Settings View Source Seeding Metrics

In some applications, telemetry is only emitted periodically or in spurts. Capturing these metrics via PromEx.Plugin.event_metrics/1 can lead to gaps in Grafana graphs after a service restart or deployment.

When an application starts, its Prometheus metrics are empty; its metrics_server_path will only show metrics that have been emitted and captured by PromEx over the course of the beam's lifespan. When a metric is not seen within a time period, Prometheus returns no data for that metric, and in Grafana rather than data points with 0 values, the graph will show No Data.

One mechanism to ensure that the graphs are populated is to seed the event metrics, i.e. emit them at least once after PromEx has started. This can be done via a Application.start_phase/3.

start-phases

Start Phases

Start phases are configured in a project's application callback in mix.exs:

  def application do
    [
      extra_applications: [],
      mod: {MyApp.Application, []},
      start_phases: [seed_prom_ex_telemetry: []]
    ]
  end

This requires adding a callback to MyApp.Application:

defmodule MyApp.Application do
  use Application

  @impl true
  def start(_type, _args) do
    children = [
      MyApp.PromEx
      # ... Repo, Phoenix, etc
    ]

    opts = [strategy: :one_for_one, name: MyApp.Supervisor]
    Supervisor.start_link(children, opts)
  end

  @impl true
  def start_phase(:seed_prom_ex_telemetry, :normal, _) do
    MyApp.PromEx.seed_event_metrics()
    :ok
  end
end

The seed_event_metrics/0 function can send any telemetry events desired, with any variations in measurements and metadata to ensure that the metrics_server_path is fully populated when it is first scraped.

def seed_event_metrics do
  # :telemetry.execute([:my_app, :event], %{}, %{status: :ok})
  # :telemetry.execute([:my_app, :event], %{}, %{status: :error})
  # :telemetry.execute([:my_app, :event_span, :start], %{count: 1}, %{})
  # :telemetry.execute([:my_app, :event_span, :stop], %{count: 1}, %{})
  # :telemetry.execute([:my_app, :event_span, :exception], %{count: 1}, %{})

  :ok
end

One thing to note is that for counter metrics, seeding the events in this way will result in the values being initialized to a count of 1. In Grafana graphs showing counters over time, the value resetting to either 0 or 1 will result in a negative delta; one may want to combine idelta with clamp_min (i.e. clamp_min(idelta(<event_name>[$__rate_interval]), 0)) to ensure that the initial value after a restart appears in the graph as no change.