spring-jfr

Production-safe JVM profiling uses JFR to collect low-overhead diagnostic recordings on live Java services. You configure always-on continuous recordings with ring buffers and dumponexit to capture incident data without significant additional load or downtime.

To capture CPU hotspots and allocation pressure, start targeted deep-dive JFR recordings via jcmd or RecordingStream. You then analyze the dumped JFR files and generate precise CPU and allocation flamegraphs using Async Profiler for detailed inspection.

JFR profiling works with Java 21 Spring Boot applications running ZGC with virtual threads. It specifically diagnoses CPU hotspots, allocation pressure, GC behavior, lock contention, and virtual-thread pinning during incidents and performance regressions.

The best way to analyze JFR files for lock contention and GC behavior is using JMC. You copy the JFR recording artifacts out of your service, such as from Kubernetes pods, and use JMC alongside Async Profiler flamegraphs for comprehensive forensic analysis.

Real-time monitoring of custom JFR events in Spring Boot involves emitting custom JFR events and streaming them via RecordingStream. This enables anomaly detection, slow-method alerts, and contention logging directly during application runtime.

Safe pod-level Kubernetes workflows require restricted actuator exposure and specific heapdump policies. You control recording depth with stackdepth configurations and use jcmd to trigger safe recordings, preventing excessive load during forensic data capture.