How to check memory usage in python application using scalene?

 



Checking Memory Usage in a Flask Application with Scalene

1. Install and Verify Scalene

  1. Ensure you’re using Python 3.6 or newer.

  2. Install Scalene via pip:

    pip install scalene

  3. Verify the install:

    scalene --version

2. Profile Your Flask App at Launch

If you normally start your app with python app.py, simply prepend scalene:

scalene --profile-interval=10 app.py

– --profile-interval=10 tells Scalene to dump a new profile report every ten seconds
– By default, each dump shows per-line memory use, Python vs. native time, and system time

3. Profile via Flask’s CLI

When you run your app with flask run, invoke Python under Scalene:

scalene --profile-interval=10 -- python -m flask run

– Everything the Flask CLI does (werkzeug server, your code, extensions) will get sampled
– Use --reduced-profile if you only care about total memory allocations and want a cleaner view

4. Attach to a Running Flask Process

If your app is already running (e.g., in a Docker container or systemd), find its PID:

ps aux | grep flask

Then tell Scalene to sample that process:

scalene --profile-interval=5 --pid=12345

Scalene will print a fresh profile every five seconds without restarting your server.

5. Interpreting the Output

Each report shows columns per source line:

  • Time % Python: time spent in your Python code
  • Time % native: time in C extensions or standard-library modules
  • Memory (MB): net memory allocated (green: low, yellow: moderate, red: high)
  • Sys %: time the OS spent doing other tasks

Use the coloration and rates to spot “hot” lines that steadily allocate memory.

Beyond Basic Memory Profiling

  • Combine Scalene with CPU profiling flags (--cpu-only, --native) to focus on performance bottlenecks.
  • Output an interactive HTML report with --outfile=profile.html and open it in a browser.
  • For long-lived services, adjust --profile-interval to balance detail vs. overhead.
  • Integrate with your CI pipeline: run short-interval Scalene profiles on PRs to catch regressions automatically.

Comments

Post a Comment