June 20, 2026
Bench Wars: Cloud Edition
PostgresBench: A Reproducible Benchmark for Postgres Services
A new Postgres speed test drops, and the comments instantly turn into a cloud fight
TLDR: PostgresBench is a new open benchmark meant to publicly compare hosted Postgres database services using the same setup so anyone can check the results. Commenters loved finally getting a shared scoreboard, but immediately argued about missing providers, whether app-side slowdowns matter more, and joked about letting AI spend a fortune tuning everything.
A new public benchmark called PostgresBench has arrived with a big promise: finally compare hosted Postgres database services in a way people can actually reproduce, check, and argue about. The company behind it says it wants the same transparent, open-book vibe as its earlier database benchmark work, using a common built-in Postgres test to measure speed and delays under pressure. In plain English: it’s a leaderboard for online database services, with the receipts supposedly included.
But the real entertainment is in the reaction. One crowd basically shouted, “Finally!” User aurareturn summed up the mood: cloud database providers all look similar and cost similar money, but nobody really knows who’s secretly fast and who’s just good at marketing. That made this launch feel less like a boring test and more like someone dropping the stopwatch at a family reunion.
Then the nitpickers, skeptics, and chaos agents entered. Schultzer pushed back with a classic comment-section twist: why obsess over the database alone when the apps and tools talking to it are often where performance problems begin? Translation: is this benchmark useful, or are we grading only half the group project? Meanwhile, whitepoplar immediately demanded more contestants, asking where PlanetScale Postgres is in all this, because apparently no benchmark is complete until someone says, “Cool, but you forgot my favorite.”
And the funniest comment may belong to datadrivenangel, who joked that with a giant AWS bill and Claude Code, they’d unleash an AI to brute-force every possible tuning setting. Nothing says 2026 like turning cloud benchmarking into an autonomous money bonfire.
Key Points
- •The article introduces PostgresBench as a public, reproducible benchmark for comparing managed Postgres services.
- •PostgresBench follows the transparent methodology of ClickBench, including public configurations, reproducible runs, and the ability for others to validate or challenge results.
- •The benchmark uses pgbench with its built-in TPC-B-like workload to simulate short, concurrent transactional workloads with frequent writes and updates.
- •Benchmark runs use 256 clients, 16 threads, a 10-minute duration, prepared statements, and two scale factors representing datasets of roughly 100 GB and 500 GB.
- •Reported metrics include average TPS, average latency, P95 latency, and P99 latency across three runs, with exact system settings and run details published for fairness and reproducibility.