What is the difference between flush, autoflush, commit, and rollback in SQLAlchemy?

I use flush when later work in the same transaction needs a generated identifier or when I want a constraint failure at a deliberate point. An order flow can flush to obtain the order id before inserting lines while one later commit still owns the complete change. I translate integrity errors carefully because a unique violation can be the expected result of a race, not a server fault. Tests distinguish flush from commit by checking visibility through another transaction and verify that failure paths cannot continue with an inactive Session.