How CoreOS sees the container industry
summary
CoreOS describes etcd as a CP replicated key‑value store built for cluster coordination: it implements compare-and-swap and streaming watches for dynamic config, a semaphore example with a lock count of three, and leases for leader election. v3 replaces JSON-over-HTTP with gRPC/protobuf, adds mini‑transactions and multi‑version reads, and targets far higher scale (from ~40,000 req/s in v2 toward tens of thousands of machines and millions of containers) while fixing snapshot pauses and reducing memory. Prometheus metrics and safe runtime reconfiguration and backups are provided.
AI-written summary · about
chapters (26)
- 0:00Intro
- 0:23Why CoreOS
- 1:00Reboot lock
- 1:40Cluster example
- 2:25Updates
- 2:35Application Configuration
- 2:57Dynamic Updates
- 3:40Safety Guarantees
- 4:03Key Requirements
- 5:07Cap Theorem
- 6:56Other providers
- 7:38Zookeeper
- 8:30Eight Cities
- 9:50TEE
- 10:15Key Value Store
- 12:30Operational Features
- 17:30Conditional Operations
- 18:40Lease
- 19:50Streaming watch
- 20:10Curator vs Zookeeper
- 21:50Zookeeper performance
- 22:30Zookeeper performance degradation
- 23:20Zookeeper performance management
- 23:40Zookeeper performance comparison
- 24:10Sub performance comparison
- 24:35Memory usage
related talks
42metcd: distributed locking and service discovery
36mWhat You Should Know about Etcd v3
26mDon't Let Your Kubernetes Cluster Go Wild: Ensuring Etcd Reliability
38metcd in 2026: Enhancements, V3.7 Roadmap...
06mKeynote: Distributed Systems Simplified Using Kubernetes
23mThe Secret Recipe of etcd
15mKubernetes and etcd Features That Unlocked 15k Node Clusters
20mLessons Learned from Operating ETCD