[sheepdog-users] resize of sheepdog device

[Liu Yuan]

On 12/12/2012 12:19 AM, Valerio Pachera wrote:
> Ok, this is something that should not happen, but I wanted to test it anyway:
> 
> Situation:
>   3 nodes.
>   1 partition of 1G for each node.
> 
> I was curios to see what was going to happen to a guest with a large
> virtual space (10G) when it reaches the end of phisical space (2.8G):
> 
>   # collie node info
>   Id      Size    Used    Use%
>    0      920 MB  916 MB   99%
>    1      920 MB  916 MB   99%
>    2      978 MB  976 MB   99%
>   Total   2.8 GB  2.7 GB   99%
> 
> Note: the guest was able to write 923M.
> 
> Sheepdog doesn't seem to give any warning in the logs.
> The guest start creying wih I/O errors as I was aspecting.
> 

Other tools can use 'collie node info' to get the high watermark
warning. Warning from logfile is not necessary from my perspective.

> But I wish to focus on another thing:
>   I stoped the cluster
>   Resized the partitions to 10G
>   restarted the cluster
> 
> *I still get a cluster of 2.8G!*
> 
> # collie node info
>   Id      Size    Used    Use%
>    0      920 MB  916 MB   99%
>    1      920 MB  916 MB   99%
>    2      978 MB  976 MB   99%
>   Total   2.8 GB  2.7 GB   99%
> 
> Do I have to re-run collie cluster format?

For now, yes. I'll look if live resize of node is possible.

> Am I going to loose vdi disks?
> 

Yes, if you format.
You can mitigate out of space problem by adding more nodes to increase
the cluster storage space.

Thanks,
Yuan