Deliver Results, Not Just Releases: Control & Observability in CD
summary
Companies decouple deployment from release using feature flags so updates deploy everywhere turned off, then ramp exposure and observe outcomes; LinkedIn runs controlled releases every five minutes (~100/day), Booking.com runs thousands of experiments concurrently and auto-reverts within a second on high error rates, Facebook Gatekeeper stages to 2% then can dial by 0.1%. Observability pairs an audit of who saw which treatment with timestamps so teams use guardrail metrics and backend SDKs to attribute behavior changes to a deployment.
AI-written summary · about
chapters (43)
- 0:00Introduction
- 1:04Agenda
- 1:53Copy the deck
- 2:15Who am I
- 8:37Why Im here
- 9:23Definitions
- 10:40Control Observability
- 11:43The payload
- 12:09Control
- 14:49Feature Flags
- 16:01Experimentation
- 17:17Code
- 18:40Observability
- 20:03Main Body
- 20:20LinkedIn
- 20:31Exellent
- 21:12Observability Metrics
- 22:31Guardrail Metrics
- 24:28Unintended Consequences
- 24:49Build for scale
- 25:25Trustworthiness
- 26:14Diversity
- 28:15Bookingcom
- 28:43Transparency
- 29:11Scale
- 31:32Deploying
- 31:59Agility
- 32:38Safety Net
- 34:26Circuit Breaker
- 34:59Auto abort
- 35:32Validate ideas
- 36:55Build measure learn loop
- 37:24Validate new ideas
- 38:30Lukas Premier
- 41:50Facebook Gatekeeper
- 42:14Complexity
- 43:04Undo Button
- 46:17Conclusion
- 46:28Deployment vs Release
- 47:51Where to Put Controls
- 49:30Where People Miss It
- 50:41Audit Trail
- 51:34Two pieces of data
related talks
53mObservability Driven Development
38mProgressive Delivery: Patterns & Benefits of Decoupling Deploy from Release
27mDEVOPS DAYS MTL 2018 - Lenore Adam
44mRelease Early, Often, Stable
06mDeploying Radically, Democratizing Release
43mTo make mistakes is human. To auto-deploy it is DevOps.
32mHow Your Deployment Strategy Influences How People Code
59mThe continuous culture