Traffic Monitor Distributed Polling
summary
Traffic Monitor is refactored so a Go-based monitor can run in distributed polling enabled mode and split thousands of caches across tm groups instead of pulling all ~5,000 caches from one box. Health polling stays first (currently 2s, target sub‑second / ~100ms) while stat collection remains on legacy monitors (stat polling disabled for distributed TMs). Peers form groups, use an optimistic consensus (all must agree to mark unhealthy), require ~3 peers for safety, and keep existing APIs and Traffic Ops as config source.
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:44About Dave
- 3:01Traffic Monitor
- 4:03Core Functionality
- 10:45Traffic Monitor Configuration
- 12:30Traffic Monitor Limitations
- 15:23High Level Requirements
- 19:12Design Decisions
- 25:45Traffic Monitor Perspective
- 27:50Control Plank Perspective
- 29:35Compromises
- 32:42Migration Strategy
- 34:30Road Map
- 37:30Resources
related talks
30mFrom Polling to Streaming: Network Monitoring with Real-Time Telemetry
06mMonitoring as a First Step to a New Service
43mMonitoring at one of the largest retail groups in the world
40mThe Prometheus monitoring system
23mAutomatic Traffic Scheduling for Internet Connectivity Failures
08mNetwork Performance Monitoring
14mMonitoring the monitor
16mClosed-loop Network Performance Monitoring and Diagnosis with SpiderMon