Deep Dive: Kubernetes (Instrumentation) SIG
summary
kube-state-metrics switches from caching Kubernetes objects to cache Prometheus metrics and cuts scrape tail latency that was 10–22 seconds on a 250-node test cluster down to a stable low value across runs; gzip raises CPU and memory so the project chooses avoid gzip by default after testing stdlib and nytimes compressors; other wins come from strings.Builder and byte slices, Go pprof/benchmarks, client-go informers, CI benchmark comparisons, and releases v1.5/v1.6.
AI-written summary · about
chapters (9)
- 0:00Start
- 3:00What is kube-state-metrics? Old version.
- 6:52Problem
- 12:01Performance improvement through caching
- 13:39Compression
- 19:11Hard code float-to-string common cases
- 23:26Memory pre-allocation
- 28:28Future
- 30:41Take aways
related talks
26mIntro and Deep Dive: Kubernetes SIG Instrumentation
41mSIG Instrumentation Deep Dive
28mIntro and Deep Dive: Kubernetes SIG Instrumentation
40mKubernetes SIG Instrumentation - Deep Dive
28mKubernetes SIG Instrumentation - Intro
37mSIG Instrumentation Intro
36mReveal Your Deepest Kubernetes Metrics - Bob Cotton, Freshtracks.io
36mDon't Be Greedy: Rightsize Your Kubernetes Cluster With Prometheus