django-channels-leak

This project is a full copy of django_channels tutorial from the official doc: https://channels.readthedocs.io/en/stable/tutorial/index.html The only changes are some commented out debug statements for memory tracing.

Deploy locally

Prerequisites

You will need a backend for channels_layer - Redis or Rabbitmq. Repo includes settings and configuration for both. All the dependencies can be installed using Poetry, they also include debug libraries like Pympler.

Setup

• Activate your local environemnt;
• Go to <project_root>/channels_leak;
• Run daphne channels_leak.asgi:application;

Steps to reproduce

• Run server locally;
• Identify PID and remember the RAM consumption before any tab is open;
• Open multiple tabs of a single chat room, like http://127.0.0.1:8000/chat/hello/ (approximately ten);
• Close all the tabs;
• Check RAM consumption. It has not been reduced, or has been reduced only by ~half since the peak amount.

Sample metrics from local debug

• Server is launched, no browser tabs open: 49.3 MB;
• First broser tab opened: 51.2 MB;
• 20 browser tabs opened: 52.2 MB;
• All browser tabs closed: 52.2 MB;
• 20 browser tabs opened: 52.6 MB;
• All browser tabs closed: 52.6 MB;

Possible locations of memory leak

Consumer instance has channels_layer property, and it looks like it leaks whatever the backend is. Both for Redis and Rabbitmq the commection map is growing larger after each tab opened.

This object's memory usage is reduced after closing tabs, but it still consumes more memory (+50-100 KB) after opening and closing multiple tabs.