June 1, 2026
Fsync and Furious
Benchmarking SurrealDB 3.x vs. Postgres, Mongo, Neo4j and Redis (With Fsync)
Database smackdown turns into a comment war over trust, tuning, and ‘just use Postgres’
TLDR: SurrealDB published new benchmark results against major databases and says this round uses safer, more realistic write settings. But commenters mostly turned it into a trust issue, arguing over missing details, fairness, licensing, and whether the answer is still simply: use Postgres.
SurrealDB tried to turn heads with fresh benchmark numbers against Postgres, MongoDB, Neo4j, and Redis, this time stressing that the tests used full durability settings — basically, the safer real-world mode where writes actually get flushed to disk instead of living dangerously in temporary memory. On paper, that sounds like a grown-up correction after last round’s eyebrow-raising omission. In the comments, though? Absolute side-eye.
The loudest reaction was a brutally simple one: why bother switching at all? One commenter boiled the whole thing down to “just use PostgreSQL,” arguing that even if SurrealDB wins some write speed, many teams care more about reads and the comfort of a giant, battle-tested ecosystem. Ouch. Others went full detective mode, demanding exact versions and settings, basically accusing the benchmark post of saying “trust us” while hiding the recipe. That fed the classic benchmark food fight: if you tweak every system differently, are you measuring the databases — or the tuning wizard behind the curtain?
And then came the purity police. One commenter swatted at the docs for calling SurrealDB “open source,” insisting it’s really source-available under a different license. Another dismissed the whole showdown with the devastatingly nerdy insult that the dataset fits in memory, meaning the test may not reflect the messy reality of huge production systems.
So yes, the article is about speed charts. But the real show is the crowd reaction: skeptical, snarky, and very ready to turn any benchmark into a courtroom drama.
Key Points
- •SurrealDB published a benchmark article for SurrealDB 3.x on May 29, 2026, authored by Tobie Morgan Hitchcock.
- •The article says benchmark fairness was addressed by using optimized, production-style configurations instead of default database settings.
- •The described test environment used 128 clients, 48 concurrent queries each, datasets with 5 to 15 million rows, and 128 GiB of RAM.
- •The workload used mixed-type records including strings, numbers, UUIDs, datetimes, booleans, large text, geospatial data, and nested objects and arrays.
- •The article states that the previous benchmark round had fsync disabled for all engines, while the new round changes that durability approach.