Intro + Deep Dive: Prometheus
summary
Prometheus runs a pull-based model storing float64 time series locally on a single-node TSDB (~1.6–1.7 bytes/sample) and expects client libraries or exporters for instrumentation. PromQL does label-aware set operations and percentiles (histogram quantiles) so you can compute request error ratios and per-path 99th percentiles. Alerting rules evaluate in Prometheus and route to Alertmanager. For durability and multicluster views use Thanos or Cortex; long-term forwarding uses remote_write (protobuf + snappy); typical scrape intervals are 15–30s.
AI-written summary · about
chapters (28)
- 0:00Intro
- 2:49History
- 3:54System Architecture
- 8:47Primitives
- 9:21Data Model
- 11:34Prometheus Query Language
- 15:03Prometheus alerting
- 16:27Operational simplicity
- 17:05Scaling
- 17:50Remote Right Protocol
- 18:29Thanos
- 19:42Dynamic environments
- 20:12Service discovery
- 21:40Conclusion
- 21:56Open QA
- 28:42Multitenancy
- 31:41Mathematic Functions
- 33:38Recommended Practices
- 36:21Data Cleanup
- 39:43Scrape Interval
- 42:14Aggregates
- 43:50Writing and scraping
- 46:46Questions
- 48:29Frontend metrics
- 49:43Replicating data
- 50:59Deadman switch
- 52:54Query API
- 55:00Deep Dive QA
