[rabbitmq-discuss] High load average with rabbitmq-server 3.1.x

Matthias Radestock matthias at rabbitmq.com
Mon Aug 19 07:52:49 BST 2013


On 19/08/13 03:58, Dan_b wrote:
> We have recently upgraded from rabbitmq-server 2.6.1 to 3.1.x (we have one
> cluster running 3.1.5 and one running 3.1.3).  On both servers we have seen
> the load average increase significantly, with it peaking up to 2x the number
> cores.  The CPU utilization and IO wait aren't affected and it seems to be
> down to internal locks.

I'd guess those FUTEXes have something to do with timers. They are 
immediately followed by fsyncs, which in rabbit are typically triggered 
by timers. So the sequence you are seeing is

1. write message to disk
2. wait for timer to expire
3. fsync


> write(27,
> "\0\0\0\0\0\3\324=\226(\16\265\355\200\307\7J\34b\4\240\"\355)\0\0\0\0\0\0\0\0"...,
> 48) = 48
> futex(0x2aaaab594524, FUTEX_WAIT_PRIVATE, 235521, NULL) = -1 EAGAIN
> (Resource temporarily unavailable)
> futex(0x2aaaab5944f8, FUTEX_WAKE_PRIVATE, 1) = 0
> fsync(27)                               = 0

So nothing to worry about.


More information about the rabbitmq-discuss mailing list