Asynchronous IO in Postgres 18

Postgres 18 just landed with a bucketload of shiny new toys—think native UUIDv7, an “Over explain” module for beefier query insights, composite index skip scans, and (drumroll) true asynchronous IO. In this episode of The Backend Engineering Show, Hussein walks us through why async IO matters, how it differs from the old sync approach, and the quirks of juggling IO in a forking database engine.

He then dives into Postgres 18’s two new io_method modes—worker threads and Linux’s blazing-fast io_uring—plus the fallback to good old sync IO. Along the way you get a behind-the-scenes look at the implementation challenges, what’s coming next (think backend writers and direct-IO support), and tips for squeezing every last drop of performance out of your database. Enjoy!

Watch on YouTube