Etcd 101: The Parts They Don't Teach You
summary
A deep dive into etcd internals and Raft consensus: how replication works, the role of the WAL, snapshot management, MVCC semantics, and why compaction is required for reliable production operation. Includes live demos of network failures, partitions, leader elections and recovery, and concludes with practical guidance on tuning, security and operating etcd in production as Kubernetes’ single source of truth.
AI-written summary · about
chapters (32)
- 0:00Introduction to etcd
- 1:27What etcd really is
- 4:36etcd’s MVCC key-value store
- 6:24Setting up etcd demo
- 9:15Raft consensus in etcd
- 11:00Connecting to etcd directly
- 12:48Watching etcd in real time
- 13:49Leases and time-based keys
- 15:12etcd transactions
- 16:57Filling the etcd database
- 17:17etcd architecture layers
- 18:03How Raft elects a leader
- 20:54Split-brain protection
- 21:15BoltDB storage engine
- 24:05In-memory index for speed
- 25:29Database size and compaction
- 27:17Defragmenting etcd
- 28:22etcd alarm management
- 29:51Latency in etcd writes
- 31:17Storage backend choices
- 32:43Authentication bottlenecks
- 34:05What happens when etcd freezes
- 35:10Quorum and cluster size
- 35:32Losing a cluster member
- 38:00Recovering from node failure
- 39:53Disaster recovery with snapshots
- 40:36Creating etcd snapshots
- 41:59Restoring from snapshots
- 43:05Upgrading etcd safely
- 43:27Essential etcd metrics
- 44:53Tuning etcd for performance
- 45:38Q&A
related talks
40mDeep Dive: etcd
20mLessons Learned from Operating ETCD
37mUnderstanding Distributed Consensus in etcd and Kubernetes - Laura Frank, CloudBees
19mDr etcd; or; How I Learned to Stop Worrying and Love the Datastore - Nick Young, VMware
23mThe Secret Recipe of etcd
31mDemystifying Kubernetes Failure Scenarios Caused by etcd
26mDon't Let Your Kubernetes Cluster Go Wild: Ensuring Etcd Reliability
42mHow Should You Effectively Use etcd Metrics