsive, both times just before it hit ~2M messages (9KB each). CPU was
pegged around 90 %, and swap grew to almost 5GB.
latest stable erlang + rabbitmq (21444) never died, even with 3.5M+
messages. CPU was 90 %, and swap never grew past ~1.7GB.
We're going to upgrade to 21444 on our host and see how it goes.
On Thu, Oct 8, 2009 at 7:20 AM, Matthew Sackman <matthew at lshift.net> wrote:
> Hi Brian,
> On Mon, Oct 05, 2009 at 11:04:21AM +0100, Matthew Sackman wrote:
> > However, if you just had billions of
> > messages in there, then even in a non clustered setup, I could believe
> > mnesia would be taking a long time to start up and that might cause the
> > above error.
> I've just come across this thread:
> which does indeed seem to support the idea that even modest mnesia
> databases can take rather more than the 30 seconds we give them to
> start up. Given that I would imagine this could happen even in v1.7
> with, say, a couple million durable queues, exchanges and bindings,
> this is certainly something we will look at as the 30 second timeout
> may well be grossly too short.
> rabbitmq-discuss mailing list
> rabbitmq-discuss at lists.rabbitmq.com
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the rabbitmq-discuss