Site availability is for everybody
summary
Shows how to reproduce outages with ApacheBench (ab) and read its output including 99th percentile ~693ms as a real signal not a misconfigured test, demonstrates common failures like non 2xx responses and Apache bench failed requests, details OS tuning steps such as ulimit/tcp tuning to avoid too many open files and ephemeral port exhaustion, describes queueing effects with unicorn workers vs a single database, and recommends monitoring tools New Relic Skylight rack mini profiler and slow query logging to find the real bottleneck.
AI-written summary · about
chapters (10)
- 0:00Intro
- 0:24What is site availability
- 4:27Load testing
- 6:08Traffic configuration
- 7:37Load test results
- 10:45Queueing theory
- 21:39Understanding bottlenecks
- 24:43Traditional tools
- 25:24Slow query log
- 26:16Load testing tools
