Continuous Deployment at Scale
summary
Etsy runs small, incremental releases — roughly 20–40 deploys a day (about 30 code deploys plus 30+ config pushes) — by shipping directly to master behind feature flags and multivariate experiments, gating changes with a local dry-run tool try and push UI deployinator. Team coordination uses an IRC push train and pushbot; monitoring is Supergrep/Supertop and Graphite dashboards; incidents get blameless postmortems recorded in MARG. Rollbacks are preferred over quick patches, and the pipeline combines tooling plus culture to keep mean time to recovery low.
AI-written summary · about
chapters (27)
- 0:00Start
- 1:19principles
- 1:33enable innovation
- 1:51optimize for purpose
- 2:03optimize for mastery
- 2:19experimentation A/B testing
- 2:51quick iteration
- 3:17continuous improvement
- 3:35low MTTR
- 4:25frequent check-ins
- 7:24continuous integration
- 7:43TryLib
- 10:27dark changes
- 10:39template changes css tweaks unreferenced code
- 11:09config pushes
- 11:41push train
- 13:08post-deploy
- 14:05dashboards
- 15:09summary
- 16:31poka-yoke
- 17:17assume best intentions
- 17:40cultivate empathy
- 17:54failure is an option
- 18:28failure is an opportunity
- 18:43blameless post-mortems
- 19:36+ remediation
- 21:15mixer
