Unleashing a heard of elephants Parallel query is close to becoming a reality in PostgreSQL! A year ago, much of the low-level infrastructure needed for parallelism, such as dynamic shared memory and dynamic background workers, had been completed, but no user-visible facilities made use of this infrastructure. Major work on error handling, transaction control, and state sharing has been completed, and further patches, including a patch for parallel sequential scan, are pending. In this talk, we will talk about parallel sequential scan itself, including performance considerations, the work allocation strategy, and the cost model; and we will also discuss the infrastructure that supports parallel sequential scan, including state sharing for GUCs, transaction state, snapshots, and combo CIDs; error handling and transaction management; and the handling of heavyweight locking. Finally, we'll discuss the future of parallelism in PostgreSQL now that the basic infrastructure is (mostly) complete.