performance
Talk
Hey, I'm using that! Fixing lock contention in OLTP.
Some transaction processing workloads end up with horrible lock contention because they end up blocked on updating the same rows. There are now databases that advertise running this workload a 1000 times faster than PostgreSQL. In this talk will discuss strategies how to manage this contention in PostgreSQL while retaining application correctness. How network latencies, different isolation levels, optimistic and pessimistic concurrency control, deadlocks and livelocks affect the capability to get work done. Working with the database allows us to take a large step closer to single-purpose database performance while staying in our familiar PostgreSQL land.