paambaati/mae

🔥Mesos app exporter for Prometheus

🔥 mae

mae collects Prometheus metrics from your Mesos apps.

See Deployment Strategy to learn how to set up mae on your Mesos cluster.

Requirements

mae requires Python (>= 2.7), and supports Python 3 (>= 3.5).

Installation

pip install mae

Usage

$ mae --help

usage: mae [-h] app_port slave_address slave_port

positional arguments:
  app_port       Port on which this exporter will run on
  slave_address  Mesos slave address
  slave_port     Mesos slave port

The logging level of the CLI can also be configured with the LOG_LEVEL environment variable. Read the logging library's levels for all the available levels.

Task Labels

Once mae is up and running, it will start collecting metrics from all apps that have the prometheus.metrics.enabled label. You can also customize how the metrics are collected —

Task Label	Description	Required?	Default
prometheus.metrics.enabled	Enables metrics collection. If the label isn't found, the app's metrics are not collected.	Yes
prometheus.metrics.port_index	The port index where your app is exposing its Prometheus metrics. This is useful for Mesos apps that use multiple ports.	No	0
prometheus.metrics.endpoint	The endpoint where your app is exposing its metrics.	No	/metrics

Deployment Strategy

mae is designed to be run as a daemon on all Mesos slave nodes. This ideally involves 2 steps —

1. Install mae as part of your base image/AMI or via user data.
2. Run mae as a daemon. For example, if your distro supports systemd, here's a sample script that runs the exporter on port 8888 —

   # Save this to /etc/systemd/system/mae.service
   [Unit]
   Description=Mesos App Exporter
   After=network.target
   
   [Service]
   Type=simple
   Restart=on-failure
   Environment="LOG_LEVEL=INFO"
   ExecStart=/usr/local/bin/mae 8888 localhost 5051 # Assuming the Mesos slave process is running on port 5051
   
   [Install]
   WantedBy=multi-user.target
   

mae's Issues

An app without any labels can crash mae's response

When the app definition does not have any labels, the response is empty. This can be fixed by adding defaults to the get() calls here —

mae/mae/mae.py

Line 48 in ab1eadb

labels = task.get('labels').get('labels')