RT @denismagda
What happens if two transactions are updating completely different fields but of the same record? Depends on how granular the database locks can be.
With row-level locking, one of the transactions waits. With column-level locking, there is no contention at all!
Follow
@FranckPachot interesting. Seems like it could be a surprise for a lot of people in many usecases though. Surely at minimum you'd need to track the column dependency graph at minimum, which could be non-trivial. Is it really a speed up that people need given the other tradeoffs?
@intrbiz the surprise that updating a column does not escalate to the row? Are people really relying on this? That's breaking physical-logical independence (think of updating through a view). In the past many RDBMS were escalating locks to pages or table and row level was considered as an evolution.