A Journey to Safe, Mission Critical Continuous Delivery at Descartes Labs
summary
Descartes Labs runs several Kubernetes clusters with dozens of services and tens of thousands of Pods and CPU cores and achieves a combined deployment frequency above 150 releases per day from a team of 30 engineers. The delivery stack uses Spinnaker, Istio & Envoy plus a dynamically templated Spinnaker pipeline (JSON beside app code) that automates canary analysis by patching Istio VirtualServices and comparing canary vs fresh baseline; result: ~3 deploys/day per service, 16 active services, and zero rollbacks for three months.
AI-written summary · about
chapters (20)
- 0:00Intro
- 0:23What is this talk about?
- 1:48Wildfire Detection
- 2:36Mission Critical Delivery?
- 3:21The value of high-velocity continuous delivery
- 5:16Istio and Kubernetes
- 8:48From Development to Production
- 10:45A developer wants to start deploying a new application
- 12:14Jinja Pipeline Templates
- 14:52k8s/deployment.yaml
- 24:07Safer deployments using canaries
- 25:55Why do we need a baseline?
- 29:15The Canary_Rollout pipeline
- 32:00Configuring Metrics & Analysis
- 33:41Dynamically steering traffic with Spinnaker and Istio
- 35:09Canary Rollout Questions
- 38:58A tale of two canaries
- 41:11How does this all work out? The numbers
- 41:58Conclusions
- 43:25Descartes Labs SRE Team
related talks
58mBuilding a Culture of Continuous Delivery
33mEvolving a Continuous Delivery Workflow
33mPanel Discussion: Spinnaker Success - CD as a Service using Spinnaker
52mContinuous Deployment to the Cloud using Spinnaker
48mContinuous Delivery with Spinnaker
35mLeveling Up Your CD: Unlocking Progressive Delivery on Kubernetes
54mFrom Code to Customer
23mThe Nine Journey: Cloud Native Transformation