But getting back into a well-defined state which you can communicate via an error is a viable recovery strategy.
This state can be "the database is corrupt", allowing for a restore from backup, it can also be "shut down gracefully" and communicate that.
Especially for databases there is a ton to concinder: what about database entries that are not committed yet? When you run into oom, should you at least try to commit the data still in ram by freeing as much space as possible?
