subaru9 / prometheus_telemetry_elixir Goto Github PK

PrometheusTelemetry is the plumbing for Telemetry.Metrics and allows the metrics passed in to be collected and exported in the format expected by the prometheus scraper.

This supervisor also contains the ability to spawn an exporter which will scrape every supervisor running for metrics and will spin up a plug and return it at /metrics on port 4050 by default, this will work out of the box with umbrella apps as well and allow you to define metrics in each umbrella app

The package can be installed by adding prometheus_telemetry to your list of dependencies in mix.exs:

def deps do
  [
    {:prometheus_telemetry, "~> 0.4"}
  ]
end

Documentation can be found at https://hexdocs.pm/prometheus_telemetry.

We can add this to any application we want:

children = [
  {PrometheusTelemetry, metrics: [
   MyMetricsModule.metrics()
  ]}
]

Then to setup an exporter, on a server application like a phoenix app or pipeline we can setup the exporter which will start the metrics server (by default on port localhost:4050):

children = [
  {PrometheusTelemetry,
    exporter: [enabled?: true],
    metrics: MyMetricsModule.metrics()
  }
]

There are built-in metrics for some erlang vm stats, phoenix, absinthe, ecto and oban. To enable them we can use the following modules:

PrometheusTelemetry.Metrics.{Ecto,Cowboy,Swoosh,Finch,Phoenix,GraphQL,Oban,VM}

(See the notes below on conveniences for configuring metrics for Ecto repos.)

children = [
  {PrometheusTelemetry,
    exporter: [enabled?: true],
    metrics: [
      PrometheusTelemetry.Metrics.Ecto.metrics(:my_ecto_app),
      PrometheusTelemetry.Metrics.Cowboy.metrics(),
      PrometheusTelemetry.Metrics.Swoosh.metrics(),
      PrometheusTelemetry.Metrics.Finch.metrics(),
      PrometheusTelemetry.Metrics.Phoenix.metrics(),
      PrometheusTelemetry.Metrics.GraphQL.metrics(),
      PrometheusTelemetry.Metrics.Oban.metrics(),
      PrometheusTelemetry.Metrics.VM.metrics()
    ]
  }
]

These are the default config settings, you can override by setting any of them

config :prometheus_telemetry,
  default_millisecond_buckets: [100, 300, 500, 1000, 2000, 5000, 10_000],
  default_microsecond_buckets: [50_000, 100_000, 250_000, 500_000, 750_000],
  measurement_poll_period: :timer.seconds(10),
  ecto_max_query_length: 150,
  ecto_known_query_module: nil

To define metrics we can create a module to group them like MyMetricsModule

defmodule MyMetricsModule do
  import Telemetry.Metrics, only: [last_value: 2, counter: 2]

  def metrics do
    [
      counter(
        "prometheus_name.to_save", # becomes prometheus_name_to_save in prometheus
        event_name: [:event_namespace, :my_metric], # telemetry event name
        measurement: :count, # telemetry event metric
        description: "some description"
      ),

      last_value(
        "my_custom.name",
        event_name: [:event_namespace, :last_value],
        measurement: :total,
        description: "my value",
        tags: [:custom_metric] # custom metrics to save, derived from :telemetry.execute metadata
      )
    ]
  end

  def inc_to_save do
    :telemetry.execute([:event_namespace, :my_metric], %{count: 1})
  end

  def set_custom_name do
    :telemetry.execute([:event_namespace, :last_value], %{total: 123}, %{custom_metric: "region"})
  end
end

Ultimately every list will get flattened which allows you to group metric modules under a single module such as

defmodule GraphQL.Request do
  def metrics do
    ...
  end
end

defmodule GraphQL.Complexity do
  def metrics do
    ...
  end
end

defmodule GraphQL do
  def metrics, do: [GraphQL.Complexity.metrics(), GraphQL.Request.metrics()]
end

For more details on types you can check telemetry_metrics_prometheus_core You can also generate these custom metrics modules using mix prometheus_telemetry.gen.metrics

This library hooks into Ecto's adapter-specific telemetry events. In order to correctly namespace events for your repo(s), you must pass an atom (or list of atoms) when injecting the metrics. E.g.: PrometheusTelemetry.Metrics.Ecto.metrics(:my_ecto_app) will produce corresponding metrics with the correct event_name of [:my_ecto_app, :repo, :query].

PrometheusTelemetry.Metrics.Ecto.metrics_for_repo/1 and PrometheusTelemetry.Metrics.Ecto.metrics_for_repos/1 can be used as a convenience for umbrellas with multiple repos and for deduplicating metrics for replicas:

iex> PrometheusTelemetry.Metrics.Ecto.metrics_for_repo(MyEctoApp.Repo)
[
  %Telemetry.Metrics.Distribution{
    description: "Gets total time spent on query",
    event_name: [:my_ecto_app, :repo, :query],
		...
  },
  ...
]

A few additional extras exist for Ecto, which include the ability to set the max query size before truncation as well as add a known module query which will be called with shorten(query)

You can set both of these via config, by default there's no known query module and the max query size is 150.

The other way to set the label for a query is directly in the Repo.all or Actions.all (if using ecto_shorts)

Repo:

Repo.all(User, telemetry_options: [label: "All Users"])

EctoShorts:

Actions.find(User, params, telemetry_options: [label: "Find User"])

Using the mix prometheus_telemetry.gen.metrics you can generate metrics modules instead of writing them manually

$ mix prometheus_telemetry.gen.metrics MyApp.Metrics.Type counter:event.name.measurement.count:event.name.count:count:tags:profile:region

The format used is the following:

<metric_type>:<metric_name>:<event_name>:<measurement_unit>:tags:<tag_name>:<tag_name>

View the Mix.Tasks.PrometheusTelemetry.Gen.Metrics module for more details

Are you looking for a new gig?? We're looking for mid-level to senior level developers to join our team and continue growing our platform while building awesome software!

Come join us at Blitz.gg