From Containers to Kubernetes Operators
summary
ElasticSearch images enforce a specific UID/GID, forbid a no latest tag workflow, use a CentOS7 base (JVM included) and discourage runtime mutation; operators are written in Go with a reconciliation loop, manage certs, security, rolling upgrades via StatefulSets and persistent volumes (EBS or local PV), and drain data before scale-down. Operator binary needs ~200–300MB RAM, is production-used and tested on GKE, and required roughly 5+ full-time engineers for a year to build the current feature set.
AI-written summary · about
chapters (14)
- 0:00Introduction and overview
- 1:11Building container images for data stores
- 5:46Why avoiding 'latest' tag is critical
- 8:13Best practices for container images
- 10:22Choosing base images for containers
- 13:12Docker Compose for local setups
- 14:47Kubernetes and Helm charts basics
- 19:58Helm charts for Elasticsearch
- 20:20What are Kubernetes Operators
- 25:39Deploying Elasticsearch with an Operator
- 30:15Running Operators securely
- 33:25Should you run Elasticsearch on Kubernetes?
- 37:31Operator vs Helm chart comparison
- 40:10Q&A on operator development effort
related talks
49mFrom Containers to Kubernetes Operators
21mStory of a Kubernetes Go Operator
34mWriting a Kubernetes Operator: the Hard Parts
44mFrom containers to Kubernetes operators
43mLessons learned writing 10+ Kubernetes Operators
42mImproving Containers Isolation in Kubernetes
36mLessons learned writing 10+ Kubernetes Operators - Lars Francke and Jannik Heyl
01h 29mTutorial: Deep Dive into the Operator Framework for Kubernetes