Migrating and running production services in Kubernetes
summary
A small Java team migrates ~20 REST microservices to Kubernetes over August–October and moves 100% traffic migrated after phased canary testing, cutting server costs by ~20% lower server costs and reducing new-service setup from eight hours to one while keeping latency unchanged. An auth service uptime rises (example: ~99.75% → ~99.999% over three months). Practical fixes: start one service at a time, use separate kube config per cluster, tune rollingUpdate to create one pod at a time (maxUnavailable=0, maxSurge=1), use HPA+cluster-autoscaler, handle SIGTERM and add a preStop hook sleep to avoid 500s from iptables/socket reuse.
AI-written summary · about
chapters (7)
- 0:00Start
- 2:34Why Kubernetes
- 5:11Lessons Learned
- 7:04Choose the Kubernetes Solution
- 7:34Making Production Check Ready Checklists
- 8:37Tuning the Rolling Updates
- 11:59The Port Life Cycle
related talks
20mKeynote: The Challenges of Migrating 150+ Microservices to Kubernetes
35mPerforming Infrastructure Migrations at Scale
38mBuilding the Perfect Infrastructure with Kubernetes
59mJava microservices from Netflix OSS to Kubernetes
45mLearning How to Pronounce Kubernetes to Production in 3 Months!
36mOptimizing Resource Usage in Kubernetes
45mHow Kubernetes Was the Secret Sauce in Our Globally Distributed Team’s Transition to Microservices
44mMigrating a Monolith to Kubernetes