Re: dm thin: How to shrink think pool device?

[Maxim Patlasov <mpatlasov@parallels.com>]

On 03/18/2015 05:51 PM, Mike Snitzer wrote:
> On Wed, Mar 18 2015 at  7:46pm -0400,
> Maxim Patlasov <mpatlasov@parallels.com> wrote:
>
>> Hi,
>>
>> It seems that a "trim" pool message was planned long time ago, but never
>> implemented. Could someone update about current status of that? The
>> question came from the following use-case:
>>
>> Suppose we have a thin pool on top of LVM volume group. The group comprises
>> of several storage drives. Now I need to (physically) remove one of them.
>> The intention is to use pvmove and vgreduce, but obviously thin pool must
>> be requested to shrink itself at first.
> We only support growing the pool.  If a drive used by the pool goes
> missing then thin_repair will need to fix things.

Do you think a feature allowing to shrink pool data device on-line 
smoothly will be useful (w.r.t all code complications needed to relocate 
data blocks inside dm-thin while handling i/o at the same time)?

>   
>> Btw, Documentation/device-mapper/thin-provisioning.txt still has a leftover
>> about it:
>>
>>> If you wish to reduce the size of your thin device and potentially
>>> regain some space then send the 'trim' message to the pool.
> I'll remove that stale documentation (commit 5ec02084f6 should've done
> that).
>
> The 'trim' message was never about reducing pool space.  The usage for
> the trim message (sent to the pool device) was:
>   trim <dev_id> <new_size_in_sectors>
>
> It was to reduce the size of the thin volume; don't think it was ever
> wired up because it wasn't needed.

Hm... "dmset load <thin-volume>" seems to do exactly that. But, yes, it 
has nothing to do with reclaiming storage space back to the system.

Thanks,
Maxim