Imperative to Declarative Infrastructure by For Developers and Operators
summary
eBay replaces imperative scripts with Kubernetes-native declarative objects and controllers, building a cluster manager called testmaster that provisions clusters on OpenStack, creates compute node pools, and self-hosts controllers. A custom NetworkScope plus an IPAM controller solves single global private IP allocation. OS images use Fedora Atomic, OSTree snapshots pushed to a container registry and Packer-built cloud images. Admission controllers enforce image scans and signing. Current clusters run ~200–300 nodes, with a target of 5k–10k nodes per availability zone, and developers keep a uniform kubectl experience.
AI-written summary · about
chapters (15)
- 0:00Start
- 2:23Kubernetes API Machinery for Declarative Models
- 4:33Specs and Controllers
- 9:08Clusters are isolated by network segmentation
- 11:00Networking Challenges
- 12:00Networking setup
- 13:07IPAM Controller
- 14:05Cluster management challenges
- 16:51TessMaster Cluster Creation Flow
- 18:24Host Runtime Challenges
- 21:00Cluster Upgrade Flow
- 22:05Application Modelling
- 23:06Admission Control specs Zone policy enforcement
- 24:26Ingress-plans
- 25:03summary
related talks
28mBuilding and Managing Kubernetes with Kubernetes
38mModel and Operate Datacenter by Kubernetes at eBay
03mFrom 5 to 1,300+ Clusters: Declarative Scaling for Private Kubernetes
29mProduction Cluster Monitoring and Remediation for High Reliability
36mBuilding a CI/CD driven infrastructure for managing Kubernetes clusters
36mOperator Framework to Manage Production Grade Stateful Workloads
45mKubernetes Operability Tooling
39mIntroduction to Kubernetes