Profiling Without Pain: Always-On Performance for the JVM

You did all the right things. The latest JVM, a fancy modern framework (Quarkus, naturally), and domain driven design. And it works - it’s just slow. At least, your colleague Ted thinks so - his overly-detailed bug reports leave little to the imagination - and the metrics from prod certainly don’t lie.

You’ve tried everything to make Ted happy again - tweaks in dev, careful microbenchmarks, hours lost to the profiler. You’ve even consulted the hallowed garbage collection tuning guides. And still, the issue won’t reveal itself. Enter continuous profiling - a tool enabled in part by the Java Flight Recorder - a way of capturing and actioning profiling data with low overhead in production, and correlating it back to all your important request metadata.

In this talk you will learn how continuous profiling can help us diagnose and resolve performance issues by tying our profiles back to actual traffic, how it works under the hood, and how it differs from traditional profiling. We’ll dive deep into the JFR - the Java subsystem that underpins Java's modern profiling support - and see how we can use open-source tools to bridge the gap from JFR to production, always-on profiling.

By the end, you’ll know how to enable, interpret, and operationalise continuous profiling in your own JVM applications - turning “it’s kinda slow sometimes” into something you can actually action.