ZFS subvol xattr "on" vs "sa"

[mailinglists]

Hi,

some time ago I migrated LXCs from PM 4.4 to 5.*

Recently I noticed that zfs subvols, that were transferred via pve-zsync do have xattr set to on, while newly created do have it set to sa.

I understand it has to do with saving attributes in such way that "sa" is faster than "on", right?
Can I just do zfs set xattr=sa /pat/to/zfsSubvol on all existing and running LXC disks?

For ZVOLs (KVM VMs), xattr, should not be set, correct?

 

[mailinglists]

Can anyone answer:
"Can I just do zfs set xattr=sa /pat/to/zfsSubvol on all existing and running LXC disks without causing problems for LXCs?"

 

[wolfgang]

Hi
See
https://pve.proxmox.com/wiki/ZFS:_Tips_and_Tricks#LXC_with_ACL_on_ZFS

 

[mailinglists]

Thank you for your reply.
I saw that. But is it safe to do while LXC is running?

 

[wolfgang]

Safe it is but I'm not sure if the data are moved.

 

[mailinglists]

Once again thank you for your reply.

What do you mean with "but I'm not sure if the data are moved"? Can you please elaborate a little. What kind of move?

I already did move data with pve-zsync a few months ago, when upgrading form 4.4, and have just recently noticed that xattr is on instead of sa.
None of the customers complained. Is it safe not to set it to sa and dnodesize=auto while LXCs are running?

 

[wolfgang]

You have two real possible values for xattr.
SA means you save the xattr in the inode
Dir == ON means you have a hidden dir where the xattr are stored (very slow).

If you change the xattr data will not be copied and can't be used anymore.
If you have a container which uses xattr you must use backup and restore to keep the xattrs.

 

[mailinglists]

Oh, now i started to get it. I somehow i just asumed xattr == posix ACL. Which is not, and my LXC instances do not use it according to their fstab.
Thank you for your time @wolfgang .