There's a lot of open issues
A long-lived project that still receives updates
Prometheus metric collector and exporter for Ruby


~> 6.0.0
~> 2.3
~> 2.0
~> 0.5.0
~> 5.15.0
~> 3.0
~> 0.8.3
~> 13.0
>= 0.69
>= 2.1.4
>= 0


 Project Readme

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
  • Migrating from v0.x
  • Installation
  • Usage
    • Single process mode
      • Custom quantiles and buckets
    • Multi process mode
    • Rails integration
      • Per-process stats
      • Sidekiq metrics
      • Shoryuken metrics
      • ActiveRecord Connection Pool Metrics
      • Delayed Job plugin
      • Hutch metrics
    • Puma metrics
    • Unicorn metrics
    • Resque metrics
    • GoodJob metrics
    • Custom type collectors
    • Multi process mode with custom collector
    • GraphQL support
    • Metrics default prefix / labels
    • Client default labels
    • Client default host
    • Histogram mode
    • Histogram - custom buckets
  • Transport concerns
  • JSON generation and parsing
  • Logging
  • Docker Usage
  • Contributing
  • License
  • Code of Conduct


Minimum Ruby of version 3.0.0 is required, Ruby 2.7 is EOL as of March 31st 2023.

Migrating from v0.x

There are some major changes in v1.x from v0.x.


Add this line to your application's Gemfile:

gem 'prometheus_exporter'

And then execute:

$ bundle

Or install it yourself as:

$ gem install prometheus_exporter


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'

# bind is the address, on which the webserver will listen
# port is the port that will provide the /metrics route
server = bind: 'localhost', port: 12345

# wire up a default local client
PrometheusExporter::Client.default = 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 ="rss", "used RSS for process")
counter ="web_requests", "number of web requests")
summary ="page_load_time", "time it took to load page")
histogram ="api_access_time", "time it took to call api")



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


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 ="load_time", "time to load page", quantiles: [0.99, 0.75, 0.5, 0.25])
histogram ="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 bind to localhost and 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(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

Custom quantiles for summaries and buckets for histograms can also be passed in.

require 'prometheus_exporter/client'

client = PrometheusExporter::Client.default
histogram = client.register(:histogram, "api_time", "time to call api", buckets: [0.1, 0.5, 1])

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

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

Ensure you run the exporter in a monitored background process:

$ bundle exec prometheus_exporter

Choosing the style of method patching

By default, prometheus_exporter uses alias_method to instrument methods used by SQL and Redis as it is the fastest approach (see this article). You may desire to add additional instrumentation libraries beyond prometheus_exporter to your app. This can become problematic if these other libraries instead use prepend to instrument methods. To resolve this, you can tell the middleware to instrument using prepend by passing an instrument option like so:

Rails.application.middleware.unshift PrometheusExporter::Middleware, instrument: :prepend

Metrics collected by Rails integration middleware

Type Name Description
Counter http_requests_total Total HTTP requests from web app
Summary http_request_duration_seconds Time spent in HTTP reqs in seconds
Summary http_request_redis_duration_seconds¹ Time spent in HTTP reqs in Redis, in seconds
Summary http_request_sql_duration_seconds² Time spent in HTTP reqs in SQL in seconds
Summary http_request_queue_duration_seconds³ Time spent queueing the request in load balancer in seconds

All metrics have a controller and an action label. http_requests_total additionally has a (HTTP response) status label.

To add your own labels to the default metrics, create a subclass of PrometheusExporter::Middleware, override custom_labels, and use it in your initializer.

class MyMiddleware < PrometheusExporter::Middleware
  def custom_labels(env)
    labels = {}

    if env['HTTP_X_PLATFORM']
      labels['platform'] = env['HTTP_X_PLATFORM']


If you're not using Rails like framework, you can extend PrometheusExporter::Middleware#default_labels in a way to add more relevant labels. For example you can mimic prometheus-client labels with code like this:

class MyMiddleware < PrometheusExporter::Middleware
  def default_labels(env, result)
    status = (result && result[0]) || -1
    path = [env["SCRIPT_NAME"], env["PATH_INFO"]].join
      path: strip_ids_from_path(path),
      method: env["REQUEST_METHOD"],
      status: status

  def strip_ids_from_path(path)
      .gsub(%r{/[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}(/|$)}, '/:uuid\\1')
      .gsub(%r{/\d+(/|$)}, '/:id\\1')

That way you won't have all metrics labeled with controller=other and action=other, but have labels such as

ruby_http_request_duration_seconds{path="/api/v1/teams/:id",method="GET",status="200",quantile="0.99"} 0.009880661998977303

¹) Only available when Redis is used. ²) Only available when Mysql or PostgreSQL are used. ³) Only available when Instrumenting Request Queueing Time is set up.

Activerecord Connection Pool Metrics

This collects activerecord connection pool metrics.

It supports injection of custom labels and the connection config options (username, database, host, port) as labels.

For Puma single mode

#in puma.rb
require 'prometheus_exporter/instrumentation'
  custom_labels: { type: "puma_single_mode" }, #optional params
  config_labels: [:database, :host] #optional params

For Puma cluster mode

# in puma.rb
on_worker_boot do
  require 'prometheus_exporter/instrumentation'
    custom_labels: { type: "puma_worker" }, #optional params
    config_labels: [:database, :host] #optional params

For Unicorn / Passenger

after_fork do |_server, _worker|
  require 'prometheus_exporter/instrumentation'
    custom_labels: { type: "unicorn_worker" }, #optional params
    config_labels: [:database, :host] #optional params

For Sidekiq

Sidekiq.configure_server do |config|
  config.on :startup do
    require 'prometheus_exporter/instrumentation'
      custom_labels: { type: "sidekiq" }, #optional params
      config_labels: [:database, :host] #optional params
Metrics collected by ActiveRecord Instrumentation
Type Name Description
Gauge active_record_connection_pool_connections Total connections in pool
Gauge active_record_connection_pool_busy Connections in use in pool
Gauge active_record_connection_pool_dead Dead connections in pool
Gauge active_record_connection_pool_idle Idle connections in pool
Gauge active_record_connection_pool_waiting Connection requests waiting
Gauge active_record_connection_pool_size Maximum allowed connection pool size

All metrics collected by the ActiveRecord integration include at least the following labels: pid (of the process the stats where collected in), pool_name, any labels included in the config_labels option (prefixed with dbconfig_, example: dbconfig_host), and all custom labels provided with the custom_labels option.

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")

# 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")
Metrics collected by Process Instrumentation
Type Name Description
Gauge heap_free_slots Free ruby heap slots
Gauge heap_live_slots Used ruby heap slots
Gauge v8_heap_size* Total JavaScript V8 heap size (bytes)
Gauge v8_used_heap_size* Total used JavaScript V8 heap size (bytes)
Gauge v8_physical_size* Physical size consumed by V8 heaps
Gauge v8_heap_count* Number of V8 contexts running
Gauge rss Total RSS used by process
Counter major_gc_ops_total Major GC operations by process
Counter minor_gc_ops_total Minor GC operations by process
Counter allocated_objects_total Total number of allocated objects by process

Metrics marked with * are only collected when MiniRacer is defined.

Metrics collected by Process instrumentation include labels type (as given with the type option), pid (of the process the stats where collected in), and any custom labels given to Process.start with the labels option.

Sidekiq metrics

There are different kinds of Sidekiq metrics that can be collected. A recommended setup looks like this:

Sidekiq.configure_server do |config|
  require 'prometheus_exporter/instrumentation'
  config.server_middleware do |chain|
    chain.add PrometheusExporter::Instrumentation::Sidekiq
  config.death_handlers << PrometheusExporter::Instrumentation::Sidekiq.death_handler
  config.on :startup do
    PrometheusExporter::Instrumentation::Process.start type: 'sidekiq'
  • The middleware and death handler will generate job specific metrics (how many jobs ran? how many failed? how long did they take? how many are dead? how many were restarted?).
  • The Process metrics provide basic ruby metrics.
  • The SidekiqProcess metrics provide the concurrency and busy metrics for this process.
  • The SidekiqQueue metrics provides size and latency for the queues run by this process.
  • The SidekiqStats metrics provide general, global Sidekiq stats (size of Scheduled, Retries, Dead queues, total number of jobs, etc).

For SidekiqQueue, if you run more than one process for the same queues, note that the same metrics will be exposed by all the processes, just like the SidekiqStats will if you run more than one process of any kind. You might want use avg or max when consuming their metrics.

An alternative would be to expose these metrics in lone, long-lived process. Using a rake task, for example:

task :sidekiq_metrics do
  server =

  PrometheusExporter::Client.default = server.collector)

  PrometheusExporter::Instrumentation::SidekiqQueue.start(all_queues: true)

The all_queues parameter for SidekiqQueue will expose metrics for all queues.

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)

Custom labels can be added for individual jobs by defining a class method on the job class. These labels will be added to all Sidekiq metrics written by the job:

  class WorkerWithCustomLabels
    def self.custom_labels
      { my_label: 'value-here', other_label: 'second-val' }

    def perform; end
Metrics collected by Sidekiq Instrumentation


Type Name Description
Summary sidekiq_job_duration_seconds Time spent in sidekiq jobs
Counter sidekiq_jobs_total Total number of sidekiq jobs executed
Counter sidekiq_restarted_jobs_total Total number of sidekiq jobs that we restarted because of a sidekiq shutdown
Counter sidekiq_failed_jobs_total Total number of failed sidekiq jobs

All metrics have a job_name label and a queue label.


Type Name Description
Counter sidekiq_dead_jobs_total Total number of dead sidekiq jobs

This metric has a job_name label and a queue label.


Type Name Description
Gauge sidekiq_queue_backlog Size of the sidekiq queue
Gauge sidekiq_queue_latency_seconds Latency of the sidekiq queue

Both metrics will have a queue label with the name of the queue.


Type Name Description
Gauge sidekiq_process_busy Number of busy workers for this process
Gauge sidekiq_process_concurrency Concurrency for this process

Both metrics will include the labels labels, queues, quiet, tag, hostname and identity, as returned by the Sidekiq Processes API.


Type Name Description
Gauge sidekiq_stats_dead_size Size of the dead queue
Gauge sidekiq_stats_enqueued Number of enqueued jobs
Gauge sidekiq_stats_failed Number of failed jobs
Gauge sidekiq_stats_processed Total number of processed jobs
Gauge sidekiq_stats_processes_size Number of processes
Gauge sidekiq_stats_retry_size Size of the retries queue
Gauge sidekiq_stats_scheduled_size Size of the scheduled queue
Gauge sidekiq_stats_workers_size Number of jobs actively being processed

Based on the Sidekiq Stats API.

See Metrics collected by Process Instrumentation for a list of metrics the Process instrumentation will produce.

Shoryuken metrics

For Shoryuken metrics (how many jobs ran? how many failed? how long did they take? how many were restarted?)

Shoryuken.configure_server do |config|
  config.server_middleware do |chain|
    require 'prometheus_exporter/instrumentation'
    chain.add PrometheusExporter::Instrumentation::Shoryuken
Metrics collected by Shoryuken Instrumentation
Type Name Description
Counter shoryuken_job_duration_seconds Total time spent in shoryuken jobs
Counter shoryuken_jobs_total Total number of shoryuken jobs executed
Counter shoryuken_restarted_jobs_total Total number of shoryuken jobs that we restarted because of a shoryuken shutdown
Counter shoryuken_failed_jobs_total Total number of failed shoryuken jobs

All metrics have labels for job_name and queue_name.

Delayed Job plugin

In an initializer:

unless Rails.env.test?
  require 'prometheus_exporter/instrumentation'
Metrics collected by Delayed Job Instrumentation
Type Name Description Labels
Counter delayed_job_duration_seconds Total time spent in delayed jobs job_name
Counter delayed_job_latency_seconds_total Total delayed jobs latency job_name
Counter delayed_jobs_total Total number of delayed jobs executed job_name
Gauge delayed_jobs_enqueued Number of enqueued delayed jobs -
Gauge delayed_jobs_pending Number of pending delayed jobs -
Counter delayed_failed_jobs_total Total number failed delayed jobs executed job_name
Counter delayed_jobs_max_attempts_reached_total Total number of delayed jobs that reached max attempts -
Summary delayed_job_duration_seconds_summary Summary of the time it takes jobs to execute status
Summary delayed_job_attempts_summary Summary of the amount of attempts it takes delayed jobs to succeed -

All metrics have labels for job_name and queue_name. delayed_job_latency_seconds_total is considering delayed job's sleep_delay parameter, so please be aware of this in case you are looking for high latency precision.

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)
Metrics collected by Hutch Instrumentation
Type Name Description
Counter hutch_job_duration_seconds Total time spent in hutch jobs
Counter hutch_jobs_total Total number of hutch jobs executed
Counter hutch_failed_jobs_total Total number failed hutch jobs executed

All metrics have a job_name label.

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).

The Amazon Application Load Balancer request tracing header is natively supported. If you are using another upstream entrypoint, you may configure your HTTP server / load balancer to add a header X-Request-Start: t=<MSEC> when passing the request upstream. Please keep in mind request time start is reported as epoch time (in seconds) and lacks precision, which may introduce additional latency in reported metrics. 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:

For Puma single mode

# puma.rb config
require 'prometheus_exporter/instrumentation'
# optional check, avoids spinning up and down threads per worker
if !PrometheusExporter::Instrumentation::Puma.started?

For Puma clustered mode

# puma.rb config
after_worker_boot do
  require 'prometheus_exporter/instrumentation'
  # optional check, avoids spinning up and down threads per worker
  if !PrometheusExporter::Instrumentation::Puma.started?

Metrics collected by Puma Instrumentation

Type Name Description
Gauge puma_workers Number of puma workers
Gauge puma_booted_workers Number of puma workers booted
Gauge puma_old_workers Number of old puma workers
Gauge puma_running_threads Number of puma threads currently running
Gauge puma_request_backlog Number of requests waiting to be processed by a puma thread
Gauge puma_thread_pool_capacity Number of puma threads available at current scale
Gauge puma_max_threads Number of puma threads at available at max scale

All metrics may have a phase label and all custom labels provided with the labels option.

Resque metrics

The resque metrics are using the method, which queries Redis internally. To start monitoring your resque installation, you'll need to start the instrumentation:

# e.g. config/initializers/resque.rb
require 'prometheus_exporter/instrumentation'

Metrics collected by Resque Instrumentation

Type Name Description
Gauge resque_processed_jobs Total number of processed Resque jobs
Gauge resque_failed_jobs Total number of failed Resque jobs
Gauge resque_pending_jobs Total number of pending Resque jobs
Gauge resque_queues Total number of Resque queues
Gauge resque_workers Total number of Resque workers running
Gauge resque_working Total number of Resque workers working

GoodJob metrics

The metrics are generated from the database using the relevant scopes. To start monitoring your GoodJob installation, you'll need to start the instrumentation:

# e.g. config/initializers/good_job.rb
require 'prometheus_exporter/instrumentation'

Metrics collected by GoodJob Instrumentation

Type Name Description
Gauge good_job_scheduled Total number of scheduled GoodJob jobs.
Gauge good_job_retried Total number of retried GoodJob jobs.
Gauge good_job_queued Total number of queued GoodJob jobs.
Gauge good_job_running Total number of running GoodJob jobs.
Gauge good_job_finished Total number of finished GoodJob jobs.
Gauge good_job_succeeded Total number of succeeded GoodJob jobs.
Gauge good_job_discarded Total number of discarded GoodJob jobs

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-listen-address

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

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

Metrics collected by Unicorn Instrumentation

Type Name Description
Gauge unicorn_workers Number of unicorn workers
Gauge unicorn_active_workers Number of active unicorn workers
Gauge unicorn_request_backlog Number of requests waiting to be processed by a unicorn worker

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 ="oldies", "old people")
    @youngies ="youngies", "young people")

  def type

  def collect(obj)
    if obj["age"] > 21

  def metrics
    [@oldies, @youngies]

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__)

Then you can collect the metrics you need on demand:

def metrics
  user_count_gauge ='user_count', 'number of users in the app')
  user_count_gauge.observe User.count

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 ="thing1", "I am thing 1")
    @gauge2 ="thing2", "I am thing 2")
    @mutex =

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

      if thing2 = obj["thing2"]

  def prometheus_metrics_text
    @mutex.synchronize do

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 = '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 ="web_requests", "number of web requests")

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

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

Exporter Process Configuration

When running the process for prometheus_exporter using bin/prometheus_exporter, there are several configurations that can be passed in:

Usage: prometheus_exporter [options]
    -p, --port INTEGER               Port exporter should listen on (default: 9394)
    -b, --bind STRING                IP address exporter should listen on (default: localhost)
    -t, --timeout INTEGER            Timeout in seconds for metrics endpoint (default: 2)
        --prefix METRIC_PREFIX       Prefix to apply to all metrics (default: ruby_)
        --label METRIC_LABEL         Label to apply to all metrics (default: {})
    -c, --collector FILE             (optional) Custom collector to run
    -a, --type-collector FILE        (optional) Custom type collectors to run in main collector
    -v, --verbose
    -g, --histogram                  Use histogram instead of summary for aggregations
        --auth FILE                  (optional) enable basic authentication using a htpasswd FILE
        --realm REALM                (optional) Use REALM for basic authentication (default: "Prometheus Exporter")
        --unicorn-listen-address ADDRESS
                                     (optional) Address where unicorn listens on (unix or TCP address)
        --unicorn-master PID_FILE    (optional) PID file of unicorn master process to monitor unicorn


The following will run the process at

  • Port 8080 (default 9394)
  • Bind to (default localhost)
  • Timeout in 1 second for metrics endpoint (default 2 seconds)
  • Metric prefix as foo_ (default ruby_)
  • Default labels as {environment: "integration", foo: "bar"}
prometheus_exporter -p 8080 \
                    -b \
                    -t 1 \
                    --label '{"environment": "integration", "foo": "bar"}' \
                    --prefix 'foo_'

You can use -b option to bind the prometheus_exporter web server to any IPv4 interface with -b, any IPv6 interface with -b ::, or -b ANY to any IPv4/IPv6 interfaces available on your host system.

Enabling Basic Authentication

If you desire authentication on your /metrics route, you can enable basic authentication with the --auth option.

$ prometheus_exporter --auth my-htpasswd-file

Additionally, the --realm option may be used to provide a customized realm for the challenge request.


  • You will need to create a htpasswd formatted file before hand which contains one or more user:password entries
  • Only the basic crypt encryption is currently supported

A simple htpasswd file can be created with the Apache htpasswd utility; e.g:

$ htpasswd -cdb my-htpasswd-file my-user my-unencrypted-password

This will create a file named my-htpasswd-file which is suitable for use the --auth option.

Client default labels

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

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

# Specify on an instance of PrometheusExporter::Client
client =
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

Client default host

By default, PrometheusExporter::Client.default connects to localhost:9394. If your setup requires this (e.g. when using docker-compose), you can change the default host and port by setting the environment variables PROMETHEUS_EXPORTER_HOST and PROMETHEUS_EXPORTER_PORT.

Histogram mode

By default, the built-in collectors will report aggregations as summaries. If you need to aggregate metrics across labels, you can switch from summaries to histograms:

$ prometheus_exporter --histogram

In histogram mode, the same metrics will be collected but will be reported as histograms rather than summaries. This sacrifices some precision but allows aggregating metrics across actions and nodes using histogram_quantile.

Histogram - custom buckets

By default these buckets will be used:

[0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1, 2.5, 5.0, 10.0].freeze

if this is not enough you can specify default_buckets like this:

Histogram.default_buckets = [0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1, 2, 2.5, 3, 4, 5.0, 10.0, 12, 14, 15, 20, 25].freeze

Specfied buckets on the instance takes precedence over default:

Histogram.default_buckets = [0.005, 0.01, 0,5].freeze
buckets = [0.1, 0.2, 0.3]
histogram ='test_bucktets', 'I have specified buckets', buckets: buckets)
histogram.buckets => [0.1, 0.2, 0.3]

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.


PrometheusExporter::Client.default will export to STDERR. To change this, you can pass your own logger: Rails.logger)

You can also pass a log level (default is Logger::WARN): Logger::DEBUG)

Docker Usage

You can run prometheus_exporter project using an official Docker image:

docker pull discourse/prometheus_exporter:latest
# or use specific version
docker pull discourse/prometheus_exporter:x.x.x

The start the container:

docker run -p 9394:9394 discourse/prometheus_exporter

Additional flags could be included:

docker run -p 9394:9394 discourse/prometheus_exporter --verbose --prefix=myapp

Docker/Kubernetes Healthcheck

A /ping endpoint which only returns PONG is available so you can run container healthchecks :


      - bin/prometheus_exporter
      - -b
      test: ["CMD", "curl", "--silent", "--show-error", "--fail", "--max-time", "3", ""]
      timeout: 3s
      interval: 10s
      retries: 5


Bug reports and pull requests are welcome on GitHub at 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.


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.