/simple-api

Primary LanguagePythonMIT LicenseMIT

Simple FastAPI service

Minimal reproducible example to showcase Issue #1624

The memory usage piles up over the time and leads to OOM

Requirements

  • Python 3.10.8
  • Poetry
  • Docker

Setup

  • building the wheel within a py 3.10 docker container
    (Note: you can skip this step, wheel available in repo) 
    docker run --rm -v $PWD:/repo -w /repo -it python:3.10 bash
pip install poetry==1.2.2
poetry build -f wheel
  • building the docker image 
    docker build -t simple_api .
  • starting up the container 
    docker run --rm -p 8080:8080 --name simple_api simple_api
  • keep an eye on docker memory usage 
    docker stats
  • send random sized requests to the service
    • only valid input 
      ./send_ok_requests.sh
    • mixed valid and invalid input 
      ./send_mixed_requests.sh

Note: the memory increase is most visible on invalid requests

Rule out other possible culprits

Check only the logic + pydantic validation

  • building the docker image 
    docker build -f Dockerfile.check_validation -t simple_api_check_validation .
  • starting up the container 
    docker run --rm --name simple_api_check_validation simple_api_check_validation
  • keep an eye on docker memory usage 
    docker stats

Note: don't see any change here

Check only the api without uvicorn server

  • building the docker image 
    docker build -f Dockerfile.check_client -t simple_api_check_client .
  • starting up the container 
    docker run --rm --name simple_api_check_client simple_api_check_client
  • keep an eye on docker memory usage 
    docker stats

Note:

  • still see the memory increase here
  • trying to kill the interactive container with CTRL+C will keep generating new PIDs, stop it with docker stop <container_id>

Track memory increse over time

  • starting the server 
    docker run --rm -p 8080:8080 -v $PWD:/repo -w /repo -it python:3.10 bash
pip install dist/*.whl
pip install memory_profiler
mprof run simple_api
  • send random sized requests to the service 
    ./send_mixed_requests.sh
  • stop the server and plot the memory increase 
    mprof plot

Sample plot: mprof_plot.png