[sheepdog-users] Testing snapshot-object-reclaim

[Hitoshi Mitake]

At Tue, 11 Mar 2014 10:36:29 +0100,
Valerio Pachera wrote:
> 
> [1  <text/plain; UTF-8 (7bit)>]
> 2014-03-10 4:01 GMT+01:00 Liu Yuan <namei.unix at gmail.com>:
> > Just want to check when remove all the snapshots include base, is all allocated
> > space released? This is the bottom line that at least we can reclaim all the
> > space.
> 
> dog vdi list
>   Name        Id    Size    Used  Shared    Creation time   VDI id  Copies  Tag
>   test         0   10 GB   72 MB  6.2 GB 2014-03-07 11:35   7c2b2a    2:1
> 
> dog node list
>   Id   Host:Port         V-Nodes       Zone
>    0   192.168.10.4:7000        75   67807424
>    1   192.168.10.5:7000        147   84584640
>    2   192.168.10.6:7000        162  101361856
> 
> dog node info
> Id      Size    Used    Avail   Use%
>  0      233 GB  5.9 GB  227 GB    2%
>  1      452 GB  5.9 GB  446 GB    1%
>  2      500 GB  5.9 GB  494 GB    1%
> Total   1.2 TB  18 GB   1.1 TB    1%
> Total virtual image size        10 GB
> 
> dog vdi delete test
> 
> dog node info
> Id      Size    Used    Avail   Use%
>  0      233 GB  2.3 GB  230 GB    0%
>  1      452 GB  2.3 GB  450 GB    0%
>  2      500 GB  2.3 GB  497 GB    0%
> Total   1.2 TB  6.8 GB  1.2 TB    0%
> Total virtual image size        0.0 MB
> 
> 
> I attach the log (debug level).
> It's size is 11M.
> 
> 2014-03-11 3:08 GMT+01:00 Hitoshi Mitake <mitake.hitoshi at gmail.com>:
> > Could you run sheeps with -l level=debug? I think this problem is
> > related to discard operation, so I want you to check whether the guest
> > OS issued discard requests or not.
> 
> I'm going to update sheep and repeat the test.
> 
> > BTW, which file system is your guest using?
> 
> xfs

I think I could reproduce your problem as a test script:
https://github.com/sheepdog/sheepdog/commit/61e396057539370254cc0219108935d356d4fdd1

The GC algorithm seems to cause a problem in a case of continuously
snapshotted and partially updated VDI. It is not related to erasure
coding, 3 replicated VDI can cause the problem. I'm digging in the
problem.

Thanks,
Hitoshi