Flame Graphs for Python Examples

This repo includes a Dockerfile and examples showing how to get flame graphs for Python code and accompanies this post that goes into more depth.

Quickstart

If you have docker installed, run the following command:

./build_and_run.sh

Which will build an image with the right tools and generate an example flame graph. If you want to play inside the container, just override the entrypoint:

docker run --rm -it --entrypoint=/bin/bash -v `pwd`:/host $TAG

If you don't have docker and are running on Ubuntu or Debian, you can mimic the Dockerfile and install the tools on your host:

apt-get install -y git wget curl vim build-essential python3 python3-pip python-is-python3
git clone --depth=1 https://github.com/brendangregg/FlameGraph
pip install flameprof 2>/dev/null

Then use any of the make_*.sh scripts to start making flame graphs.

Scripts

There are a few scripts demonstrating generating a flame graph in different ways:

1. make_flameprof_flamegraphs.sh: Generic script to make a flame graph based on the specified python file
2. make_self_profiled_flamegraph.sh: A script showing how to do the same thing when the script contains code invoking cProfile (as in example.py)
3. make_inferno_flamegraphs.sh: Generic script that uses inferno instead of the classic FlameGraph tools.

make_all.sh invokes all of them and shows how to use them, but it takes longer to complete.

Example Targets

Included are three examples that show different kinds of Python programs where profiling is helpful:

1. basic-example/example.py: A simple example to demonstrate flame graphs and the use of cProfile to profile specific parts of the code
2. sqlite-example/test_insert.py: Inserts a large number of random rows in a sqlite3 db to show where the bottleneck really is...
3. aoc-2023-19/a.py: An example solution to a specific Advent of Code challenge that takes a while.
4. aoc-2023-19/a-opt.py: The same file after a simple optimization made based on looking at the flame graph (runs about 2x faster).

Thanks!

I hope this helps, feel free to open an issue if there's something you want to see, or follow me on GitHub or Twitter for more.