Parallelism in PostgreSQL 15 | Citus Con 2023

Video recording of a talk about the way PostgreSQL plans and executes parallel queries given by Thomas Munro at Citus Con: An Event for Postgres 2023. In contrast to the distributed multi-server parallelism that the Citus database extension provides, this talk is about stock PostgreSQL using multiple CPU cores on a single machine to run a single query. This talk illustrates the key concepts and problems by working through simple examples of workloads that can and can't benefit from CPU parallelism. There are also many cases where parallelism could help, but doesn't yet. Some of the opportunities for future development are discussed. Thomas Munro is a PostgreSQL developer and committer based in New Zealand. He began working full time on PostgreSQL 7 years ago, first at Enterprise DB and now at Microsoft. Before that he worked with Unix and relational databases in the web, finance, and software industries for a couple of decades. Some of Thomas’s PostgreSQL interests include query parallelism, taming resource management, transaction machinery, portability, and modernizing database/operating system interfaces. His other non-Postgres interests include hacking on the FreeBSD operating system, trying to learn other languages, and trying to ride various things with wheels. ► Video bookmarks: ⏩ 00:00 Introducing Thomas ⏩ 01:35 Thomas’ talk begins: Parallelism in PostgreSQL ⏩ 04:12 History of parallelism in relational databases ⏩ 06:15 The free lunch is over ⏩ 07:14 Embarassingly parallelizable problems ⏩ 11:12 Cost-based optimizer ⏩ 17:55 How do we control parallelism? ⏩ 24:37 Selected plan types ⏩ 29:16 Hash join memory struggles ⏩ 35:13 Q&A ✅ Learn more: Watch more Citus Con talks: https://aka.ms/cituscon-playlist 📕 Everything you need to know about Citus Con: An Event for Postgres can be found at: https://aka.ms/cituscon 📌 Let’s connect:   Twitter – @CitusCon,   / cituscon   #CitusCon #PostgreSQL #Database