One idea about a potential fix:
- Make a backup of the db files every now and then - perhaps once a day
- Restore the backup if there’s an error with a db file
- Mail the operator that there was an error, backup was restored and they may have lost a bit of non-essential statistics data
This way a node will almost never go offline because of a db corruption - and since the db files are non-essential it’s not a big deal if some data in them is lost.