[rabbitmq-discuss] Connection blocked by "flow" for more than 600 seconds

[Matthias Radestock]

On 09/10/13 00:09, Choo wrote:
> So, it seems like memory is enough, file_descriptor is enough, and disk_free
> should still be ok.  But the connection is still blocked.  But good thing
> is, the block duration seems shorter.
>
> 10.95.212.10:42336 -> 10.95.212.12:5672	blocked	0	flow	17.306112
> 10.95.212.10:42338 -> 10.95.212.12:5672	blocked	0	flow	108.213992
> 10.95.212.10:42359 -> 10.95.212.12:5672	blocked	0	flow	36.799473
> 10.95.212.10:42360 -> 10.95.212.12:5672	blocked	0	flow	146.325601
> 10.95.212.10:42361 -> 10.95.212.12:5672	blocked	0	flow	53.655967
> 10.95.212.10:42374 -> 10.95.212.12:5672	blocked	0	flow	121.259312
> 10.95.212.10:42375 -> 10.95.212.12:5672	blocked	0	flow	139.258805
> 10.95.212.10:42376 -> 10.95.212.12:5672	blocked	0	flow	137.210049
> 10.95.212.10:42377 -> 10.95.212.12:5672	blocked	0	flow	51.628542
>
> Any other areas I should check?

You might just be seeing normal flow control behaviour. Looks like you 
have lots of durable queues and are publishing persistent messages. That 
puts a huge strain on disk i/o, so that's likely the bottleneck which 
causes flow control to kick in.

Matthias.