Monarch, Google’s Planet Scale Monitoring Infrastructure
summary
Monarch is Google’s planet-scale monitoring system that ingest and serves tens of billions of measurements from roughly 100 million targets by using a language-aware metric model, the streamy library, ingestion routers and memory-backed “leaves” with checkpoints and replication. Queries and aggregations (percentiles, joins, top-k) push through Zone and root mixers to align, preaggregate and fetch historical data from a long-term repo. It supports automatic alerting, retention policies, CI-backed configuration, and powers Stackdriver—deployed and used widely inside Google with redundant instances that monitor each other.
AI-written summary · about
chapters (14)
- 0:00Introduction to SRE
- 1:20Meet the Speaker
- 2:40Scale and Infrastructure
- 5:08The Need for Monarch
- 6:22Local Monarch Architecture
- 7:44Metrics and Targets
- 9:57Data Ingestion and Storage
- 12:54Querying and Alerts
- 14:29Global Architecture
- 16:23Query Execution Flow
- 20:10Using Monarch
- 23:07Unified System Benefits
- 26:55Lessons Learned
- 29:47Q&A Session
related talks
12mMonarch: An API To Your Supercomputer
24mScaling Telemetry To Terabytes: Applying Google's Principles To OpenTelemetry...
30mZenMonitor: Scaling Distributed Monitoring
16mMicroMon: A Monitoring Framework for Tackling Distributed Heterogeneity
39mThe Monitoring Path 2.0: Everything Has Changed
41mMonitoring Solr at Scale
24mMonarch: A Distributed Execution Engine for PyTorch
18mBuilding Resilient Monitoring at Meta