Scaling Resilient Systems: A Journey into Slack's Database Service
Talkat KubeCon + CloudNativeCon North America 2019 · on Tuesday, 2019-11-19· at 11:50· Ballroom Sec 20AB - San Diego Convention Center, Upper Level·
summary
Claims a Vitess-based database fleet scales from zero to 2.7 Million QPS and runs on 2000+ nodes with plans for 3500, deployed on AWS with cells that map to availability zones. Architecture moves from a single shared Consul to Consul per-topology and per-AZ cells to limit blast radius, uses semi-synchronous replication with one ACK for durability, Orchestrator for MySQL leader failover, and replica reads to cut latency (reported ~300 µs p50 win on some tables).
AI-written summary · about
chapters (18)
- 0:00Start
- 1:45Agenda
- 2:26Current status
- 2:54Why are we migrating?
- 3:34shards
- 9:11Durability through replication
- 11:12How we run Vitess
- 12:38architecture
- 14:07standard deployment
- 15:42Slack cloud infrastructure
- 16:58Vitess initial deployment
- 21:20We fixed things!
- 22:44Resilient systems
- 24:16Current deployment
- 26:19Performance wins
- 27:23Complex system failures
- 29:40Interested in learning more?
- 32:38Vitess configuration
related talks
36mHorizontal Scaling with Vitess
40mVitess: Intro and New Features
05mVitess: Unlimited Database Scalability
06mProject Lightning Talk: Vitess in Production: Powering Applications at Massive Sca...
45mDeep Dive: Vitess
41mScaling Databases with Vitess
28mThe Cloud is Inevitable
34mInside Vitess: Building Resilient, Distributed Databases