delete on snapshot does not free space

delete on snapshot does not free space

[Michael DiDomenico]

this is probably expected behavior, but i'd like to confirm as i'm not
a device-mapper expert

we're using COW snapshots overtop of a readonly block device.  this
seems to be working fine, however, when a person deletes files the
blocks are not freed.  case in point a RO block device with a COW
filled up to 100% as shown in df.  when i deleted some files to free
space the usage never decreased

to create the mount we do

echo "0 <realsize> linear /dev/loop1 0" | dmsetup create roimage-base-real
echo "0 <cowsize> linear /dev/loop2 0" | dmsetup create roimage-snap-cow
echo "0 <realsize> snapshot-origin /dev/mapper/roimage-base-real" |
dmsetup create roimage-base
echo "0 <realsize> snapshot /dev/mapper/roimage-base-real
/dev/mapper/roimage-snap-cow n 1" | dmsetup create roimage-cow

mount /dev/mapper/roimage-cow /roimagemntpnt

this is working as the origin device is Read-Only, but we can change files.

so two questions come up

1 when we delete files, how do we reclaim the space
2 should/can we roll the snapshots to reclaim space?
- if i wanted to create a new snapshot and discard the existing one,
do i basically just to the fourth step again to a different device?  i
haven't found a "clear" how-to on this.  most people just stop after
the fourth line and then tear it all down

ideally what i'm looking to do is create a filesystem where the base
is read-only, but we can make changes.  i'm okay with all those
changes getting lost after a reboot/unmount/etc.  the issue right now
is when df shows 100% usage it seems the only recourse is reboot(aka
clear it all out) and restart.  i'm hoping there's something more
graceful

Re: delete on snapshot does not free space

[Zdenek Kabelac]

Dne 30. 04. 24 v 19:54 Michael DiDomenico napsal(a):
> this is probably expected behavior, but i'd like to confirm as i'm not
> a device-mapper expert
> 
> we're using COW snapshots overtop of a readonly block device.  this
> seems to be working fine, however, when a person deletes files the
> blocks are not freed.  case in point a RO block device with a COW
> filled up to 100% as shown in df.  when i deleted some files to free
> space the usage never decreased
> 
> to create the mount we do
> 
> echo "0 <realsize> linear /dev/loop1 0" | dmsetup create roimage-base-real
> echo "0 <cowsize> linear /dev/loop2 0" | dmsetup create roimage-snap-cow
> echo "0 <realsize> snapshot-origin /dev/mapper/roimage-base-real" |
> dmsetup create roimage-base
> echo "0 <realsize> snapshot /dev/mapper/roimage-base-real
> /dev/mapper/roimage-snap-cow n 1" | dmsetup create roimage-cow
> 
> mount /dev/mapper/roimage-cow /roimagemntpnt
> 
> this is working as the origin device is Read-Only, but we can change files.
> 
> so two questions come up
> 
> 1 when we delete files, how do we reclaim the space
> 2 should/can we roll the snapshots to reclaim space?

Please consider using  thin-pools - this should be giving you the features you 
are likely looking for.

Note - you can use thin-pool to  hold only snapshots  and use your read-only 
volume as the external origin for thin volume.

Regards

Zdenek