Statistical profiler which outputs in format suitable for FlameGraph (
Switch branches/tags
Nothing to show
Clone or download
Latest commit 6b70f90 Sep 28, 2017


A simple statistical profiler which outputs in format suitable for FlameGraph.


Simply run:

pip install git+


Run your script under the profiler:

python -m flamegraph -o perf.log --your-script args here

Or, run the profiler from your script:

flamegraph.start_profile_thread(fd=open("./perf.log", "w"))

Run Brendan Gregg's FlameGraph tool against the output: --title "MyScript CPU" perf.log > perf.svg

Enjoy the output:

Attic create flame graph


Sometimes you may want to exclude a method (for example in a server the method that waits for a new request) or you may want to profile only a subset of your code (a particular method and its children which are performance critical).

Filtering can be done by passing a python regular expression to the -f or --filter command line option which will restrict output to only those lines which match. Filtering is done against the entire line so you can filter by function name, thread name, both, or even by more complex filters such as function ABC calls DEF (ABC.*DEF).

Alternatively since the output is stackframes each on a line by itself, this can simply be done with a simple grep filter.:


  grep -v waiting_method perf.log > removed_waiting.log


  grep function_name perf.log > filtered.log

Then run the script against the filtered file.

yCanta webapp full profile of PDF export

Full profile output of yCanta webapp PDF export. Most time is spent in wait state and graph is not very helpful.

yCanta webapp filtered for PDF export format function.

Filtered profile output of yCanta webapp PDF export. Filtering was on the pdf format function so time spent in wait state has been excluded and the graph is now helpful.