At Mon, 23 May 2011 20:45:51 +0200,
krimson wrote:
> 
> I've been trying to reproduce this but managed to do so only once. 
> However, I can reproduce a similar error with a different scenario and 
> now have the feeling it is some sort of timing issue. Here's what I do:
> 
> root@styx:~# killall sheep
> root@styx:~# sheep /sheep
> root@styx:~# collie vdi list
>    name        id    size    used  shared    creation time   vdi id
> ------------------------------------------------------------------
> failed to read a inode header 1131155, 0, 42
> failed to read a inode header 2701199, 0, 42
> failed to read a inode header 13864185, 0, 42
> 
> But then the next time I repeat the list command (after say one second) 
> it works fine:
> 
> root@styx:~# collie vdi list
>    name        id    size    used  shared    creation time   vdi id
> ------------------------------------------------------------------
>    deb01        1  8.0 GB  136 MB  0.0 MB 2011-05-23 20:38   114293
>    xxx          1  5.0 GB  0.0 MB  0.0 MB 2011-05-23 20:37   29378f
> 
> Does this make any sense ? Hopefully you can reproduce this. I am using 
> a git checkout from yesterday and
> currently testing with a 2 node setup (copies=2), on an ext4 filesystem 
> on both nodes with user_xattr flag. sheep.log does not show anything 
> except for an "accepted" and "closed" message when I do the list command.

I've reproduced the same error.  This happens when Sheepdog is not
ready for I/O operations because of, e.g., node membership change or
data object recovery.  I/O requests from VMs are retried in this case,
but requests from collie are not retried and result in errors.

Thanks for you report.  I'll fix the problem.

Kazutaka
-- 
sheepdog mailing list
[email protected]
http://lists.wpkg.org/mailman/listinfo/sheepdog

Reply via email to