Why would anyone willingly spend weeks chasing a slow query, knowing they might hit dead ends along the way? In Episode 36 of Talking Postgres, Tomas Vondra—Postgres committer and long‑time performance contributor—joins Claire to explain why hacking on Postgres performance is not just hard, but also fun. We dig into the process of investigating why queries are slow, how iteration and “wrong turns” are part of performance work, and why Tomas prefers meaningful performance puzzles over toy problems. Along the way, we talk about using benchmarks to build an understanding of a problem. Tomas also shares how even small changes in code can have outsized impact when that code is used a lot, and how the mathematics embedded in the Postgres query planner/executor makes the work especially rewarding.

Previously on Talking Postgres:

• Talking Postgres Ep31: What went wrong (& what went right) with AIO with Andres Freund
• Talking Postgres Ep24: Why mentor Postgres developers with Robert Haas

Links mentioned in this episode:

• PGConf.dev 2026: Schedule
• GitHub repo: PostgreSQL Monthly Hacking Workshop, organized by Robert Haas 
• Nordic PGDay 2026: Tomas talk on approximating percentiles
• Video of POSETTE 2025 talk: Performance Archaeology – 20 years of improvements
• Video of PGConf EU 2025 talk: Fast-path locking improvements in PG18
• Conference: Prague PostgreSQL Developer Day
• Discord: PostgreSQL Hacking Discord
• GitHub repo: tvondra/tdigest
• Brendan Gregg’s site: perf Linux profiler examples
• Docs: pgbench for running benchmarks on PostgreSQL
• Blog: Tomas Vondra blog
• Postgres Patch Ideas: List on Tomas Vondra blog
• Calendar invite: LIVE recording of Ep37 of Talking Postgres to happen on Wed Mar 18, 2026