Telemetry.Metrics definitions for the events emitted by Agens.

The list returned by metrics/0 is intended to be supplied to a Telemetry.Metrics.Reporter (for example TelemetryMetricsPrometheus, TelemetryMetricsStatsd, or your own reporter) in your application's supervision tree.

Wiring into a parent application

Add a Telemetry.Metrics.Supervisor (or reporter of your choice) to your application's supervision tree and feed it Agens.Metrics.metrics/0:

defmodule MyApp.Application do
  use Application

  def start(_type, _args) do
    children = [
      {Agens.Supervisor, name: Agens.Supervisor},
      {TelemetryMetricsPrometheus, metrics: Agens.Metrics.metrics()}
      # ...your own children
    ]

    Supervisor.start_link(children, strategy: :one_for_one)
  end
end

To combine Agens metrics with metrics from your own app:

def metrics do
  Agens.Metrics.metrics() ++ my_app_metrics()
end

Event coverage

Metrics are emitted for the following event prefixes:

• [:agens, :serving, ...] - Serving lifecycle, enqueue, and result duration / exceptions (via :telemetry.span/3).
• [:agens, :job, ...] - Job lifecycle, status changes, yields and errors.
• [:agens, :node, ...] - Per-Node start, result, and retry events.
• [:agens, :sub, ...] - Sub-Job lifecycle from the parent's perspective: dispatch (:start), Agens.Serving.handle_sub/3 invocation (:handle), completion routed back to parent (:done), and propagated failure (:error).
• [:agens, :tool, :call, ...] - Tool calls with duration and exception coverage (span).
• [:agens, :resource, :load, ...] - Resource loads with duration and exception coverage (span).