I have a ZFS pool that I made on proxmox. I noticed an error today. I think the issue is the drives got renamed at some point and how its confused. I have 5 NVME drives in total. 4 are supposed to be on the ZFS array (CT1000s) and the 5th samsung drive is the system/proxmox install drive not part of ZFS. Looks like the numering got changed and now the drive that used to be in the array labeled nvme1n1p1 is actually the samsung drive and the drive that is supposed to be in the array is now called nvme0n1.
root@pve:~# zpool status
pool: zfspool1
state: DEGRADED
status: One or more devices could not be used because the label is missing or
invalid. Sufficient replicas exist for the pool to continue
functioning in a degraded state.
action: Replace the device using 'zpool replace'.
see: https://openzfs.github.io/openzfs-docs/msg/ZFS-8000-4J
scan: scrub repaired 0B in 00:07:38 with 0 errors on Sun Oct 13 00:31:39 2024
config:
NAME STATE READ WRITE CKSUM
zfspool1 DEGRADED 0 0 0
raidz1-0 DEGRADED 0 0 0
7987823070380178441 UNAVAIL 0 0 0 was /dev/nvme1n1p1
nvme2n1p1 ONLINE 0 0 0
nvme3n1p1 ONLINE 0 0 0
nvme4n1p1 ONLINE 0 0 0
errors: No known data errors
Looking at the devices:
nvme list
Node Generic SN Model Namespace Usage Format FW Rev
--------------------- --------------------- -------------------- ---------------------------------------- --------- -------------------------- ---------------- --------
/dev/nvme4n1 /dev/ng4n1 193xx6A CT1000P1SSD8 1 1.00 TB / 1.00 TB 512 B + 0 B P3CR013
/dev/nvme3n1 /dev/ng3n1 1938xxFF CT1000P1SSD8 1 1.00 TB / 1.00 TB 512 B + 0 B P3CR013
/dev/nvme2n1 /dev/ng2n1 192xx10 CT1000P1SSD8 1 1.00 TB / 1.00 TB 512 B + 0 B P3CR010
/dev/nvme1n1 /dev/ng1n1 S5xx3L Samsung SSD 970 EVO Plus 1TB 1 289.03 GB / 1.00 TB 512 B + 0 B 2B2QEXM7
/dev/nvme0n1 /dev/ng0n1 19xxD6 CT1000P1SSD8 1 1.00 TB / 1.00 TB 512 B + 0 B P3CR013
Trying to use the zpool replace command gives this error:
root@pve:~# zpool replace zfspool1 7987823070380178441 nvme0n1p1
invalid vdev specification
use '-f' to override the following errors:
/dev/nvme0n1p1 is part of active pool 'zfspool1'
where it thinks 0n1 is still part of the array even though the zpool status command shows that its not.
Can anyone shed some light on what is going on here. I don’t want to mess with it too much since it does work right now and I’d rather not start again from scratch (backups).
I used smartctl -a /dev/nvme0n1 on all the drives and there don’t appear to be any smart errors, so all the drives seem to be working well.
Any idea on how I can fix the array?
Weird. I suspect your disk is dead, but in case it’s not, I’d mark that disk for replacement, then “replace” it at the software level and allow resilver to see what happens.
No idea how to do this in proxmox or the commands but I know how to do it in truenas with the GUI.
I don’t know anything about ZFS, but in the future you might want to address them by /dev/disks/by-uuid/… or by-id and not by /dev/nvme…
That is definitely true of zfs as well. In fact I have never seen a guide which suggests anything other than using the names found under /dev/disk/by-id/ or /dev/disk/by-id/uuid and that is to prevent this very problem. If the proper convention is used then you can plug the drives in through any available interface, in any order, and zfs will easily re-assemble the pool at boot.
So now this begs the question… is proxmox using some insane configuration to create drive clusters using the name they happen to boot up with???
Is there a way to change this on an existing zpool?
Generally, you just need to export the pool with
zpool export zfspool1, then import again withzpool import -d /dev/disk/by-id zfspool1.I believe it should stick after that.
Whether that will apply in its current degrated state I couldn’t say.
OP – if your array is in good condition (and it looks like it is) you have an option to replace drives one by one, but this will take some time (probably over a period of days). The idea is to remove a disk from the pool by its old name, then re-add the disk under the corrected name, wait for the pool to rebuild, then do the process again with the next drive. Double-check, but I think this is the proper procedure…
zpool offline poolname /dev/nvme1n1p1zpool replace poolname /dev/nvme1n1p1 /dev/disk/by-id/drivenameCheck zpool status to confirm when the drive is done rebuilding under the new name, then move on to the next drive. This is the process I use when replacing a failed drive in a pool, and since that one drive is technically in a failed state right now, this same process should work for you to transfer over to the safe names. Keep in mind that this will probably put a lot of strain on your drives since the contents have to be rebuilt (although there is a small possibility zfs may recognize the drive contents and just start working immediately?), so be prepared in case a drive does actually fail during the process.
When you say “the drives were renamed”, do you mean you renamed them while the array was online? That sounds like what this means.
In that case, you can find out which drive is the problem, clear it, and repair the array. Should be pretty quick.
I didn’t rename them. I suspect it happened during a reboot or maybe a bios update that I may have done last month.
How do I clear or repair it?
The device names and aliases in /dev don’t just simply change between reboots. Something else happened here.
What are the path or IDs of the drives that are in there now under /dev/nvme*?
Are you sure about that? Ever hear about this supposed predictable network names in recent linux versions? Yeah those can change too. I was trying to set up a new firewall with two internal NICs plus a 4-port card, and they kept moving around. I finally figured out that if I cold-booted the NICs would come up in one order, and if I warm-booted they would come up in a completely different order (like the ports on the card would reverse which order they were detected). This was completely the fault of systemd because when I installed an older linux and used udev to map the ports, it worked exactly as predicted. These days I trust nothing.