It's not in Production Unless it's Monitored
summary
Argues that production is incomplete without instrumentation and treats monitoring as “tests for ops,” showing concrete stacks and patterns that work in practice: ActiveSupport::Notifications for Rails metrics, statsd on each host (a tiny UDP daemon with almost zero cost) feeding Graphite + Whisper or OpenTSDB for storage, Logstash/Papertrail for logs, and deploy-event overlays for correlation. Demonstrates immediate post-deploy error spikes and fast fixes, warns about Graphite capacity needs, recommends alert hysteresis/rearm windows and using triple exponential smoothing to detect seasonal anomalies.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:22Infrastructure shift and monitoring
- 10:07Monitoring model for modern systems
- 13:18Collection as monitoring foundation
- 18:57Aggregation and centralization
- 23:22Visualization and correlation
- 27:39Dashboards and shared understanding
- 30:21Alerting strategies
- 33:57Conclusion and key takeaways
related talks
32mWhat’s my App *Really* Doing in Production?
28mMonitoring Rails apps using Open Source
41mRuby Monitoring State of the Union
21mBuilding a Performance Analytics Tool with ActiveSupport
28mProduction Testing through Monitoring
06mMonitoring as a First Step to a New Service
13mForget Production: Observability a Non-Production Necessity
39mKeeping tabs on production