nesquivelr / compare_python_api_frameworks

Code examples and resources from the blog post on https://medium.com/@n.esquivelrivera/apis-using-grpc-and-rest-b57f811da319

Comparing sending one message in GRPC, Flask and FastAPI

The idea of this test is to know which of these frameworks can handle a message faster.

Disclaimer: This test may be completely wrong as each server uses default values, it could be the case that the default value of workers on a server is greater than the number on a different server.

Create environment and install libraries

conda create -n test_frameworks python=3.12
conda activate test_frameworks
pip install -r requirements.txt

Tests

These tests were performed on a machine with AMD Ryzen 9 7900 and 32GB of RAM.

GRPC

Run python server.py in one terminal and in another run:

ipython
from client import run
%timeit -n 1000 run()
1.2 ms ± 9.87 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

Flask

Run python server.py in one terminal and in another run:

ipython
from client import run
%timeit -n 1000 run()
673 µs ± 77.6 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

FastAPI

Run python server.py in one terminal and in another run:

ipython
from client import run
%timeit -n 1000 run()
735 µs ± 4.36 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

Improvement ideas

• Automate the tests
• Create response time histogram.

Useful commands when developing

Compile proto file

python -m grpc_tools.protoc locus_message.proto --proto_path=. --python_out=. --pyi_out=. --grpc_python_out=.