prometheus

Python prometheus library for django and django rest framework. This helps in monitoring the application on a granular level. You can customize which part of the application you want to monitor. Through this you can monitor a REST API, a python function , a code segment.

Usage

Requirements

• Django >= 1.8
• djangorestframework >= 3.0
• prometheus_client >= 0.7.1

Installation

Install with:

pip install prometheus-python

Or, if you're using a development version cloned from this repository:

git clone https://github.com/harshittrivedi78/prometheus.git
python prometheus/setup.py install

This will install Django >= 1.8 and djangorestframework >= 3.0 and prometheus_client as a dependency if not installed already.

Quickstart

In your settings.py:

INSTALLED_APPS = [
   ...
   'prometheus',
   ...
]

In your urls.py:

urlpatterns = [
    ...
    url('', include('prometheus.urls')),
]

In your views.py:

from rest_framework import generics, status
from rest_framework.response import Response
from prometheus import monitor

class TestAPIView(generics.RetrieveAPIView):

    @monitor(app_name="test") # app_name should be unique through out the application.
    def retrieve(self, request, *args, **kwargs):
        data = {}
        return Response(data, status=status.HTTP_200_OK)

So as you can see in the above example I have decorated the retrieve function by our monitor decorator which will provide monitoring metrics for this function only. And you can identify how much time this function is taking to execute, how many requests are in progress currently, how many request totally served till now.

Metrics are exposed to:

http://localhost:8000/metrics

Default list of monitored metrics

* request_count
* request_latency
* request_in_progress
* response_by_status_total

Configuration

Prometheus uses Histogram based grouping for monitoring latencies. The default buckets are here: https://github.com/prometheus/client_python/blob/master/prometheus_client/core.py

You can define custom buckets for latency, adding more buckets decreases performance but increases accuracy: https://prometheus.io/docs/practices/histograms/

In your settings.py

PROMETHEUS_LATENCY_BUCKETS = (.1, .2, .5, .6, .8, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.5, 9.0, 12.0, 15.0, 20.0, 30.0, float("inf"))

Monitor in multiprocess mode (uWSGI, Gunicorn)

In your settings.py

PROMETHEUS_MULTIPROC_MODE = True # default is False
PROMETHEUS_MULTIPROC_DIR = /path/to/prometheus_multiproc_dir # default it will save db files in prometheus/multiproc_dir/

Monitoring of Batch Jobs

So in prometheus legacy system we have to collect the metrics and push those metrics to the pushgateway and then prometheus server has to scrape those metrics from push gateway. But now I have modified this apporach. Now I have exposed an endpoint in this prometheus client to push your metrics.

So as usual you must be running prometheus client with server (Django, Django Rest Framework).

In settings.py: these settings is actually where your server is running.

PROMETHEUS_METRICS_PROTOCOL = "HTTP" # or HTTPS
PROMETHEUS_METRICS_HOST = "127.0.0.1"
PROMETHEUS_METRICS_PORT = "8000"
PROMETHEUS_PUSH_METRICS_URL = "/push/metrics"

In your any batch_job.py

from prometheus import batch_monitor

@batch_monitor(app_name="sum")
def sum(a,b):
   return a+b

sum(10, 20)

So here this batch_monitor decorator will push the metrics to you server and add monitored metrics into your server's metrics.

Default Batch Job Monitored Metrics

* request_count
* time_taken
* last_success
* Last_failure

These metrics can be seen at /metrics endpoint.