I had RAM go bad after running 18T (5 HDDs in raid1) btrfs system in closet for years. Btrfs of course noticed it and fixed most of them automatically when some

I had RAM go bad after running 18T (5 HDDs in raid1) btrfs system in closet for years. Btrfs of course noticed it and fixed most of them automatically when some of blocks were corrupt. But eventually the system failed: the tree that contains the checksums for all the other trees corrupted itself on both copies of one node. Fixed the HW problem and then had to use hex editor to set the checksum manually to correct value (I modified the kernel to print the expected value). Now the system has been again stable for 3 years.