[PVE-User] Some thought about zfs based configurations

[Pongrácz István]

Hi,

>> I have a proxmox node, which is using zfs in such a way I described (except 
>> btrfs) and that node is up an running for more than 657 days now.
>> 
> uptime for a server <=> security or stability. Long uptimes also means
> a lot of active kernel bugs fixed in more resent kernels.

This is an example about how stable zfs is. It is not about security. Of course, you are right in general.


> For your claim of been able to recover a node from crash quickly as a
> consequence of using ZFS this can be easily achieved with the current
> default install of Proxmox when sticking to these three rules:
> 
> 1) A node is simply used as runtime for VM's/CT's.
> 2) All virtual disks for VM's/CT's resides on shared storage.
> 3) Backups are stored on shared storage as well.
> 
> Then to recover a node is simply a matter of reinstall from backup.

For higher budget, it can be true. For lower budget (1 or 2 servers + backup) maybe not.
Anyway, if a compute node crashes, its filesystem could be left inconsistent, which means, even the shared storage did not crash and up and running, when you start a new node and access to the same shared virtual disk, the newly started system will fsck it.
If you have a big storage, it will take a lot of time. Over the network on 1G takes much more.
If you use btrfs on the VM, it will be very quick.
Of course, zfs on zfs also a possibility, but needs tuning to avoid double cache and waste of resources.

Bye,
István