I did an upgrade from Unraid 7.0.1 to 7.1.1. Before the upgrade, everything worked fine for weeks. The upgrade went without any problems, after rebooting the server I ran the array and found that one ZFS pool was broken: "Unmountable: wrong or no filesystem".

 

 

I run from shell:

 

 

But it didn't help anything. What should I do?  

On this pool I have the configuration of all the applications (appdata), the whole server now is dead, I can't run anything. 😞

Please help.

 

 

------------ after some time

 

After execute import the pool:

 

But in IU the pool is still unmountable:

 

 

mordor-diagnostics-20250510-1232.zip

Edited May 10, 20251 yr by jaclas

There's an issue with one of the partitions, post the output from:

 

wipefs /dev/nvme1n1p1

and

fdisk -l /dev/nvme1n1

 

Note that the wipefs command as written won't wipe anything.

@JorgeB 

 

root@mordor:~# wipefs /dev/nvme1n1p1
DEVICE    OFFSET       TYPE       UUID                 LABEL
nvme1n1p1 0x4000       zfs_member 11242959326477942528 ssdraid
nvme1n1p1 0x44000      zfs_member 11242959326477942528 ssdraid
nvme1n1p1 0xe8e0c04000 zfs_member 11242959326477942528 ssdraid
nvme1n1p1 0xe8e0c44000 zfs_member 11242959326477942528 ssdraid

root@mordor:~# fdisk -l /dev/nvme1n1
Disk /dev/nvme1n1: 931.51 GiB, 1000204886016 bytes, 1953525168 sectors
Disk model: Samsung SSD 980 PRO 1TB                  
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disklabel type: dos
Disk identifier: 0x00000000

Device         Boot Start        End    Sectors   Size Id Type
/dev/nvme1n1p1       2048 1953525167 1953523120 931.5G 83 Linux

 

Edited May 10, 20251 yr by jaclas

That looks normal, the easiest solution for now, and since the pool is a mirror, would be to reimport the pool degraded, then wipe the other device and add it back to recreate the mirror, if you want to do that, see below, other option is to downgrade to 7.0.1 for now.

 

Stop the array and offline the device with the "bad partition" by typing 

echo 1 > /sys/block/nvme1n1/device/device/remove

 

Wait a couple of seconds and refresh the GUI, that device should now be missing, start the array with the remaining pool device and it should import the pool degraded

 

If all looks normal, disable array auto-start if enabled, reboot to bring the old device back, check that it's still nvme1n1, since device IDs can change with a reboot, if it changed, adjust the command, then wipe it with:

blkdiscard -f /dev/nvme1n1

 

Now add assign it back to the pool, start array to begin resilver, that should do it.

 

 

@JorgeB

Thank you for your advice, I will look into it in a while and follow these instructions.

But I have a question in connection with this:
- why did this situation arise at all, and why doesn't Unraid solve this on its own? When migrating from 6 to 7 there was no such problem. Is this a bug of Unraid and the next version will fix it?

 

 

--- after a few minutes

 

After remove drive (by echo) the device disapear but the array can't be run without it:

 

Edited May 10, 20251 yr by jaclas

20 minutes ago, jaclas said:

why did this situation arise at all, and why doesn't Unraid solve this on its own? When migrating from 6 to 7 there was no such problem. Is this a bug of Unraid and the next version will fix it?

There's a change in 7.1 on how it detects partitions, but the problem is not Unraid, the partition is not showing all the details as it should with lsblk, and this is a Linux utility, not Unraid, what LT may be able to do is to try and ignore that for partitions that are not working as they should.

 

22 minutes ago, jaclas said:

After remove drive (by echo) the device disapear but the array can't be run without it:

Check

 

1 hour ago, JorgeB said:

That looks normal, the easiest solution for now, and since the pool is a mirror, would be to reimport the pool degraded, then wipe the other device and add it back to recreate the mirror, if you want to do that, see below, other option is to downgrade to 7.0.1 for now.

 

Stop the array and offline the device with the "bad partition" by typing 

echo 1 > /sys/block/nvme1n1/device/device/remove

 

Wait a couple of seconds and refresh the GUI, that device should now be missing, start the array with the remaining pool device and it should import the pool degraded

 

 

 

After start Array the pool is not degraded, it's look ok, but has only one disk now:

 

 

Is this a good symptom?

 

-- after some time

 

It's getting worse... strange things are happening. Suddenly the disk is too small!

 

 

How is this possible? The two drives had been stacked in an array for many months.

 

Edited May 10, 20251 yr by jaclas

Sorry, I missed one device was smaller than the other, they were both showing as 1TB on the screenshot, my suggestion would be to create a new zfs single pool with the smaller device, copy everything from the other pool to that one, then remove the larger pool and that add device to the smaller one to create a mirror.

 

 

 

 

1 hour ago, JorgeB said:

Sorry, I missed one device was smaller than the other, they were both showing as 1TB on the screenshot, my suggestion would be to create a new zfs single pool with the smaller device, copy everything from the other pool to that one, then remove the larger pool and that add device to the smaller one to create a mirror.

 

A little uncool 😕

Ok, but I have to get out of it. I have an idea and you tell me if it makes sense or what to improve.
I will create a second pool of ssdraid2 on this second disk, which is now unused. I will recreate the dataset and directory structure on it, copy the data from the first pool. Then I will delete the ssdraid pool, clean this disk and then try to add it to the ssdraid2 pool (extending the pool to the mirror).
Does this make sense? What should I pay attention to?

17 hours ago, jaclas said:

Does this make sense?

That should do it, if you are using datasets/snaphots for all shares, you can also use zfs replication to easily transfer everything, including any snapshots