A conference talk from Oracle engineers demonstrating an intelligent JVM monitoring system that combines JDK Flight Recorder (JFR), Java Management Extensions (JMX), and AI. The system uses a Java agent to stream JFR events from microservices to a central monitoring service, which uses LangChain4j with Claude to analyze performance data and autonomously trigger corrective actions (like enabling batch processing or clearing caches) via JMX. Key techniques covered include creating JFR recording agents with custom event configurations, implementing dynamic MBeans with metadata for AI-discoverable actions, prompt engineering with system/user prompt separation and chain-of-thought reasoning, LangChain4j AI Services with tool/function calling, and confidence-threshold gating with human-in-the-loop for critical actions.
Sort: