Continuously Deliver your Kubernetes Infrastructure
summary
Zalando runs ~84 Kubernetes clusters across 366 AWS accounts and consistently upgrades clusters from v1.4 to v1.9 using CloudFormation, Container Linux, an external etcd stack and a custom cluster lifecycle manager that watches Git and a cluster registry. Every PR triggers Jenkins pipelines that create three test clusters and run the upstream conformance test (144 tests) plus statefulset (Redis) and ingress/external-dns checks. Node upgrades use taints, pod disruption budgets and a Postgres operator for safe failover; flannel backend migration and a saturated Docker registry caused real outages.
AI-written summary · about
chapters (17)
- 0:00Intro
- 0:50ZALANDO AT A GLANCE
- 2:04INFRASTRUCTURE @ ZALANDO
- 6:07PHILOSOPHY
- 9:58CLUSTER SETUP
- 12:59CLUSTER METADATA CLUSTER-REGISTRY
- 14:06CLUSTER METADATA (CLUSTER-REGISTRY)
- 15:09CLUSTER LIFECYCLE MANAGER CLM
- 18:45CLUSTER CHANNELS
- 27:06HINTS FOR RUNNING E2E TESTS
- 28:50NAÏVE NODE UPGRADE STRATEGY
- 29:45PROBLEMS WITH THE NAIVE STRATEGY
- 31:39NODES READY?
- 32:36POD DISRUPTION BUDGETS
- 33:25POSTGRES OPERATOR
- 35:08ROLLING UPGRADE OF NODES
- 39:30OPEN SOURCE
related talks
27mRunning Kubernetes in Production A Million Ways to Crash Your Cluster
35mKubernetes on AWS: Practices & Opinions
35mKubernetes journey
25mAutomating Kubernetes Upgrades for Zero Downtime
35mTo Infinite Scale and Beyond: Operating Kubernetes Past the Steady State
19mZuul CI for Kubernetes on the Workday Private Cloud
19mPains and gains of Kubernetes for Developers, story from experience
41mKubernetes — A Year in Production