Cassandra Serving Netflix @ Scale
summary
Netflix runs Cassandra with 250+ clusters, 10,000+ nodes, 3+ PB and serves millions of ops/sec with multiple nines of availability. Operations replace cron checks with a Mantis streaming health check that pushes node heartbeats, runs a finite state machine and cuts alert noise from ~300 pages/day to <100/week. A Priam sidecar automates bootstrapping, token assignment and backups to survive instance, AZ and region failures. IndyBench benchmarks long-running, dynamic loads for certification; Winston auto-remediates common faults.
AI-written summary · about
chapters (10)
- 0:00Introduction to Cassandra
- 3:04Monitoring challenges
- 8:59Limitations of polling
- 13:54Mantis streaming solution
- 19:43Visualization and gains
- 23:50Priam sidecar tool
- 27:13Resiliency strategies
- 33:18Benchmarking with NdBench
- 38:24The architecture summary
- 41:43Q&A and future steps
related talks
52mHow Netflix Provisions Optimal Cloud Deployments of Cassandra
34mHow Netflix Delivers Key-Value and Time-Series Storage at Any Scale
44mTechniques Netflix Uses to Weather Significant Demand Shifts
25mImproving Bad Partition Handling in Apache Cassandra
26mBuilding Financial Systems on Eventually Consistent DBs
50mFast Delivery
24mHow Netflix enables
36mStories of Cassandra in Adobe Audience Manager