Re: Disk too full

[Sven Gaerner <sgaerner@xxxxxxx>]

On Wed, May 22, 2013 at 11:57:57AM +0200, Michael Neumann wrote:
> Am 22.05.2013 10:32, schrieb Konrad Neuwirth:
> >Dear reader,
> >
> >there is probably something very, very obvious that I am missing, but one server of ours is claiming so much more disk space is filled than current use warrants and I don't see how I can free that again.
> >
> >It's a drive that contains five pfs; four master and one slave.
> >
> >I've reblocked the pfs, but still there's too much overhead.
> >
> >How can I investigate where the blocks went, and particularly: How can I free them?
> >
> >Thank you,
> >  Konrad
> 
> Hammer takes snapshots and keeps them for a while. You can configure
> this using "hammer config pfs".
> If you then run "hammer cleanup", this history retention policy is
> applied (i.e. snapshots are taken,
> old snapshots are freed, all non-snapshotted data is purged, and the
> filesystems in question are
> reblocked...).
> 
> You probably either don't run "hammer cleanup" regularily, or your
> history retention policy keeps too
> many snapshots (I think the default is to keep 60 days worth of
> snapshot with a 1-day granularity). If
> this is the case, either "hammer config" and use a lower value (e.g.
> "1d 30d" to keep 30 days, or "2d 30d"
> to only keep every second day...).
> 
> "hammer reblock" doesn't help as it only brings the blocks in-order
> again, but does not free space.
> "hammer prune" is your friend.
> 
> If you want to release all space taken by snapshots, use "hammer
> prune-everything filesystem" but with CARE!
> All your historical data will be LOST!
> 
> Regards,
> 
>   Michael

You should also keep in mind that all of these daily jobs have a
maximum runtime. Depending on the size of your PFS this runtime
might be too short. Pruning is by default only run for 5 minutes.

Regards,
Sven