https://github.com/outbrain-inc/torch

A Prometheus metrics aggregator
https://github.com/outbrain-inc/torch

Last synced: 1 day ago
JSON representation

A Prometheus metrics aggregator

• Host: GitHub
• URL: https://github.com/outbrain-inc/torch
• Owner: outbrain-inc
• License: mit
• Created: 2016-12-13T20:13:48.000Z (about 8 years ago)
• Default Branch: master
• Last Pushed: 2023-03-31T12:20:50.000Z (over 1 year ago)
• Last Synced: 2024-11-12T17:02:33.120Z (about 1 month ago)
• Language: Python
• Homepage:
• Size: 15.6 KB
• Stars: 23
• Watchers: 17
• Forks: 7
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Torch: A Prometheus metrics aggregator

Torch is a web application meant to fill in for a use case the the Prometheus Pushgateway explicitly does not support, aggregating metrics.  This is useful for scenarios where standard code instrumentation fails such as a python web application which uses multiple worker processes.  If we use standard instrumentation, we would only get the results from whichever worker serves the page that Prometheus will scrape rather than the combined metrics from all of the worker processes.

## How to use

Install the torch python package. It will create a script named torch. You must set the port for torch to run on by setting the environment variable `SERVICE_PORT`

By default a metric/label combination which has not been updated will be removed after 24 hours.  Set the environment variable `TORCH_TTL` to change the number of hours to wait before cleaning up old metrics.

###Usage

    SERVICE_PORT=9092 torch

## How it works

Torch works via a simple HTTP interface.

Metrics are expected to be sent as JSON blobs with the following format:

    {

    	name: "The name of the metric",

    	description: "The description of the metric",

    	labels: {label1: "value1", label2: "value2"},

    	value: 1.2

    }

If the metric is a histogram you can define the bucket values via an additional "bucket" field to be sent as an array. If the buckets are not defined Torch will use these values by default:

    [.005, .01, .025, .05, .075, .1, .25, .5, .75, 1.0, 2.5, 5.0, 7.5, 10.0, +Inf]

If custom bucket values are sent which do not end wit +Inf, torch will automatically add +Inf as the last bucket value.

Example of a histogram metric blob with custom bucket values

    {

    	name: "The name of the metric",

    	description: "The description of the metric",

    	labels: {label1: "value1", label2: "value2"},

    	value: 1.2,

    	buckets: [1, 2, 3, 5, 8, 13, +Inf]

    }

*Note: Torch does not support dynamic quantile calculation for the Summary metric type*

## Clients

A python client is included in the torch package, though implementing a client in other languages should be pretty easy.

## Routes

###	/metrics/

This will generate the scrape page formatted how Prometheus expects

###	/metrics

This will generate the scrape page formatted how Prometheus expects

###	/metrics/counter

Increment Counter

###	/metrics/gauge/inc

Increment Gauge

###	/metrics/gauge/dec

Decrement Gauge

###	/metrics/gauge/set

Set Gauge

###	/metrics/summary

Update Summary

###	/metrics/histogram

Update Histogram