Enabling Code Profiler for Python on App Service Linux

2 minute read | By Kendrick Dubuisson

The App Service Linux Python images now come pre-configured with a code-profiler. This quickly allows us to collect traces for troubleshooting High Memory & High CPU against our running Python process without installing any additional packages within your application manually.

Once enabled & run, it’ll output our trace as a JSON file directly onto the file system for us to download. Additionally, we will use VizTracer to review the timeline & visualization of execution/utilization while providing insights into the active call stack.

This blog will walk through enabling the profiler on App Service Linux & reviewing locally.

Enabling & Gathering a trace on App Service Linux

  1. The code profiler is currently an opt-out feature, meaning it is enabled by default. There is no need to add any additional app settings.

  2. Validate the running process to profile. In this example, we’re using gunicorn running in forked mode so it’s expected to see multiple processes, but we can use variations of ps to target the process-id consuming cpu time. In general the second Gunicorn process serves the request and should be profiled. This same approach can be used if other python servers are used to serve your application.

     #checking for the running python process id
     ps -aux
     ps -C gunicorn fc -o ppid,pid,cmd

    Gunicorn Process

  3. Use the following command to start the profile. Upon completion additional steps will be provided in the console that we will also cover in our next steps.

     code-profiler --attach <PID> -t <Number of seconds to profile>

    Gunicorn Process

  4. Navigate to file manager of Kudu site of the App Service & validate profiler output. This can be downloaded directly from the File manager to review locally.

    • Kudu New UI: https://<site-name>.scm.azurewebsites.net/newui/fileManager
    • Default Trace Location: /home/LogFiles/CodeProfiler/profiler_trace.json

    NOTE: If there is no profiler_trace.json present after running the code-profiler, this can be caused if the incorrect PID is choosen.

Reviewing the trace locally with VizTrace

VizTracer is a low-overhead logging/debugging/profiling tool that can trace and visualize your python code execution. The front-end UI is powered by Perfetto.

  1. Ensure Python is installed in your Linux machine / Windows Subsystem for Linux (WSL)

  2. Run the following command to install viztracer: pip install viztracer
  3. On your local machine (Linux / WSL), run the following command
    vizviewer --port 8080 profiler_trace.json
  4. This will open vizviewer & your trace. Below we can see the my “secondMethod” function was called which also consumed CPU visible in the timelime.
    1. Timeline
    2. Process Graph
    3. CPU & Memory Utilization
    4. Function Selection Gunicorn Process