[rabbitmq-discuss] RabbitMQ writes non stop to disk
raphael at rightscale.com
Thu Oct 27 05:18:56 BST 2011
This is the output of the same command line on the broker with high i/o:
Device: rrqm/s wrqm/s r/s w/s rsec/s wsec/s avgrq-sz
avgqu-sz await svctm %util
dm-0 0.00 0.00 0.00 83.62 0.00 6217.08 74.35
0.06 0.78 0.50 4.20
So you're right I was looking at blocks written per second. Is there a way
to easily retrieve the average ingress rate across all queues for a broker?
Maybe what's happening is a rogue client is spamming that specific server.
I'd like to see the average ingress rate across all our brokers to validate
Thanks for the help troubleshooting!
On Wed, Oct 26, 2011 at 6:29 PM, Matthias Radestock
<matthias at rabbitmq.com>wrote:
> On 27/10/11 01:47, Matthias Radestock wrote:
>> Can you find out what files the 6k/s writes are going to? It can't be
>> just the persistent message store since, as you observed, that's only
>> writing 16M about every couple of minutes. And while I would expect to
>> see some writes to the queue index journal files, these shouldn't add up
>> to anywhere near 6k/s with the traffic levels you are reporting.
> Btw, are you definitely looking at the correct column in the iostat output?
> As a test I ran 'MulticastMain -f persistent -q 1 -r 17 -s 50000' (i.e.
> 17msg/s, persistent, qos=1, 50k message size), and the same with "-r 8" -
> which roughly corresponds to the rates I saw reported on the two queues in
> the system with non-negligible traffic. I adjusted rabbit's memory
> thresholds s.t. the target_ram_count of the queues dropped to zero but w/o
> raising the memory alarm. 'iostat -d -x 10' shows:
> Device: rrqm/s wrqm/s r/s w/s rsec/s wsec/s avgrq-sz
> avgqu-sz await svctm %util
> sda 0.00 219.70 0.00 57.70 0.00 2219.20 38.46
> 0.61 10.57 10.55 60.90
> So that's just 57.70 writes/s.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the rabbitmq-discuss