ZFS zpool & dataset for pve-zsync

[basstscho]

Hello,

I'm new to proxmox and we'd like to migrate from ESXI to Proxmox. Now I do some trials of different scenarios to get a better understanding on test systems. Currently I want to test pve-zsync and got confused with my whole ZFS setup.

What I did on the test-server:
- There is a zpool called "local_10011" (created on the "host")
- On the Datacenter I created a dataset "local-10011" and a dataset "replication"
- Then I installed a VM on the dataset "local-10011" (ID 101)
- Now I wanted to pve-zsync a VM from another host and started to struggle:

I thought that this will do the trick: pve-zsync sync --source 10.10.100.100:109 --dest local_10011/replication --verbose --maxsnap 5 --name pve06-109, but it shows the error: cannot open 'local_10011/replication': dataset does not exist

But it works with pve-zsync sync --source 10.10.100.100:109 --dest local_10011 --verbose --maxsnap 5 --name pve06-109

A zfs list also only shows the pool and no datasets

local_10011                 170G  1.59T    96K  /local_10011
local_10011/vm-101-disk-0   752K  1.59T   592K  -
local_10011/vm-101-disk-1   170G  1.59T   149G  -

From my understanding I was expecting, that the vm is lolcated in "local_10011/local-10011/vm-101*"

Did I configured something wrong or do I have a completly wrong understanding of the ZFS-setup? Looking to https://pve.proxmox.com/wiki/Storage:_ZFS and https://pve.proxmox.com/wiki/PVE-zsync I thougt I'm rigth. Could it be, that the creating of a dataset in the "Datacenter" is not really a dataset?

Thank you and best,
Johannes

 

[waltar]

Yes, you last assumption is right as in proxmox you usually use zvol and not datasets which is seen in your "code window" as it's not mounted.
But nevertheless you could manually create zfs dataset and let define that in proxmox storage as dir´, just tested and uses qcow2, works fine.
Did 2 same alma8 vm's for comparing, looks nearly all the same in I/O, just when you force really hard sync writes in the vm it's 2,5x faster (3 vs 7,5) with dataset. Could seen in "vm summeries" zvol writes max300MB/s and 80% cpu with zvol and max400MB/s while using max70% cpu with vm on dataset - but this is not an application test - nor it's the default as proxmox expect this ...

 

[basstscho]

Ah ok, so I got the main idea but the GUI didn't do what I was expected. Got it working with creating the dataset by shell: zfs create local_10011/replication

Interesting to know, that there is a performance difference. Did you configured the dataset with different parameters?

 

[waltar]

No, cmd zfs create with no options, new storage dir ... as it was just a test to let pve install in a dataset and see what could be the differences in functionality (formats, pve snapshots, ...) and last few tests to zvol and dataset "under the vm's" and other from inside the vm's, all just for fun