Prometheus with Molten

molten has built-in support for exporting the following Prometheus metrics:

• http_request_duration_seconds{method,path} -- a histogram of the request duration percentiles by request method and path,
• http_requests_total{method,path,status} -- a counter of the total number of requests by method, path and status,
• and http_requests_inprogress{method,path} -- a gauge of the number of requests in progress by method and path.

Let's say that you have a basic molten app. Something like this:

from molten import App, Route


def index():
    return {}

app = App(
    routes=[
        Route("/", index),
    ],
)

To start tracking metrics, add prometheus_middleware to your app's middlewares list:

from molten import App, ResponseRendererMiddleware, Route
from molten.contrib.prometheus import prometheus_middleware


def index():
    return {}

app = App(
    middleware=[
        prometheus_middleware,
        ResponseRendererMiddleware()
    ],
    routes=[
        Route("/", index),
    ],
)

The app will then begin keeping track of metrics in memory, but won't expose them anywhere. Prometheus uses a pull-based model to gather metrics from servers, which means you have tell it where (i.e. which servers/ports) to look for metrics and your server needs to be able to expose those metrics in a way that Prometheus can understand.

To start exposing metrics, add expose_metrics to your routes:

from molten import App, ResponseRendererMiddleware, Route
from molten.contrib.prometheus import expose_metrics, prometheus_middleware


def index():
    return {}

app = App(
    middleware=[
        prometheus_middleware,
        ResponseRendererMiddleware()
    ],
    routes=[
        Route("/", index),
        Route("/metrics", expose_metrics),
    ],
)

If you run your app now and visit the index handler a couple of times and then visit /metrics, you should see all the metrics that were gathered up until that point.

One caveat related to expose_metrics is it should only be used in single-process configurations. If your application uses multiple processes to server requests (for example, multiple gunicorn workers) then you should use expose_metrics_multiprocess instead.

Assuming you're running the app on localhost:8000, you can then use a scrape config such as this one to let Prometheus know where to look for metrics:

scrape_configs:
  - job_name: 'a-molten-app'
    scrape_interval: 15s
    static_configs:
      - targets: ['localhost:8000']

Save that to a file called prometheus.yml, run prometheus and then visit http://localhost:9090 and you should be able to start querying the metrics I mentioned above. Check out Prometheus' querying documentation to find out how to construct queries.

While Prometheus' dashboard is nice for inspecting metrics, it's not meant for long-lived graphs and dashboards. For that, you should take a look at Grafana.