https://github.com/parttimenerd/tiny-profiler
A tiny CPU profiler for Java written completely in Java 17.
https://github.com/parttimenerd/tiny-profiler
java profiler
Last synced: 4 months ago
JSON representation
A tiny CPU profiler for Java written completely in Java 17.
- Host: GitHub
- URL: https://github.com/parttimenerd/tiny-profiler
- Owner: parttimenerd
- License: other
- Created: 2023-03-26T10:23:12.000Z (almost 3 years ago)
- Default Branch: main
- Last Pushed: 2023-11-27T08:40:49.000Z (about 2 years ago)
- Last Synced: 2025-06-30T01:38:01.353Z (7 months ago)
- Topics: java, profiler
- Language: Java
- Homepage: https://mostlynerdless.de/blog/2023/03/27/writing-a-profiler-in-240-lines-of-pure-java/
- Size: 925 KB
- Stars: 92
- Watchers: 4
- Forks: 19
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
Tiny Profiler
=============
A tiny CPU profiler for Java written completely in Java 17.
This is an educational sampling profiler that helps demystify profilers, but is not limited to it.
This profiler is a sampling profiler with can output method tables and
flame graphs (via [d3-flame-graph](https://github.com/spiermar/d3-flame-graph)).
The sampling is based on the `Thread.getAllStackTraces()` method, making it simple but also safepoint-biased.
The profiler should work on any platform with a JDK 17 or newer ([JDK 11 version](https://github.com/StSchnell/tiny-profiler) by Stefan Schnell). The usage is fairly simple:
```sh
# build it
mvn package
# run your program and print the table of methods sorted by their sample count
# and the flame graph, taking a sample every 10ms
java -javaagent:target/tiny-profiler.jar=flamegraph=flame.html ...
```
Example:
Profiling the included fibonacci benchmark:
```sh
> java -javaagent:target/tiny-profiler.jar=flamegraph=flames.html Fib.java 40
===== method table ======
Total samples: 54
Method Samples Percentage On top Percentage
com.sun.tools.javac.launcher.Main.run 53 98,15 0 0,00
com.sun.tools.javac.launcher.Main.main 53 98,15 0 0,00
com.sun.tools.javac.launcher.Main.compile 28 51,85 0 0,00
java.lang.reflect.Method.invoke 25 46,30 0 0,00
```
Profiling the [Renaissance](https://renaissance.dev) benchmark suite:
```sh
# download a benchmark
> test -e renaissance.jar || wget https://github.com/renaissance-benchmarks/renaissance/releases/download/v0.14.2/renaissance-gpl-0.14.2.jar -O renaissance.jar
> java -javaagent:./target/tiny-profiler.jar=flamegraph=flames.html -jar renaissance.jar dotty
...
===== method table ======
Total samples: 11217
Method Samples Percentage On top Percentage
dotty.tools.dotc.typer.Typer.typed 59499 530.44 2 0.02
dotty.tools.dotc.typer.Typer.typedUnadapted 31050 276.81 7 0.06
scala.runtime.function.JProcedure1.apply 24283 216.48 13 0.12
dotty.tools.dotc.Driver.process 19012 169.49 0 0.00
dotty.tools.dotc.typer.Typer.typedUnnamed$1 18774 167.37 7 0.06
dotty.tools.dotc.typer.Typer.typedExpr 18072 161.11 0 0.00
scala.collection.immutable.List.foreach 16271 145.06 3 0.03
...
```
This results in the following flamegraph:
The overhead for this example is around 2% on my MacBook Pro 13" for a 10ms interval, which makes the profiler usable
when you ignore the safepoint-bias.
Structure
---------
- Main class: Entry point of the Java agent
- Options: Parses and stores the agent options
- Profiler: Contains the profiling loop
- Store: Stores and outputs the collected results
License
-------
MIT, Copyright 2023 SAP SE or an SAP affiliate company, Johannes Bechberger
and tiny profiler contributors
*This project is a prototype of the [SapMachine](https://sapmachine.io) team
at [SAP SE](https://sap.com)*