gcavalcante8808/gunicorn_cpu_profiler repository overview

⁠Gunicorn Cpu Monitor

This is a simples project that aims to provide an ondemand cpu_profiler for gunicorn worker processes and its threads by using Uber Pyflame⁠ tools.

⁠How To use

Just spin a container pointing to the gunicorn container that will be profiled:

    docker run -it --rm -e FLAMECHARTS_FOLDER=/charts -v $(pwd)/charts:/charts --pid="container:gunicorn_container" gcavalcante8808/gunicorn_cpu_monitor

After a few seconds, some files with pattern "/<FLAMECHARTS_FOLDER>/proccess_.profile" will be dumped into the volume mounted on FLAMECHARTS_FOLDER.

These files can be viewed with Chrome Dev Tools Performance analizer (https://developers.google.com/web/tools/chrome-devtools/evaluate-performance⁠).

⁠Using with Docker-compose

With a working docker-compose.yaml file, you can just create an docker-compose.override.yaml with the following statements:

version: <SAME_VERSION_OF_DOCKER_COMPOSE_DOT_YAML>

services:

  cpu_monitor:
    image: gcavalcante8808/gunicorn_cpu_profiler
    pid: "service:<GUNICORN_COMPOSE_SERVICE>"
    command: ["/usr/local/bin/dumb-init","python","gunicorn_cpu_monitor.py"]
    cap_add:
      - SYS_PTRACE
    environment:
      FLAMECHARTS_FOLDER: /charts
    volumes:
     - /my-folder:/charts
    depends_on:
      - <GUNICORN_COMPOSE_SERVICE>

Update <GUNICORN_COMPOSE_SERVICE> to point to your gunicorn service and <SAME_VERSION_OF_DOCKER_COMPOSE_DOT_YAML> to use the same version of your docker-compose.yaml original file, otherwise docker-compose will return errors about the version.

For more information about docker-compose file override, take a look at https://docs.docker.com/compose/extends.

⁠Settings

The following environment variables can be used to tune or change program`s behaviour:

• FLAMECHARTS_FOLDER: Folder where flamecharts json files will be created;
• CPU_THRESHOLD: By Default, all processes that have more than 90 percent usage of a processor are profilled;
• CPU_READ_INTERVAL: By Default, pyflame will read sequentially 5 seconds of cpu use for each subprocess/worker for gunicorn. If you change this interval, be sure to check if the need time to do all reads (subprocess * CPU_READ_INTERVAL is lesser than SCAN_INTERVAL);
• SCAN_INTERVAL: By Default, the scan for threads will occur every 60 seconds;
• GUNICORN_PARENT_PID: By Default, the cpu_monitor will try to use the first pid that is a child from pid 1. This is a behaviour when you are using an supervisor or an minimal init system to initialize your gunicorn instance.

⁠Author

Author: Gabriel Abdalla Cavalcante Silva ([email protected]⁠)