flask-monitor

A Prometheus middleware to add basic but very useful metrics for your Python Flask app.

Metrics

As valid Big Brother library it exposes the following metrics:

request_seconds_bucket{type, status, isError, method, addr, le}
request_seconds_count{type, status, isError, method, addr}
request_seconds_sum{type, status, isError, method, addr}
response_size_bytes{type, status, isError, method, addr}
dependency_up{name}
application_info{version}

In detail:

1. The request_seconds_bucket metric defines the histogram of how many requests are falling into the well defined buckets represented by the label le;

2. The request_seconds_count is a counter that counts the overall number of requests with those exact label occurrences;

3. The request_seconds_sum is a counter that counts the overall sum of how long the requests with those exact label occurrences are taking;

4. The response_size_bytes is a counter that computes how much data is being sent back to the user for a given request type. It captures the response size from the content-length response header. If there is no such header, the value exposed as metric will be zero;

5. The dependency_up is a metric to register weather a specific dependency is up (1) or down (0). The label name registers the dependency name;

6. Finally, application_info holds static info of an application, such as it's semantic version number;

Labels

For a specific request:

1. type tells which request protocol was used (e.g. grpc, http, <your custom protocol>);
2. status registers the response status code;
3. isError let you know if the request's response status is considered an error;
4. method registers the request method (e.g. GET for http get requests);
5. addr registers the requested endpoint address;
6. and version tells which version of your service has handled the request;

How to

Add this package as a dependency:

pip install bb-flask-monitor

or

pipenv install bb-flask-monitor

HTTP Metrics

Use it as middleware:

from flask import Flask
from prometheus_client import make_wsgi_app
from werkzeug.middleware.dispatcher import DispatcherMiddleware
from werkzeug.serving import run_simple
from flask_monitor import register_metrics

app = Flask(__name__)
app.config["APP_VERSION"] = "v0.1.2"

register_metrics(app)
# Plug metrics WSGI app to your main app with dispatcher
dispatcher = DispatcherMiddleware(app.wsgi_app, {"/metrics": make_wsgi_app()})

One can optionally define the buckets of observation for the request_second histogram by doing:

register_metrics(app, buckets=[0.1]); // where only one bucket (of 100ms) will be given as output in the /metrics endpoint

Other optional parameters are also:

1. error_fn: an error callback to define what you consider as error. 4** and 5** considered as errors by default;

Dependency Metrics

For you to know when a dependency is up or down, just provide a health check callback to be watch_dependencies function:

app = Flask(__name__)
app.config["APP_VERSION"] = "v0.1.2"

register_metrics(app)
# Plug metrics WSGI app to your main app with dispatcher
dispatcher = DispatcherMiddleware(app.wsgi_app, {"/metrics": make_wsgi_app()})

def check_db():
    try:
        response = req.get("http://localhost:7000/bd")
        if response.status_code == 200:
            return 1
    except:
        traceback.print_stack()
    return 0

watch_dependencies("Bd", check_db, app=app)

Other optional parameters are also:

watch_dependencies has the following parameters:

1. dependency: the name of the dependency;

2. func: the health check callback function;

3. time_execution: the interval time in seconds that func will be called.

Now run your app and point prometheus to the defined metrics endpoint of your server.

More details on how Prometheus works, you can find it here.

Example

In the example folder, you'll find a very simple but useful example to get you started. On your terminal, navigate to the project's root folder and type:

cd example
pipenv install

and then

python app.py

On your browser, go to localhost:5000 and then go to localhost:5000/metrics to see the exposed metrics.

Big Brother

This is part of a more large application called Big Brother.