English Wikipedia gives the following wording read committed
This is a leased-assumption of the DBMS implementation of the transaction.
I do not fully understand why the requirement is being made to keep the blocking at the isolation level, which allows reading the changed data. As far as I understand, the database cannot write unencrypted data directly on top of existing ones just physically, because otherwise a transaction cannot be rolled back, and it needs staging area / wal anyway, so the option is to keep these locks so that no one can see uncommitted data, me also seems wrong. Why does this isolation level still require retention of write locks? Because otherwise it is contrary to the atomic nature of the transaction?
