[sheepdog-users] Write till the end of the cluster

[Liu Yuan]

On 12/13/2012 09:34 PM, Valerio Pachera wrote:
> Well, went forward with the messing :-)
> 
> I killed a node and added new bigger node of 10G.
> The recovery has been fast.
> 	collie node info
> 	Id      Size    Used    Use%
> 	0      982 MB  956 MB   97%
> 	1      982 MB  568 MB   57%
> 	2      10.0 GB 1.4 GB   14%
> 	Total   12 GB   2.9 GB   24%
> 
> Without rebooting the guest I tried to write a small amount of data by dd
>   dd if=/dev/zero of=/mnt/sda1/smallfile bs=1M count=128 but
> but I had to stop it because it was very slow.
> 

Sheepdog will return NO SPACE to QEMU and QEMU actually returns EIO to
the VM. So in your case, your VM got the EIO for dd command. 'dd' looks
slow because it might be busy retry because of EIO and the progress is
halted.

> I tried to poweroff the guest but i hanged up, so I had to kill it.
> After the reboot, it was not possible to mount sda1
>    '/dev/sda1 can't read superblock'.
> I tried then
>    xfs_repair -L /dev/sda1
>    xfs_repair: pwrite64 failed: Input/output error
>    <repeated severl times>
> dmesg is showing I/O errors
> sheep.log
> 	Dec 13 11:59:05 [io 7284] prealloc(303) failed to preallocate space,
> No space left on device
> 	Dec 13 11:59:05 [io 7284] err_to_sderr(129) diskfull, oid=7c2b25000002f1
>         <repeated several times>
> After that I was able to mount sda1.
> I tried to run dd again with small data amount but this time I was not
> even able to stop it.
> I had to kill the guest. Note: before write anything, I removed all
> file from /mnt/sda1.
> I', trying now to check the disk consistency
>   collie vdi list
>     Name        Id    Size    Used  Shared    Creation time   VDI id
> Copies  Tag
>     test         1   10 GB  3.4 GB  0.0 MB 2012-12-13 11:46   7c2b25     2
>   collie vdi check test
>       Failed to read, No object found
> 

Good catch. It is a bug inside block driver of QEMU about sheepdog. I
have sent a fix against it to the QEMU list.

Thanks,
Yuan