This is a simple example app showcasing how to monitor a Django app with Prometheus, when app is running behind UWSGI and Nginx. Grafana is used for dashboard. The whole stack is running on Docker.

• Docker
• Docker Compose
• K6 (optional, required for load testing)

In the project root, run the following command. --build flag is optional, but might be required if you make changes to the project between runs:

docker-compose up [--build]

This will start five Docker containers:

You can stop the stack with Ctrl+C.

Apps are now available in the following ports:

By default, Grafana does not have any dashboards but you can use the provided example.

• Add a new Datasource. Call it Prometheus and use URL http://prometheus:9090
• Go back to datasource configuration and get the data source ID from the URL
• Create a new empty dashboard
• Go to Dashboard settings -> JSON Model and paste the content from dashboard.json
• Run find-and-replace and replace Aa93neR4k with your Prometheus data source ID.

Project is using the standard Nginx -> uWSGI -> Django stack. This is explained in more detail in the uWSGI documentation.

Prometheus is running as a standalone app. It works by pulling metrics from different places - called targets. Targets are just HTTP endpoints (or sockets) that provide the metrics in specific format.

In this project there are two targets. Django application provides the metrics in http://localhost:8000/metrics. For this, django-prometheus package is used. It makes it very easy to export HTTP, database and model metrics from the Django app.

Another target is uwsgi-exporter container, which is running in http://localhost:9117/metrics. uWSGI does have it's own metrics endpoint, but it is not compatible with Prometheus so uwsgi-exporter helps with that by working between uWSGI and Prometheus. This way we can get metrics from uWSGI workers.

Prometheus provides some dashboard capabilities, but to make really fancy dashboards, Grafana is used. Basically, Grafana is fetching the data from Prometheus and visualizes them.

Django-prometheus package makes it really easy to get metrics from the database and from models. Everything is explained in the README but here is the short summary.

To get database metrics, all you need to do is to change the ENGINE in the Django app configuration. For example, if you are using django.db.backends.sqlite3 replace it with django_prometheus.db.backends.sqlite3. Now database metrics are available.

For models, you need to make your models to extends ExportModelOperationsMixin("model_name") class. model_name is the name that is used in the metrics to identify the model. Check Dog and Cat for an example.

This project also includes a load testing example with K6.

To run the test, first run docker-compose in the project root:

docker-compose up [--build]

Then, in another terminal go to k6 directory and run the following command

k6 run --vus 1 --duration 5s load_test.js

This will run a 5 second load testing using just one virtual user. If you want to bombard the system properly, you can use a load test that will increase virtual users ever 15 seconds until there are 160 users.

k6 run --vus 5 --stage 15s:10,15s:20,15s:40,15s:80,2m:160 load_test.js

This will quite likely make your local stack to choke, but it is also a good way to see the changes in the dashboard.