[sheepdog-users] WARN wait_forward_request messages

Liu Yuan namei.unix at gmail.com
Mon Oct 13 07:17:12 CEST 2014


On Sat, Oct 11, 2014 at 02:27:34PM +0200, Maxim Terletskiy wrote:
> Hi!
> 
> I see similar messages in log:
>  WARN [gway 18794] wait_forward_request(413) poll timeout 1, disks
> of some nodes or network is busy. Going to poll-wait again
> 
> Some of nodes have Infiniband for IO connection. This nodes have
> much less warnings in logs(0-10 per day with IB, 10-20 per hour
> without IB). So looks like the problem is in network. But after
> couple of days monitoring network activity with vnstat I can't see
> any traffic peaks which could be the reason of this messages.
> 
> Where can be the bottleneck?
> 

The disks. poll timeout for every 30 seconds. It means that you had disks
don't process the requests in 30 seconds.

This is a warning some disks of your nodes are too slow (or too many requests
hitting on in a burst).

Sheep will take care of this problem and resend the requests after timeout.

Thanks
Yuan



More information about the sheepdog-users mailing list