Prometheus Exporter

Prometheus Exporter allows you to aggregate custom metrics from multiple processes and export to Prometheus. It provides a very flexible framework for handling Prometheus metrics and can operate in a single and multiprocess mode.

To learn more see Instrumenting Rails with Prometheus (it has pretty pictures!)

Requirements
Installation
Usage
Transport concerns
JSON generation and parsing
Contributing
License
Code of Conduct

Requirements

Minimum Ruby of version 2.3.0 is required, Ruby 2.2.0 is EOL as of 2018-03-31

Installation

Add this line to your application's Gemfile:

gem 'prometheus_exporter'

And then execute:

$ bundle

Or install it yourself as:

$ gem install prometheus_exporter

Usage

Single process mode

Simplest way of consuming Prometheus exporter is in a single process mode.

require 'prometheus_exporter/server'

# client allows instrumentation to send info to server
require 'prometheus_exporter/client'
require 'prometheus_exporter/instrumentation'

# port is the port that will provide the /metrics route
server = PrometheusExporter::Server::WebServer.new port: 12345
server.start

# wire up a default local client
PrometheusExporter::Client.default = PrometheusExporter::LocalClient.new(collector: server.collector)

# this ensures basic process instrumentation metrics are added such as RSS and Ruby metrics
PrometheusExporter::Instrumentation::Process.start(type: "my program", labels: {my_custom: "label for all process metrics"})

gauge = PrometheusExporter::Metric::Gauge.new("rss", "used RSS for process")
counter = PrometheusExporter::Metric::Counter.new("web_requests", "number of web requests")
summary = PrometheusExporter::Metric::Summary.new("page_load_time", "time it took to load page")
histogram = PrometheusExporter::Metric::Histogram.new("api_access_time", "time it took to call api")

server.collector.register_metric(gauge)
server.collector.register_metric(counter)
server.collector.register_metric(summary)
server.collector.register_metric(histogram)

gauge.observe(get_rss)
gauge.observe(get_rss)

counter.observe(1, route: 'test/route')
counter.observe(1, route: 'another/route')

summary.observe(1.1)
summary.observe(1.12)
summary.observe(0.12)

histogram.observe(0.2, api: 'twitter')

# http://localhost:12345/metrics now returns all your metrics

Custom quantiles and buckets

You can also choose custom quantiles for summaries and custom buckets for histograms.

summary = PrometheusExporter::Metric::Summary.new("load_time", "time to load page", quantiles: [0.99, 0.75, 0.5, 0.25])
histogram = PrometheusExporter::Metric::Histogram.new("api_time", "time to call api", buckets: [0.1, 0.5, 1])

Multi process mode

In some cases (for example, unicorn or puma clusters) you may want to aggregate metrics across multiple processes.

Simplest way to achieve this is to use the built-in collector.

First, run an exporter on your desired port (we use the default port of 9394):

$ prometheus_exporter

And in your application:

require 'prometheus_exporter/client'

client = PrometheusExporter::Client.default
gauge = client.register(:gauge, "awesome", "amount of awesome")

gauge.observe(10)
gauge.observe(99, day: "friday")

Then you will get the metrics:

$ curl localhost:9394/metrics
# HELP collector_working Is the master process collector able to collect metrics
# TYPE collector_working gauge
collector_working 1

# HELP awesome amount of awesome
# TYPE awesome gauge
awesome{day="friday"} 99
awesome 10

Rails integration

You can easily integrate into any Rack application.

In your Gemfile:

gem 'prometheus_exporter'

In an initializer:

unless Rails.env == "test"
  require 'prometheus_exporter/middleware'

  # This reports stats per request like HTTP status and timings
  Rails.application.middleware.unshift PrometheusExporter::Middleware
end

Ensure you run the exporter in a monitored background process:

$ bundle exec prometheus_exporter

Per-process stats

You may also be interested in per-process stats. This collects memory and GC stats:

# in an initializer
unless Rails.env == "test"
  require 'prometheus_exporter/instrumentation'

  # this reports basic process stats like RSS and GC info
  PrometheusExporter::Instrumentation::Process.start(type: "master")
end

# in unicorn/puma/passenger be sure to run a new process instrumenter after fork
after_fork do
  require 'prometheus_exporter/instrumentation'
  PrometheusExporter::Instrumentation::Process.start(type:"web")
end

Sidekiq metrics

Including Sidekiq metrics (how many jobs ran? how many failed? how long did they take? how many are dead? how many were restarted?)

Sidekiq.configure_server do |config|
   config.server_middleware do |chain|
      require 'prometheus_exporter/instrumentation'
      chain.add PrometheusExporter::Instrumentation::Sidekiq
   end
   config.death_handlers << PrometheusExporter::Instrumentation::Sidekiq.death_handler
end

To monitor Sidekiq process info:

Sidekiq.configure_server do |config|
  config.on :startup do
    require 'prometheus_exporter/instrumentation'
    PrometheusExporter::Instrumentation::Process.start type: 'sidekiq'
  end
end

Sometimes the Sidekiq server shuts down before it can send metrics, that were generated right before the shutdown, to the collector. Especially if you care about the sidekiq_restarted_jobs_total metric, it is a good idea to explicitly stop the client:

  Sidekiq.configure_server do |config|
    at_exit do
      PrometheusExporter::Client.default.stop(wait_timeout_seconds: 10)
    end
  end

Delayed Job plugin

In an initializer:

unless Rails.env == "test"
  require 'prometheus_exporter/instrumentation'
  PrometheusExporter::Instrumentation::DelayedJob.register_plugin
end

Hutch Message Processing Tracer

Capture Hutch metrics (how many jobs ran? how many failed? how long did they take?)

unless Rails.env == "test"
  require 'prometheus_exporter/instrumentation'
  Hutch::Config.set(:tracer, PrometheusExporter::Instrumentation::Hutch)
end

Instrumenting Request Queueing Time

Request Queueing is defined as the time it takes for a request to reach your application (instrumented by this prometheus_exporter) from farther upstream (as your load balancer). A high queueing time usually means that your backend cannot handle all the incoming requests in time, so they queue up (= you should see if you need to add more capacity).

As this metric starts before prometheus_exporter can handle the request, you must add a specific HTTP header as early in your infrastructure as possible (we recommend your load balancer or reverse proxy).

Configure your HTTP server / load balancer to add a header X-Request-Start: t=<MSEC> when passing the request upstream. For more information, please consult your software manual.

Hint: we aim to be API-compatible with the big APM solutions, so if you've got requests queueing time configured for them, it should be expected to also work with prometheus_exporter.

Puma metrics

The puma metrics are using the Puma.stats method and hence need to be started after the workers has been booted and from a Puma thread otherwise the metrics won't be accessible. The easiest way to gather this metrics is to put the following in your puma.rb config:

# puma.rb config
after_worker_boot do
  require 'prometheus_exporter/instrumentation'
  PrometheusExporter::Instrumentation::Puma.start
end

Unicorn process metrics

In order to gather metrics from unicorn processes, we use rainbows, which exposes Rainbows::Linux.tcp_listener_stats to gather information about active workers and queued requests. To start monitoring your unicorn processes, you'll need to know both the path to unicorn PID file and the listen address (pid_file and listen in your unicorn config file)

Then, run prometheus_exporter with --unicorn-master and --unicorn-listen-address options:

prometheus_exporter --unicorn-master /var/run/unicorn.pid --unicorn-listen-address 127.0.0.1:3000

# alternatively, if you're using unix sockets:
prometheus_exporter --unicorn-master /var/run/unicorn.pid --unicorn-listen-address /var/run/unicorn.sock

Note: You must install the raindrops gem in your Gemfile or locally.

Custom type collectors

In some cases you may have custom metrics you want to ship the collector in a batch. In this case you may still be interested in the base collector behavior, but would like to add your own special messages.

# person_collector.rb
class PersonCollector < PrometheusExporter::Server::TypeCollector
  def initialize
    @oldies = PrometheusExporter::Metric::Counter.new("oldies", "old people")
    @youngies = PrometheusExporter::Metric::Counter.new("youngies", "young people")
  end

  def type
    "person"
  end

  def collect(obj)
    if obj["age"] > 21
      @oldies.observe(1)
    else
      @youngies.observe(1)
    end
  end

  def metrics
    [@oldies, @youngies]
  end
end

Shipping metrics then is done via:

PrometheusExporter::Client.default.send_json(type: "person", age: 40)

To load the custom collector run:

$ bundle exec prometheus_exporter -a person_collector.rb

Global metrics in a custom type collector

Custom type collectors are the ideal place to collect global metrics, such as user/article counts and connection counts. The custom type collector runs in the collector, which usually runs in the prometheus exporter process.

Out-of-the-box we try to keep the prometheus exporter as lean as possible. We do not load all Rails dependencies, so you won't have access to your models. You can always ensure it is loaded in your custom type collector with:

unless defined? Rails
  require File.expand_path("../../config/environment", __FILE__)
end

Then you can collect the metrics you need on demand:

def metrics
  user_count_gague = PrometheusExporter::Metric::Gauge.new('user_count', 'number of users in the app')
  user_count_gague.observe User.count
  [user_count_gauge]
end

The metrics endpoint is called whenever prometheus calls the /metrics HTTP endpoint, so it may make sense to introduce some type of caching. lru_redux is the perfect gem for this job: you can use LruRedux::TTL::Cache, which will expire automatically after N seconds, thus saving multiple database queries.

Multi process mode with custom collector

You can opt for custom collector logic in a multi process environment.

This allows you to completely replace the collector logic.

First, define a custom collector. It is important that you inherit off PrometheusExporter::Server::CollectorBase and have custom implementations for #process and #prometheus_metrics_text methods.

class MyCustomCollector < PrometheusExporter::Server::CollectorBase
  def initialize
    @gauge1 = PrometheusExporter::Metric::Gauge.new("thing1", "I am thing 1")
    @gauge2 = PrometheusExporter::Metric::Gauge.new("thing2", "I am thing 2")
    @mutex = Mutex.new
  end

  def process(str)
    obj = JSON.parse(str)
    @mutex.synchronize do
      if thing1 = obj["thing1"]
        @gauge1.observe(thing1)
      end

      if thing2 = obj["thing2"]
        @gauge2.observe(thing2)
      end
    end
  end

  def prometheus_metrics_text
    @mutex.synchronize do
      "#{@gauge1.to_prometheus_text}\n#{@gauge2.to_prometheus_text}"
    end
  end
end

Next, launch the exporter process:

$ bin/prometheus_exporter --collector examples/custom_collector.rb

In your application send metrics you want:

require 'prometheus_exporter/client'

client = PrometheusExporter::Client.new(host: 'localhost', port: 12345)
client.send_json(thing1: 122)
client.send_json(thing2: 12)

Now your exporter will echo the metrics:

$ curl localhost:12345/metrics
# HELP collector_working Is the master process collector able to collect metrics
# TYPE collector_working gauge
collector_working 1

# HELP thing1 I am thing 1
# TYPE thing1 gauge
thing1 122

# HELP thing2 I am thing 2
# TYPE thing2 gauge
thing2 12

GraphQL support

GraphQL execution metrics are supported and can be collected via the GraphQL collector, included in graphql-ruby.

Metrics default prefix / labels

This only works in single process mode.

You can specify default prefix or labels for metrics. For example:

# Specify prefix for metric names
PrometheusExporter::Metric::Base.default_prefix = "ruby"

# Specify default labels for metrics
PrometheusExporter::Metric::Base.default_labels = { "hostname" => "app-server-01" }

counter = PrometheusExporter::Metric::Counter.new("web_requests", "number of web requests")

counter.observe(1, route: 'test/route')
counter.observe

Will result in:

# HELP web_requests number of web requests
# TYPE web_requests counter
ruby_web_requests{hostname="app-server-01",route="test/route"} 1
ruby_web_requests{hostname="app-server-01"} 1

Client default labels

You can specify a default label for instrumentation metrics sent by a specific client. For example:

# Specify on intializing PrometheusExporter::Client
PrometheusExporter::Client.new(custom_labels: { hostname: 'app-server-01', app_name: 'app-01' })

# Specify on an instance of PrometheusExporter::Client
client = PrometheusExporter::Client.new
client.custom_labels = { hostname: 'app-server-01', app_name: 'app-01' }

Will result in:

http_requests_total{controller="home","action"="index",service="app-server-01",app_name="app-01"} 2
http_requests_total{service="app-server-01",app_name="app-01"} 1

Transport concerns

Prometheus Exporter handles transport using a simple HTTP protocol. In multi process mode we avoid needing a large number of HTTP request by using chunked encoding to send metrics. This means that a single HTTP channel can deliver 100s or even 1000s of metrics over a single HTTP session to the /send-metrics endpoint. All calls to send and send_json on the PrometheusExporter::Client class are non-blocking and batched.

The /bench directory has simple benchmark, which is able to send through 10k messages in 500ms.

JSON generation and parsing

The PrometheusExporter::Client class has the method #send-json. This method, by default, will call JSON.dump on the Object it recieves. You may opt in for oj mode where it can use the faster Oj.dump(obj, mode: :compat) for JSON serialization. But be warned that if you have custom objects that implement own to_json methods this may not work as expected. You can opt for oj serialization with json_serializer: :oj.

When PrometheusExporter::Server::Collector parses your JSON, by default it will use the faster Oj deserializer if available. This happens cause it only expects a simple Hash out of the box. You can opt in for the default JSON deserializer with json_serializer: :json.

Contributing

Bug reports and pull requests are welcome on GitHub at https://github.com/discourse/prometheus_exporter. This project is intended to be a safe, welcoming space for collaboration, and contributors are expected to adhere to the Contributor Covenant code of conduct.

License

The gem is available as open source under the terms of the MIT License.

Code of Conduct

Everyone interacting in the PrometheusExporter project’s codebases, issue trackers, chat rooms and mailing lists is expected to follow the code of conduct.

jcomeaux/prometheus_exporter