zpool import -T hangs

[Wondertan]

Type	Version/Name
Distribution Name	Arch
Kernel Version	5.14.15-arch1-1
Architecture	amd64
OpenZFS Version	zfs-kmod-2.1.1-1

Hello everyone. I have two FSs mounted on:

• /dir
• /dir/subdir

I had to migrate /dir to another FS and execute mv /dir <new_path>. Unfortunately, that also touched /dir/subdir with terabytes of data. Once I realized that the /subdir is getting moved as well I immediately interrupted the mv. However, since /subdir contains more subdirectories, critical files there were deleted during the move due to the nature of mv.

To recover deleted files. I looked into zpool history and found valid txg to execute zpool import -T. Unfortunately, it hangs like for an hour already with no symptoms of life.

Pls note:

• I don't have snapshots
• It is impossible to get data from the place where it was moved.
• Nothing was written to the pool under /dir/subdir after the move, so I hope it is possible to recover deleted data.
• The moved data is absolutely critical

Any ideas on why it may hang and maybe any tips on resolving the issue?

[rincebrain]

Using -T/-X/... by default walks every single referenced metadata+data block before importing. This is rather time consuming, and I don't think it uses anything like the sequential scrub logic to reorder things, so it's going to be similar to the runtime of the pre-sequential scrub code.

You can adjust this behavior with the tunables spa_load_verify_data (which can turn off reading all data blocks) and spa_load_verify_metadata (which will turn off reading all metadata or data first) - of course, you can run into inconsistent things on-disk if you don't verify them before importing, so it's a tradeoff.

I would pretty strongly recommend using -o readonly=on when playing with -T/-F/-X, though if you've already done it without doing that, you may have already done some damage...well, we'll see.

As the man page says, "WARNING: This option can be extremely hazardous to the health of your pool and should only be used as a last resort.", so be aware, it can go awry.

You said, though that there's "no symptoms of life" - does that mean no output (that's expected), or no IO on the pool at all (not so much). Is there anything in dmesg that might suggest it found something upsetting on trying to import and died?