Observability Beyond the Three Pillars — Continuous Profiling With Alibaba Dragonwell
summary
Shows a production-grade continuous profiler built on JFR that runs via Java agent collectors and stores sampled profiles for aggregation, query and flame graphs. Claims CPU/allocation/lock profiling with end-to-end correlation by appending trace/span ID in thread-local context to JFR events so blind spots link to traces. Measures overhead on SPECjbb2015 dropping from >15% to ~2% when using the HotSpot memory-fence flag, warns allocation and old-object sample costs on older JDKs, and demos finding a Spring split hotspot and deoptimization sites.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:52Basics of observability
- 6:54Continuous profiling explained
- 10:34Java Flight Recorder basics
- 16:46Alibaba Dragonwell introduction
- 23:11Correlating tracing and profiling
- 30:26Lessons learned from implementation
- 33:00Demo: Object allocation profiling
- 44:02Demo: Deoptimization profiling
- 48:49Summary and Q&A
related talks
56mContinuous Profiling the missing piece in your observability puzzle
41mUnleash the Power of Open-Source Java Profilers
42mProfiling and monitoring with the JDK
46mContinuous Profiling - the 4th Pillar of Observability
52mThe missing piece in your observability puzzle!
45mJava Performance Analysis and Optimization
33mWriting a Java Profiler in Pure Java
41mA glimpse into Alibaba Drogonwell: Towards a Java Runtime for Cloud Computing