[sheepdog-users] vdi and dynamic allocation

[Liu Yuan]

On 12/14/2012 04:13 PM, Valerio Pachera wrote:
> 2012/12/14 Liu Yuan <namei.unix at gmail.com>:
>> Yes, it is possible for us to support shrink the image offline. If this
>> is desired, we can implement it.
> 
> I think it may be a good way to optimize the cluster.
> Think of this scenario: you sell a virtual private server (vps) with X
> G of vdi space.
> The vps administrator do something wrong (like too many backups) that
> will use lot's of space.
> Later he will remove the unnecessary data.
> We have a lot of space allocated by the vdi of this vps that can't be
> used by other vdis in the cluster.
> 
> The word 'shrink' may be misunderstood like 'to reduce the virtual
> space of a disk'.
> So it may be better use something else like 'optimize' or 'free_allocated"
> 
> There may be two ways to implement the shrinking:
> - explicit command like 'collie vdi optimize <vdiname>
> - or as an option of the cluster.
> 
> In the first case, the "clean up" is done manually by the cluster
> administrator, disk by disk.
> 
> In the second case, sheep dog it self is going to take care of the
> cleaning up of all the vdis.
> The I/O of removing unused chunks may be done with a very low priority
> so to not impact the cluster performance.
> I do not know if the 'optimize' option may be given when we
> create/format the cluster, or later by a command.
> That probably depends on how sheepdog works internally.
> 

This is similar to linux file punch holes to safe disk space.

Okay, "optimize" is indeed a better wording. let's firstly implement a
mechanism to optimize the underlying storage allocation via 'collie vdi
optimize'. This feature is in the plan.

Thanks,
Yuan