Profiling Java — Flamegraph Link to heading

Flamegraph is a terrific visualization for profiling a program. Unfortunately, it only supports native apps (such as compiled binaries)…

Profiling Java — Flamegraph Link to heading

Flamegraph is a terrific visualization for profiling a program. Unfortunately, it only supports native apps (such as compiled binaries) and do not support apps running on VMs, such as Java or C#.

In this article, let’s take a look at how to obtain flamegraph from a Java app. First, clone async-profiler

$ git clone https://github.com/async-profiler/async-profiler

Then, we need to install debug symbols if you are using OpenJDK. On Ubuntu, you can run

$ sudo apt install -y openjdk-11-dbg

To verify the tool has been successfully installed, you can verify with

$ gdb $JAVA_HOME/lib/server/libjvm.so -ex 'info address UseG1GC'

where $JAVA_HOME points to the OpenJDK java directory, e.g., /usr/lib/jvm/java-11-openjdk-amd64. If you see a message Symbol "UseG1GC" is static storage at address …, then you are good to proceed.

Next, we can build it

$ make -j

If the build is successful, you should see a file ./build/lib/libasyncProfiler.so. With this, we can now profile your app by adding the java option

-agentpath:/path/to/async-profiler/build/lib/libasyncProfiler.so=start,event=cpu,file=profile.html

For example, the following command

$ java -cp /path/to/classes -agentpath:/path/to/async-profiler/build/lib/libasyncProfiler.so=start,event=cpu,file=profile.html my_awesome_app

will profile my_awesome_app and generate profile.html file, which can be opened with a browser and you will see a flamegraph!