Monitoring the Health of Your App
summary
Argument: averages lie — web response times are long‑tailed so 95th percentile and histograms matter. Implementation: an MVP service (Skylight) instruments Rails via notifications, subtracts GC noise (subtract GC time), and shows distribution heatmaps and endpoints ordered by importance. Surprises: many apps have one in twenty requests far slower than the mean, cache misses and GC spikes dominate slow tails. Outcome: concrete alerts from business metrics (dollars per minute, cart adds, checkouts) pinpoint which pipeline step actually breaks.
AI-written summary · about
chapters (9)
- 0:00Start
- 8:26Mental Model
- 8:41Measure Business Value
- 9:01Measure $ Earned
- 9:32Compare with Historical Data
- 11:43Performance is Business Value
- 12:42Statistics 101
- 14:53Standard Deviation Variance
- 35:48TILDE
related talks
18mLook at Your Data
37mLies, Damn Lies, and Metrics
30mHow to collect application metrics if it needs to be done yesterday
32mWhat’s my App *Really* Doing in Production?
32mSoftware Performance: A Shape Not a Number
19mOops my app performance sucks... what if observability could be a game changer?
36mHow to Build a Smart Profiler for Rails
30mRails Performance monitoring: A primer for developers