Proxmox ZFS with Ceph?

[stevefan1999]

Ceph does have a ZFS module to use it as backend.
Wonder if it can be cherrypicked or ported to Linux so I don't have to use the ZVOL method (that essentially means running LVM on top of a ZVOL which is quite ridiculous).
I think ZFS and Ceph will have good chemistry, consider ZFS is one of the best FS out there (bit rot protection, incremental snapshot, L2ARC all the goodies) and Ceph can leverage it to have safe, performant and hyperconverged file store.

 

[dcsapak]

why exactly would you want this? ceph already does everything you'd expect from zfs (snapshots/checksums/redunancy/etc) but using whole disks (with lvm) will be much faster than putting the (cow) data on a zfs (which is again cow)

also:

These APIs are not meant to be public, but are documented so that it is clear what the tool is doing behind the scenes. Do not alter any of these values.

so this is not intended to be used yet...

 

[stevefan1999]

why exactly would you want this? ceph already does everything you'd expect from zfs (snapshots/checksums/redunancy/etc) but using whole disks (with lvm) will be much faster than putting the (cow) data on a zfs (which is again cow)

also:


so this is not intended to be used yet...

I mean Ceph could also offload these features to ZFS, leaving the distribution layer to be taken care of, that means the datastore and COW (as native ZFS filesystem), snapshot and synchronization (SLOG) is all done by ZFS, and Ceph just need to send metadata around clusters.

ZFS itself is nowt cluster-aware I know, so that's where Ceph could come in.

 

[dcsapak]

I mean Ceph could also offload these features to ZFS, leaving the distribution layer to be taken care of, that means the datastore and COW (as native ZFS filesystem), snapshot and synchronization (SLOG) is all done by ZFS, and Ceph just need to send metadata around clusters.

that is absolutely not how ceph works and would mean a complete rewrite of ceph

but basically what you describe is gluster on zfs afaics ?

ZFS itself is now cluster-aware I know, so that's where Ceph could come in.

where did you read that zfs is cluster aware?

 

[stevefan1999]

where did you read that zfs is cluster aware?

mistyped sorry