March 26, 2026

Speed freaks vs. backup huggers

Show HN: Turbolite – a SQLite VFS serving sub-250ms cold JOIN queries from S3

Blazing speed has HN buzzing—big warning label, rival shout‑outs, and cache fights ensue

TLDR: A new tool makes a small database read straight from cloud storage in under a quarter‑second. The crowd loves the speed but debates the scary warning label, cache eviction details, and rival ideas like Graft—signaling a bigger shift toward fast, cheap cloud storage over local disks.

Hacker News saw a classic “speed vs. sanity” showdown as Turbolite promised quarter‑second “cold” queries—first‑time reads with nothing cached—from cloud storage. In plain English: it’s a tiny database reading straight from Amazon‑style buckets and still feeling fast. The catch? The project is experimental and says it may corrupt your data. Cue the popcorn.

The builder, russellthehippo, jumped in with the brainy backstory, shifting the convo from “can it read over the network?” to “what breaks when storage is the cloud instead of a disk?” He teased a big rethink around data layout and proudly noted the goal was to beat another startup’s ~500ms cold starts—mission accomplished. That swagger sparked cheers and side‑eye in equal measure.

Meanwhile, practical folks demanded receipts on real‑world survival: alex_hirner pressed for an eviction strategy—non‑tech translation: how do you throw old stuff out of the cache without wrecking performance? And then rival energy crackled when carlsverre dropped a link to Graft, pitching a different take that keeps writes small and favors “partial replication.”

The vibe: speed freaks are giddy, backup huggers are clutching their USB drives, and everyone’s comparing notes with Litestream/LiteFS, Turso, Neon, and more. Jokes flew about “S3 in S3conds,” while others warned not to store your wedding photos here just yet. Still, seeing sub‑250ms pulls from the cloud is a mood—like watching the gap between local disks and cloud buckets evaporate in real time.

Key Points

  • Turbolite is an experimental Rust-based SQLite VFS that serves point lookups and JOINs directly from S3 with sub-250ms cold latency.
  • It provides page-level zstd compression and AES-256 encryption, usable independently of S3.
  • The tool ships as a Rust library, a SQLite loadable extension, and language packages for Python and Node.js (Go via GitHub), supporting S3-compatible backends (AWS S3, Tigris, R2, MinIO).
  • Benchmarks on 1M rows (~1.5GB) show cold query latencies on S3 Express One Zone as low as 75–188ms for various indexed queries and 591ms for a full scan; Tigris cold latencies are higher (159–681ms; 921ms for full scan).
  • Design principles optimize for object storage constraints: minimize request count, maximize bandwidth, account for per-operation costs, treat objects as immutable, and deprioritize space efficiency.

Hottest takes

“what assumptions break once the storage layer is object storage instead of a filesystem?” — russellthehippo
“What are your thoughts on eviction” — alex_hirner
“keep writes fast and small and optimize for partial replication” — carlsverre
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.