It's Not in Production Unless it's Monitored
summary
Monitoring is mandatory for production: make collection near-zero friction and treat instrumentation like unit tests. Use cheap collectors such as ActiveSupport::Notifications and UDP-based statsd (the daemon is ~319 lines) to emit counters/timers, aggregate into a storage backend like Graphite/whisper or OpenTSDB, and surface correlated dashboards. Add deploy/event annotations, tune alerts with cancel thresholds and rearm windows, and apply time-series models (Holt‑Winters, EWMA) for anomaly detection so alerts stay actionable.
AI-written summary · about
chapters (40)
- 0:00Introduction
- 0:47Where the quote came from
- 1:42Context
- 2:20Bootstrap
- 2:54Infrastructure
- 4:03Continuous Deployment
- 5:16Continuous Integration
- 7:13Travis CI
- 8:06Bacon
- 8:37Monitoring Tools
- 9:43Monitoring Sucks
- 10:17Better Model
- 10:31Metrics
- 10:49Application Performance
- 11:51Collection Stage
- 12:54Separation of Concerns
- 13:11Collection
- 13:42Logging
- 14:25Log Stash
- 14:54Active Support Notifications
- 16:20StatsD
- 16:59StatsD Setup
- 17:53StatsD Clients
- 18:23StatsD Instrument
- 18:53Collection and aggregation
- 19:28Round robin
- 20:21Graphite
- 20:53Whisper
- 21:55OpenTimeSeries
- 23:07Visualization
- 23:38Events
- 25:04Dashboards
- 25:51Marketing Guy
- 26:21Dashboard
- 27:06Dashboard Example
- 27:34Nother Talk
- 27:48Alerts
- 30:05Alert thresholds
- 31:16Integrations
- 33:50Conclusion
related talks
41mRuby Monitoring State of the Union
21mBuilding a Performance Analytics Tool with ActiveSupport
39mKeeping tabs on production
58mObserving Applications in Production
47mDid My Deploy Degrade Production?
22mService Monitoring Manual—2018 Edition
28mMonitoring Rails apps using Open Source
59mMonitoring A-Z