exception type std::runtime_error: FIXME: bailing out here, need to fix this further up the call stack about bees HOT 2 CLOSED

The current bees algorithm loops too many times when many blocks in many files are identical. This is fairly common in BMP files. The exception just breaks the loop. If the loop is allowed to run until completion it can take several days to process each file.

Can you say more about what exactly is going wrong with the filesystem?

from bees.

When it first happened I asked on the BTRFS mailing list for help to recover (https://www.spinics.net/lists/linux-btrfs/msg75143.html) but I was using zstd compression so couldn't use any of the normal tools, so I gave up. That instance of corruption was particularly bad, as it was only noticed after the FS went read only.
The second time (during the rebuild) I was being a bit more proactive about running scrub every day or two, which is how I found the corruption. There were no IO errors or other kernel messages beforehand. I wasn't capturing the output from bees so I don't know if it emitted any errors.
I tried to fix the FS, but btrfs check couldn't do it. Then I tried to btrfs check --init-extent-tree which seemed to be working, but I stopped it after millions of errors and 26 hours.
I'm rebuilding for the 3rd time now and no errors so far. I haven't deleted any snapshots with bees running (I remember a while back that would crash bees). Just create snapshot, rsync data, and repeat.
I may have also rebooted while bees was running, and if the drive didn't unmount in time then a bunch of dirty metadata would get lost.
Basically I have no idea where the problem is at the moment, still looking. I doubt it's bees though, more likely the kernel dedupe ioctls. If I have anything to report I'll let you know

from bees.