Deathmatch Bundler vs Rubygems.org
summary
The team reimplements the Bundler API, measures with Librato and Papertrail, and scales on Heroku; the result is an index and service that make rubygems.org respond server 100× faster and speed bundle resolution from bundle install 32s → 8s. Postgres outperforms Redis for this workload, moving median app-stack latency ~80ms→20ms and dropping the 95th percentile 600ms → 225ms; external probes reveal 5% external timeouts that drive design choices (Puma deadlocks, Thin/Unicorn tradeoffs, webhooks plus periodic sync, append-only index prototype).
AI-written summary · about
chapters (11)
- 0:00Start
- 12:36knowledge
- 15:27big postgres is faster
- 16:40redis caching is slow
- 18:48threading is great!
- 21:59usage is tricky
- 22:31metrics are tricky
- 23:53capacity is tricky
- 25:08monitor all of them
- 26:09response time is all lies
- 28:13check it from
related talks
29mFaster Bundler, Happier Developers
45m`bundle install` Y U SO SLOW: Server Edition
37m`bundle install` Y U SO SLOW: Server Edition
47mbundle install Y U SO SLOW: Server Edition
37mExtreme Makeover: Rubygems Edition
01mExtreme Makeover - Rubygems Edition
46mHeroku 2014: A Year in Review
43mReal World Ruby Performance at Scale
