Graphite@Scale or How to Store Millions of Metrics per Second
summary
Booking.com pushes Graphite-compatible stack to handle millions of unique metrics/sec, ingesting thousands of gigabits/s and storing hundreds of terabytes in Whisper. A custom Carbon relay + local relays at producers replaces upstream bottlenecks; a new query frontend cuts render latency by 15x query speedup. Jump Consistent Hashing drives shard imbalance under 1% distribution variance. Experiments on replication factor, tagging service (current-state only), quota enforcement, differential disk-usage diffs, and open-source releases complete the operational story.
AI-written summary · about
chapters (11)
- 0:00Introduction to metrics
- 1:38Graphite infrastructure
- 3:29Scaling and system challenges
- 5:25Customizing the stack
- 7:45Improving data access
- 9:48Distribution and performance
- 12:23Replication and hardware
- 14:07Tagging and system stability
- 16:45Lessons in user management
- 20:11Differential data analysis
- 23:40Future goals and conclusion
related talks
36mMetrics - What's your code actually doing?
20mHow We are Dealing with Metrics at Scale on GitLab.com
59mGathering Metrics With StatsD And Graphite
27mEasy App Metrics
30mCloud Foundry Monitoring and Metrics Collection Using Sensu and Graphite
39mHow To Take Prometheus Planet Scale: Massively Large Scale Metrics Installations
51mCollectd and Graphite: Monitoring PostgreSQL w/Style
31mHow the Metrics Backend Works at Datadog