Ouch, that must have been a pain to recover from…

I’ve had almost the opposite experience to yours funnily. Several years ago my HDDs would drop out at random during heavy write loads, after a while I narrowed down the cause to some dodgy SATA power cables, which sadly I could not replace at the time. Due to the hardware issue I could not scrub the filesystem successfully either. However I managed to recover all my data to a separate BTRFS filesystem, using some “restore” utility that was mentioned in the docs, and to the best of my knowledge all the recovered data was intact.

While that past error required a separate filesystem to perform the recovery, my most recent hardware issue with drives dropping out didn’t need any recovery at all - after resolving the hardware issue (a loose power connection) BTRFS pretty much fixed itself during a scheduled scrub and spat out all the repairs in dmesg.

I would suggest enabling some kind of monitoring on BTRFS’s counters if you haven’t, because the fs will do whatever it can to prevent interruption to operations. In my previous two cases, performance was pretty much unaffected, and I only noticed the hardware problems due to the scheduled scrub & balance taking longer or failing.

Don’t run a fsck - BTRFS essentially does this to itself during filesystem operations, such as a scrub or a file read. The provided btrfs check tool (fsck) is for the internal B-tree structure specifically AFAIK, and irreversably modifies the filesystem internally in a way that can cause unrecoverable data loss if the user does not know what they are doing. Instead of running fsck, run a scrub - it’s an online operation that can be done while the filesystem is still mounted