Fixing Twitter: Improving the Per
summary
Describes concrete fixes: centralized metrics at 10‑second intervals feeding a 1,200 metrics dashboard and Ganglia deploy graphs to spot regressions; automated review-board SVN hooks before config pushes; a live alert script called WhaleWatcher that counts 503s; moving heavy work into daemons and a custom queue called Kestrel; switching to a native libmc with the fnv hash and a C gem to cut web CPU by ~50%; Mill to kill runaway MySQL queries and functional partitioning instead of master writes.
AI-written summary · about
chapters (25)
- 0:00Intro
- 0:26Operations
- 1:41752% 2008 Growth
- 2:56Mantra!
- 3:40Find the Weakest Point
- 4:19Instrument Everything
- 4:48Monitoring
- 5:18Dashboards
- 6:01Analyze
- 7:33Deploys
- 8:20Whale-Watcher
- 8:55Take Action !
- 9:36Configuration Management
- 10:49Process through Reviews Review Board
- 11:54Improve Communication
- 13:41Make an attack plan.
- 16:42Disk is the new Tape.
- 19:34Memcached • memcached isn't perfect.
- 20:23API + Caching (search)
- 20:55Relational Databases not a Panacea
- 22:34Asynchronous Requests
- 23:16Don't make services dependent
- 23:34Daemons
- 24:49Database Replication
- 26:24Key Points
