The Gotchas of Zero-Downtime Traffic /w Kubernetes
Talkat KubeCon + CloudNativeCon North America 2019 · on Thursday, 2019-11-21· at 14:25· Ballroom Sec 20AB - San Diego Convention Center, Upper Level·
summary
Kubernetes defaults cause intermittent 503/502s during rollouts because endpoints are removed before pods finish draining and SIGTERM gets sent; kube-proxy and Ingress update asynchronously and the default terminationGracePeriod is 30s. Practical fixes shown: use HTTP readiness probes with conservative timeouts, avoid aggressive liveness probes, set an explicit preStop sleep (demo used 20s) and adjust rollingUpdate/maxSurge and minReadySeconds. Live demo on kind + siege + Nginx and uWSGI shows zero-downtime once probes and preStop sleep are applied.
AI-written summary · about
chapters (4)
related talks
38mCaulking the deployment gap: absolutely zero downtime deployments in Kubernetes
37mZero-downtime Deployment Strategies (...) [ENG]
39m502 Bad Gateway: Zero Downtime Is More Than a Config
40mZero-Downtime Java Enterprise Applications With Kubernetes
25mCrossing the Streams! Rollout Strategies to Keep Your Users Happy!
35mKubernetes, Lessons Learnt
40mStrategies for Zero Down Time, Frequent Deployments
49mZero-Downtime Deployment in Kubernetes & Istio