[sheepdog-users] vdi problem

[MORITA Kazutaka]

At Wed, 12 Jun 2013 15:43:24 +0200,
Valerio Pachera wrote:
> 
> 2013/6/12 MORITA Kazutaka <morita.kazutaka at gmail.com>:
> > you find any errors in sheep.log when you get the above errors?
> 
> Logs are big because of recovery.
> Anyway, I can see that
> 
> sheepdog001
> ...
> Jun 11 13:13:32 [main] recover_object_main(625) done:329320
> count:329322, oid:c8d12800012667
> Jun 11 13:13:32 [main] recover_object_main(625) done:329321
> count:329322, oid:c8d1280002d7a9
> Jun 11 13:13:32 [main] recover_object_main(625) done:329322
> count:329322, oid:c8d12a0002fc78
> Jun 12 05:14:46 [gway 8311] wait_forward_request(176) poll timeout 1,
> disks of some nodes or network is busy. Going to poll-wait again
> 
> sheepdog002
> ...
> Jun 11 13:13:14 [main] recover_object_main(625) done:289439
> count:289441, oid:c8d12800012667
> Jun 11 13:13:14 [main] recover_object_main(625) done:289440
> count:289441, oid:c8d1280002d7a9
> Jun 11 13:13:14 [main] recover_object_main(625) done:289441
> count:289441, oid:c8d12a0002fc78
> Jun 11 20:10:31 [gway 29882] wait_forward_request(176) poll timeout 1,
> disks of some nodes or network is busy. Going to poll-wait again
> 
> sheepdog004
> has no 'wait_forward_request'. Since I remove data from it, I dind't
> run any guest.

I guess I/O timeout happens on the guest OS.  I think we really need
to consider how to prevent recovery traffic from affecting VM I/Os.

> 
> > How did you delete the cache?
> 
>   collie vdi cache delete backup_data

Yuan, is there any possiblity that the node could flush the invalid
object cache before the above collie command deletes it?

Thanks,

Kazutaka