Continuous Profiling in the Cloud with eBPF
summary
Describes building continuous production profiling across a cloud fleet at Odnoklassniki to be able to retrieve a CPU/profile for any app at any past time. Covers difficulties of rolling out continuous profilers (you can’t just run async-profiler everywhere), the use of eBPF as the core technology in the solution, and the infrastructure for collecting, storing and visualizing profiles, ending with a demo of the user-facing result.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 12:32Continuous profiling problem
- 27:18eBPF technology overview
- 37:19Profile visualization with венгров
- 51:04System issues and optimizations
- 1:00:05Results and production deployment
- 1:06:12Conclusion
related talks
34mBeyond the Hype: Cloud Native eBPF
37meBPF: Revolutionizing Observability and Telemetry for Modern Applications
32mDeveloping eBPF profiler for polyglot cloud-native appli...
27mLow-Overhead Tracing Using eBPF for Observability into Kubernetes Apps and Services
29mBuzzing Across the Cloud Native Landscape with eBPF
36mContinuous Profiling with Parca- Beneath the Kernel to Beyond...
31mZero instrumentation observability with eBPF
36mObservability and performance analysis with BPF