1 minute ago, poeterdebier said:

That is something that you recommend to do before replacing the disk1 entirely?

Yes, first attempt to repair the file system; rebuilding the disk without repairing it will rebuild the same corrupt filesystem.

tower-diagnostics-20251231-1622.zipSo I did a repair on disk1 through the webUI (Check Filesystem Status section). I had to 'zero log' due to 'dirty log detected'. After that I got a file system corruption fixed.

So current disk1 is back online. I did a scrub of the cache pool and that is still the same (2056 uncorrectable errors). I also reset the pool device status (I assume that this is what you mean with 'reset pool stats'.

At the moment I do not know what to do with the uncorrectable errors. The scrub should fix this if there is a correct copy available (I guess there isn't right now). I looked into the syslog to attempt to see what files are corrupt to remove/replace them but the only 'corrupt' message I can find is:

BTRFS error (device nvme0n1p1): bdev /dev/nvme0n1p1 errs: wr 0, rd 0, flush 0, corrupt 7368, gen 4

Do you have an idea how to identify the files that are corrupt? To get this uncorrectable error count back to zero?

Edited December 31, 2025Dec 31 by poeterdebier
added diagnostics

Corrupt file(s) should be listed in the syslog during/after the scrub, for example:

Dec 31 15:14:50 Tower kernel: BTRFS error (device nvme0n1p1): unable to fixup (regular) error at logical 4659085312 on dev /dev/nvme1n1p1 physical 4638113792

Dec 31 15:14:50 Tower kernel: BTRFS warning (device nvme0n1p1): checksum error at logical 4659085312 on dev /dev/nvme1n1p1, physical 4638113792, root 5, inode 523165, offset 7588679680, length 4096, links 1 (path: isos/homeassistant/haos_ova-7.1.qcow2)

In this case path: isos/homeassistant/haos_ova-7.1.qcow2

Completely missed that one (the path part). Thanks. Removed already part and the errors went down. Tomorrow continue.

Thanks for the help so far, to be continued.

Happy New Year 🎆

Hi Jorge,

I decided to focus on disk1. So I did a file system check of disk1. I got an 'dirty log error'. I went ahead with 'zero log' and this fixed the file corruption on disk1. Started the array after that and disk1 mounted without issues.

I did another scrub of the cache pool but this kept giving 2056 uncorrectable errors. As mentioned by you I started replacing/deleting some of the files that the syslog mentioned and this removed some of the uncorrectable errors in the cache pool. Not all unfortunately, 664 remaining. Also, I cannot identify them the same way as the isos were (with a mentioned file name). So regarding the 664 uncorrectable errors I am stuck at the moment.

I was considering to repair the cache pool in the 'check filesystem status' but the help section was really specific with mentioning that this only needed to be done on advise of a community expert. So I only did the readonly check (but forgot to take a screenshot).

(did a parity check with correction after disk1 was reinstated. Many corrections. Did a parity check again after that without correction checked. Finding 0 errors.)

tower-diagnostics-20260102-2227.zip

Edited January 2Jan 2 by poeterdebier
forgot to mention the parity checks that were done.

If the remaining errors don't list files, they could be in the metadata; in that case, I would recommend backing up and recreating the pool

recreating the pool could be done by:

• moving all data to the array

• removing both cache from array / format

• adding both nvme back as cache

or am I then recreating errors?

Yes, the moved data should be all OK, assuming the issue is no longer occurring.

Hi Jorge,

removed all data from cache (moved to array). Removed and formatted the nvme's and recreated a pool. Scrub gives 0 errors and pool device stats are all at 0. Server operational without issues. Will closely monitor the server oncoming time. Also learned some things and will have a bit different approach on backing up files/settings in the future.

I really appreciated the help and support.

regards Piet

(not sure if you want to close this post)

Edited January 4Jan 4 by poeterdebier