Deep Dive: Vitess
summary
Vitess runs a sharded MySQL fabric across three Kubernetes clusters using VTGate and VTTablet plus Orchestrator and a topo server (etcd/zookeeper); the demo brings up shards with no persistent disks, uses S3 backups, Prometheus metrics and semi‑sync replication, kills pods and an entire EU region, and automatically elects new masters. Recovery takes about 2 minutes in the demo (can be tuned down toward 9 seconds), and at YouTube scale Vitess ran 256 shards with 80–120 replicas across 20 cells, proving automated failover works end‑to‑end.
AI-written summary · about
chapters (40)
- 0:00Introduction
- 1:02Who are you
- 2:28Vitess Architecture
- 10:00The App
- 11:30The Architecture
- 15:00Test Dashboard
- 15:45Automatic Ratings
- 16:05VT Gate Interface
- 16:40Orchestrator
- 17:05Replicas
- 17:20Planned Reparent
- 17:45Repack
- 18:10Switching between nodes
- 18:40Things fail
- 19:30Deleting
- 20:35Rate
- 20:55Health checks
- 21:25Master tablet fail
- 21:45Orchestrator failed
- 22:00Killing the master
- 23:00Inserts fail
- 23:30Orchestrator recovery
- 23:55New master selection
- 25:10What if a whole data center goes down
- 25:20Example
- 27:30Failover Mode
- 28:50Recovery Procedure
- 29:20VT Gate
- 29:35Back to Terminal
- 30:00Google shenanigans
- 30:20Database recovery
- 31:00Questions
- 31:25catastrophic failures
- 33:05What happens if I lose the master
- 34:55Compatibility
- 36:35Sharding
- 39:25Buffering
- 40:30GT IDs
- 41:25Will it work with managed database systems
- 42:50Orchestration configuration
related talks
35mExploring High Availability in Kubernetes with Vitess
34mInside Vitess: Building Resilient, Distributed Databases
44mIntro: Vitess
31mVitess: Running MySQL with confidence on Kubernetes
35mExploring High Availability in Kubernetes with Vitess
43mVitess and a Kubernetes Operator
44mIntro: Vitess
28mVitess: Introduction and New Features