Reflections on Haskell@Meta
summary
A Haskell anti‑abuse engine called Sigma grows to over a million lines, runs millions of requests per second with three to four nines reliability, and supports near‑instant deploys (commit to production ~30 minutes) with a team of about a dozen. Key inventions: the Haxel data‑fetching library and applicative do translation to recover concurrency. Practical costs include GHC runtime patches, hard‑to‑diagnose deterministic compilation bugs, and a ~1.5 person‑year compiler upgrade burden; Sigma is sunset around 2024 and much work moves into Hack while Glean (a separate Haskell code indexer) continues in production.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 3:37Haskell at Meta growth
- 10:19Sigma anti-abuse platform
- 22:15Applicative Do in Sigma
- 32:15Haskell challenges at Meta
- 43:50Debugging Haskell services
- 51:35Sigma sunset reasons
- 55:47Glean code indexing
- 1:04:04Haskell advocacy lessons
- 1:15:10Q&A
related talks
46mPainless software development with Haskell
47mIndustrial Strength Laziness: What's Next?
54mHaskell Project Showcase and Community Contributions
01h 07m50,000 Lines of LLM Haskell: What Broke, What Held, and Letting Go
22mHaskell Tools, Compilers, Libraries and Community Projects
52mProject Presentations - Haskell
49mHaskell On Rails
21mProduction Haskell Demystified