How we sped up Notion in the browser with WASM SQLite

Original Article ↗ Hacker News Discussion ↗

WASM SQLite in the Browser (Notion & Others)

  • Notion uses WASM SQLite with OPFS as a local cache; others report similar production use with Kotlin/JS and SQLDelight.
  • Reported benefits: powerful local querying (joins, full-text search), large local caches, and noticeable navigation speedups compared to server-only data fetches.
  • Significant engineering cost: awkward OPFS tooling/debugging, concurrency pitfalls, single-worker constraints, non-idiomatic official JS bindings, and complex multi-tab coordination.
  • Some describe WASM SQLite + OPFS as “incredibly powerful”; others call the stack “a nightmare” due to multi‑MB WASM downloads, corruption edge cases, and complex write coordination.

IndexedDB, LocalStorage, and WebSQL Debates

  • IndexedDB is widely criticized: verbose “NoSQL-ish” API, per-row overhead, inconsistent and buggy implementations, hard debugging, sometimes blocking the main thread even via workers, and no strong persistence guarantees.
  • LocalStorage: too small (≈10 MB), synchronous, contention issues, and unreliable with multiple tabs.
  • WebSQL/SQLite-in-browser: some argue it was a missed opportunity and would be far better than the current IndexedDB ecosystem; others argue standardizing on SQLite would freeze an old version and massively expand attack surface.
  • Several note that browsers already use SQLite internally but do not expose it, which is seen as a “missed chance” by some and a deliberate safety choice by others.

OPFS and File-System Concerns

  • OPFS gives more disk and better perf but has tricky concurrency; a “multiple readers/writers” proposal is implemented in Chrome, welcomed by some but not widely adopted.
  • Some wish browsers would expose a more “real” file system or align with WASI; others strongly oppose giving web pages direct file access for security reasons.

Performance, Local-First, and Cloud Costs

  • Multiple commenters praise local SQLite (and DuckDB for OLAP) for being able to scan tens of MBs–GBs in milliseconds on SSDs.
  • There’s debate over cloud IOPS pricing: some see local-offload as economically compelling; others argue million‑IOPS disk is overkill and real workloads are CPU/network bound.

Notion-Specific UX Concerns

  • Despite the new cache, users still complain about slow loading, laggy tables (especially on mobile), lack of true offline mode, and growing UI bloat.