Intelligent JVM Monitoring: Combining JDK Flight Recorder with AI
summary
A Java agent starts JDK Flight Recorder with a custom JFC and streams ~90 default events (pruned to a custom set) to a central service, which stores events in a database and runs anomaly detection. The system discovers controllable settings via JMX MBeans (dynamic descriptors carrying action names and confirmation flags), builds prompts and calls an LLM through LangChain4j (Anthropic Hayo 4.5) which returns JSON reasoning plus a tool call; a 70% confidence action is rejected, a 90% run toggles batch mode successfully.
AI-written summary · about
chapters (26)
- 0:00Intelligent JVM monitoring intro
- 2:20Galaxy Cafe microservices
- 3:45Manual JFR monitoring issues
- 4:54Centralized intelligent monitoring proposal
- 5:16Java agent for JFR recording
- 6:21Low-overhead JFR event config
- 8:52Custom JFR event filtering
- 9:56JMX for runtime control intro
- 10:40JMX concepts and MBeans
- 12:06Self-healing MBeans for microservices
- 15:20JMX for Galaxy Cafe
- 16:45Self-healing MBeans for batch
- 19:36Remote MBean discovery
- 20:44JFR data with AI integration
- 21:46JFR anomaly detection with AI
- 22:30LLM prompt design
- 25:24MBeans metadata for AI actions
- 27:55JConsole MBean metadata
- 28:59LLM output format rules
- 30:24AI decisions via tool calling
- 31:50JFR data preprocessing
- 34:24AI services with LangChain4j
- 37:42AI decisions via JMX client
- 39:06Full monitoring workflow
- 41:35Summary of solution
- 42:38Q&A
related talks
50mIntelligent JVM Monitoring: Combining JDK Flight Recorder with AI
46mContinuous Monitoring with JDK Flight Recorder
39mProgrammer's Guide to JDK Flight Recorder
31mJava Mission Control: To Infinity and Beyond…
51mThe Power of JDK Flight Recorder: Efficient Profiling and Troubleshooting for Java Applications
39mProfiling and Monitoring with the JDK
56mTaking Off With #JDK #FlightRecorder
42mProfiling and monitoring with the JDK