Scaling to 5000+ Unique K8s Deployments, How We Did It
summary
Company migrates 60,000 WordPress sites off VMs by replacing 300,000 per-site PHP deployments with a custom Go PHP app server and a pooling model that reduces needed Kubernetes deployments to 6,000 deployments across regional clusters. Automation uses Helm/Tiller plus a custom listramos operator that creates Helm releases from VM-representing custom resources; target rollout time drops from >24 hours to ~1 hour while encountering a Helm/Tiller concurrency bug that they work around.
AI-written summary · about
chapters (10)
- 0:00Start
- 0:33WordPress
- 1:45What are we doing
- 3:05Kubernetes 1.8 Limits
- 4:04Custom PHP Application Server
- 6:59Deployment Problems
- 7:48Operators
- 8:36Custom Resources
- 21:17Challenges
- 24:07Examples of Other Use Cases
related talks
28mLessons Learned Deploying Traditional Web Applications on Top of Kubernetes
22m10 Kubernetes is hard!
47mDeploying Like a Boss: From 2 Servers to 20,000
50mAutoscaling, Rolling Deploys with Drupal on AKS
33mScaling k8s Deployments: Achieving Elite DORA Metrics
34mKubernetes from Dev to Prod at GoEuro
35mKubernetes journey
32mDon't Start With 500 Clusters: Patterns in Scaling Multi-Cluster Kubernetes…