[sheepdog-users] Sheepdog and slow node

[Liu Yuan]

On Wed, Jun 25, 2014 at 11:51:59AM +0200, Valerio Pachera wrote:
> 2014-06-25 11:45 GMT+02:00 Valerio Pachera <sirio81 at gmail.com>:
> 
> > Hi, if a node is slow to respond I see these warning messages in the log
> >
> 
> Jun 24 17:02:15   WARN [gway 12981] wait_forward_request(389) poll timeout
> 1, disks of some nodes or network is busy. Going to poll-wait again
> Jun 24 17:02:15   WARN [gway 12920] wait_forward_request(389) poll timeout
> 1, disks of some nodes or network is busy. Going to poll-wait again
> Jun 24 17:02:15   WARN [gway 12975] wait_forward_request(389) poll timeout
> 1, disks of some nodes or network is busy. Going to poll-wait again
> Jun 24 17:02:15   WARN [gway 12986] wait_forward_request(389) poll timeout
> 1, disks of some nodes or network is busy. Going to poll-wait again
> Jun 24 17:02:16   WARN [gway 12957] wait_forward_request(389) poll timeout
> 1, disks of some nodes or network is busy. Going to poll-wait again
> 
> One of 3 nodes has a hardware raid controller.
> The controller cache is temporarily disabled and the write speed drop down
> to 10M/s.
> 
> What does sheepdog do in such case?
> Are all guests' requests going to be slow down?
> May that cause any known problem?

Probably yes since our data are evenly distributed across nodes then IO requests are
evenly distributed too. So it is highly likeyly that all the guests has to deal
with slow node.

But it won't cause any problem because requests will automatically be retried
after timeout.

Thanks
Yuan