[SOLVED] ZFS pool almost full despite reducing zvol size

[jpaugh]

I have a PVE host that is currently reporting that "local-zfs" is nearly full. There is only one guest with one virtual disk stored in this pool; the virtual disk was originally 50TB and I have since reduced it to 43TB using

zfs set volsize=43T rpool/data/vm-301-disk-0

. Anyone experience this before, or can share some advice on how to fix this?

 

[justinclift]

Ahhh. It's probably worth checking if there are any snapshots of the rpool ZFS pool hanging around.

Try running zfs list -t snapshot -r rpool and seeing if it returns anything.

If there are snapsnots, that'll probably be why you're not seeing the newly freed up space be available.

To destroy a snapshot, then (one at a time) run zfs destroy NAME_OF_SNAPSHOT. ie zfs destroy rpool/data/something@something_else

 

[jpaugh]

Ahhh. It's probably worth checking if there are any snapshots of the rpool ZFS pool hanging around.

Try running zfs list -t snapshot -r rpool and seeing if it returns anything.

If there are snapsnots, that'll probably be why you're not seeing the newly freed up space be available.

To destroy a snapshot, then (one at a time) run zfs destroy NAME_OF_SNAPSHOT. ie zfs destroy rpool/data/something@something_else

Thank you for reply, there are no snapshots listed with the above command (output attached).

 

[justinclift]

Damn. In that case I have no idea, sorry dude.

 

[jpaugh]

Damn. In that case I have no idea, sorry dude.

No problem, I appreciate you taking the time to help. This issue has been a real head scratcher, and I have scoured multiple forums trying different things to diagnose it

 

[leesteken]

A 4-drive RAIDz1 only has about 50% usable space because of the volblocksize mismatch. As discussed before: RAIDz1 is not like hardware RAID5. There is a lot of padding (and write multiplication, which it also makes it slow for running VMs) and ZFS does not show the expected usable size (when it's just created) which catches users by surprise.

 

[jpaugh]

A 4-drive RAIDz1 only has about 50% usable space because of the volblocksize mismatch. As discussed before: RAIDz1 is not like hardware RAID5. There is a lot of padding (and write multiplication, which it also makes it slow for running VMs) and ZFS does not show the expected usable size (when it's just created) which catches users by surprise.

I checked out the OpenZFS documentation to reaffirm what you are saying and I think I understand it better now, and will likely use this chart over the PVE GUI in the future: https://openzfs.github.io/openzfs-docs/Basic Concepts/RAIDZ.html.

I wrongfully assumed that going from 80TB to 60TB was the total cost of the parity because one disk worth of capacity made sense from a RAID5 perspective. However, as you pointed out and I checked with the OpenZFS documentation this is not the case, and my fears of write-amplification are confirmed.

Thank you for your time and input, I appreciate having a second set of eyes look at this problem for me!