Reveal Your Deepest Kubernetes Metrics - Bob Cotton, Freshtracks.io
summary
Kubernetes exposes thousands of metrics and this session narrows which ones matter by applying the USE and RED methods and by showing Prometheus queries. It lists sources: node_exporter for node CPU/memory, cAdvisor for per-container CPU/memory/throttling and IO, API server and etcd request/rate/latency metrics, and recording rules that aggregate container series up to pod/deployment/namespace/cluster. Examples include CPU throttling counters used to detect bad CPU limits and etcd leader/commit latency to flag storage problems.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:28About Me
- 2:41Four Golden Signals
- 3:40USE Method
- 6:04Node Metrics from node exporter
- 7:02USE for Node CPU
- 9:45USE for Node Memory
- 12:31Container Metrics from Advisor
- 13:57USE for Container CPU
- 15:44USE for Container Memory
- 21:50RED for Kubernetes API Server
- 26:32Core Metrics Aggregation
related talks
23mLearning Kubernetes Through the Lens of Metrics
35mDeep Dive: Kubernetes (Instrumentation) SIG
29mKubernetes-Defined Monitoring [I]
08mTalk Deep Dive
19mKeynote: Monitoring Kubernetes Clusters with Prometheus
16mShrinking the Observability Bill
30mCrash course in Kubernetes monitoring
32mUnderstanding Observability in Kubernetes