The RED Method: How To Instrument Your Services [B]
summary
Defines RED = requests, errors, duration as the three service metrics to record, uses a single Prometheus histogram with labels (method, route pattern, status) plus middleware and HTTP-snoop to collect them, and warns against high-cardinality raw paths. Demonstrates queries: a dev endpoint at ~36 QPS and a 90th-percentile latency around 6 ms. Publishes generated Grafana dashboards called Klumps (JSONnet) for cluster- and service-level views. Notes gaps: saturation and hardware error counters are harder to measure.
AI-written summary · about
chapters (21)
- 0:00Introduction
- 2:40Use Method
- 3:54Checklist Approach
- 4:44CPU Utilization
- 6:40Memory
- 11:06Read Method
- 13:32Prometheus
- 14:13HTTP snoop
- 16:33Rate of requests
- 20:39Microservice architecture
- 21:01Dashboards
- 21:44Averages
- 23:00Logs
- 23:21Golden Signals
- 23:36Saturation
- 23:51Full Services
- 24:05Resource Requests
- 29:27Recap
- 29:54Other Methods
- 31:01Microphone
- 31:13Battery
