Effortless Profiling on Kubernetes
Talkat KubeCon + CloudNativeCon North America 2021 · on Wednesday, 2021-10-13· at 17:25· Room 408 AB + Online·
summary
Tool demonstrates live, no-restart profiling on Kubernetes by launching a sidecar profiler pod that shares PID and filesystem and copies out flamegraphs. It chooses profilers per language: async-profiler for JVM, eBPF for Go, py-spy for Python, rbspy for Ruby; captures in 10–30s and returns SVG/HTML results. Demo profiles a Java ad service and a Python recommendation service on minikube without image rebuilds. Roadmap items: Kubernetes F-Containers, eBPF improvements and continuous profiling systems.
AI-written summary · about
chapters (9)
- 0:00Start
- 0:41Introduction to What Is Profiling
- 1:41Flame Graph
- 2:31Why Profiling Is So Hard
- 7:33Demo
- 7:51Modify the Container Images To Include Profiler
- 12:51Recap
- 13:27Execute into the Container and Invoke the Profiler
- 19:17Continuous Profiling
related talks
01h 02mProfiling Node Applications
29mProfiling Node Applications
01h 12mA New Way to Profile Async Activity in Node.js
23mOpenTelemetry Auto Instrumentation for Effortless Observability
10mLightning Talk
22mEmbedded Software Development Efficiency Improvement with Docker and Kubernetes
23mDig Deep V8 profiling
37mIntroduction to SpinKube