Break ZFS temporarily, safely

[neuron]

I have a Proxmox node with 2x 240GB ZFS mirrored drives for the system itself, and 6x 480GB SSDs in a RAIDz1.

The RAIDz1 array has faulty SSDs that are wearing out too fast and right now I'm dealing with corruption that's breaking partition tables on some of the VMs. This particularly seems to happen around the freeze/thaw of backups to PBS for some reason. I'm needing to throw in a single drive to move the VMs off the faulty array.

Is it safe to just pull one of the drives out of the RAIDz1 array, pop in a new drive, format it as an LVM and then start moving the VMs over? The RAIDz1 will be Degraded status but still working. I do have PBS backups offsite which is a slow link and that's my last resort if it doesn't work.

My concern is ZFS might grab the new drive and start rebuilding the array automatically. I don't want that to happen.

 

[Dunuin]

ZFS shouldn't replace failed disks on its own if there is no hot spare disk defined.

As soon you remove one disk of that raidz1 it can't handle any errors any longer. If one of that disks is already causing errors I would only remove this failing drive.

 

[neuron]

Thanks for the reassurance - to clarify, the RAIDz1 is currently healthy and scrub never finds errors. The problem we've been having is the SSDs can't seem to keep up with the performance and we have had to recover partitions inside the VM three times now. We've just been having lots of problems with the SSDs themselves and simply want to migrate the VMs off them. The server is physically full with 6 hot swap disks in place to make the RAIDz1. So we need to free up one of the drive bays in order to facilitate the moves of the VM off the RAIDz1 onto a single SSD until we can get the array itself replaced.

I just mostly needed reassurance that ZFS won't go rogue on me and grab the new SSD to rebuild the array automatically.

 

[UdoB]

The server is physically full

Personally I would avoid to remove the only existing parity bit from that RaidZ Pool. To circumvent the lack of a free port I would use an USB Adapter. (And I say this although I have very bad experiences with USB drives which I use for a second-level PBS.)
For my VMs the new destination should have some redundancy also. Your single LVM-disk approach seems to lack this feature.

Another way for copying that content is over-the-network of course.

Maybe I did not understand your goal...

 

[manuelkamp]

I would rather add a PCIe controller card für HBA and add the additional drive there. Never rip off a disk of the raidz1 if it is still working, as this could destroy your pool if another disk has an error and you need to pull backups from your pbs