Cluster: Host node ZFS redundancy with additional (storage) replication for VMs /CTs - best practice?

[tuathan]

I am looking at the option of configuring a PVE cluster with storage replication (using a ZFS pool across the cluster nodes) so VMs and CTs can be moved to another node if the original host node has a failure.

This is probably a PVE independent question but: when I create the initial ZFS provision on each node (for pooling) I was wondering if I should also select a RAID level (i.e. mirrored for nodes with 2 physical disks), RAIDZ[N] for nodes with additional disks)? Would this be best practice or overkill, given the VM/CT replication (to another physical node on the cluster) occurring on each node?

Thanks

 

[LnxBil]

Depends on what you optimise in your cluster setup: If you want redundancy at any level, I'd go with anything but stipping. Disks fail and it is the only part that fails regularly in a computer, so adding RAID will benefit this a lot. If you want speed, go with mirroring, if you want space, go with raidz.

Normally, I want a cluster that is running all the time, so that failed disks and PSUs are hot-plugged, but everything is still working fine.

 

[tuathan]

Depends on what you optimise in your cluster setup: If you want redundancy at any level, I'd go with anything but stipping. Disks fail and it is the only part that fails regularly in a computer, so adding RAID will benefit this a lot. If you want speed, go with mirroring, if you want space, go with raidz.

Normally, I want a cluster that is running all the time, so that failed disks and PSUs are hot-plugged, but everything is still working fine.

Thanks for your answer.

One of the disadvantages of RAIDZ for me is that you cannot extended the RAIDZ (number of drives used) on a node but with a zfspool across the physical cluster nodes (with storage replication) I think I can just add more physical disks to any node and the zfspool at any time. Is this correct?

 

[LnxBil]

a node but with a zfspool across the physical cluster nodes (with storage replication) I think I can just add more physical disks to any node and the zfspool at any time. Is this correct?

ZFS is local, you cannot strip it across more than one host.

One of the disadvantages of RAIDZ for me is that you cannot extended the RAIDZ (number of drives used) on a node

As with any RAID, just plan ahead and increase the capacity with the same setup as you previously had, e.g. RAID5 over 3 Disks, add another 3 disks. RAID10 with 4 disks, add another 2 disks. With ZFS, it's the same. I've never seen someone expanding a RAID5 with one disk, it makes no sense to me.