syntropy

defmodule SyntropyWeb.Telemetry do
  @moduledoc """
  Telemetry supervisor: periodic VM measurements plus the in-process
  Prometheus reporter backing `GET /metrics`.

  Metrics cover the BEAM (memory, run queues), Phoenix endpoint latency,
  Ecto query timing (only emitted when persistence is enabled), and the
  custom runtime events defined in `Syntropy.Telemetry` — task runs, LLM
  requests and token usage, webhook deliveries, and event recorder batch
  flushes.
  """

  use Supervisor

  import Telemetry.Metrics

  @reporter :syntropy_prometheus_metrics

  @duration_buckets_ms [
    10,
    50,
    100,
    250,
    500,
    1_000,
    2_500,
    5_000,
    10_000,
    30_000,
    60_000,
    120_000
  ]

  @spec start_link(keyword()) :: Supervisor.on_start()
  def start_link(opts) do
    Supervisor.start_link(__MODULE__, opts, name: __MODULE__)
  end

  @impl true
  def init(_opts) do
    children = [
      # telemetry_poller also emits its default VM measurements
      # ([:vm, :memory], [:vm, :total_run_queue_lengths], ...) on this period.
      {:telemetry_poller, measurements: [], period: 10_000, name: :syntropy_telemetry_poller},
      {TelemetryMetricsPrometheus.Core, metrics: metrics(), name: @reporter}
    ]

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

  @doc """
  Renders the current metric values in Prometheus text exposition format.
  """
  @spec scrape() :: String.t()
  def scrape do
    TelemetryMetricsPrometheus.Core.scrape(@reporter)
  end

  @spec metrics() :: [Telemetry.Metrics.t()]
  def metrics do
    [
      # BEAM
      last_value("vm.memory.total", unit: :byte),
      last_value("vm.total_run_queue_lengths.total"),
      last_value("vm.total_run_queue_lengths.cpu"),
      last_value("vm.total_run_queue_lengths.io"),

      # Phoenix
      distribution("phoenix.endpoint.stop.duration",
        unit: {:native, :millisecond},
        reporter_options: [buckets: @duration_buckets_ms]
      ),

      # Ecto (no-op unless persistence is enabled)
      distribution("syntropy.repo.query.total_time",
        unit: {:native, :millisecond},
        reporter_options: [buckets: @duration_buckets_ms]
      ),

      # Task runs. The histogram carries run count (by status/strategy),
      # total duration, and latency buckets.
      distribution("syntropy.task_run.stop.duration_ms",
        tags: [:status, :strategy],
        reporter_options: [buckets: @duration_buckets_ms],
        description: "Terminal run duration by status and strategy"
      ),
      sum("syntropy.task_run.stop.agent_count",
        tags: [:status],
        description: "Total agents dispatched across terminal runs"
      ),

      # LLM requests
      distribution("syntropy.llm_request.stop.duration_ms",
        tags: [:provider, :status],
        reporter_options: [buckets: @duration_buckets_ms],
        description: "LLM completion call latency by provider and outcome"
      ),
      sum("syntropy.llm_request.tokens.input_tokens", tags: [:provider]),
      sum("syntropy.llm_request.tokens.output_tokens", tags: [:provider]),
      sum("syntropy.llm_request.tokens.total_tokens", tags: [:provider]),

      # Webhook deliveries
      counter("syntropy.webhook_delivery.stop.count",
        tags: [:status],
        description: "Webhook delivery count by final status"
      ),
      sum("syntropy.webhook_delivery.stop.attempts",
        tags: [:status],
        description: "Total delivery attempts (including retries)"
      ),

      # Event recorder. The histogram carries flush count and total flush time.
      sum("syntropy.event_recorder.flush.batch_size",
        description: "Total events flushed to Postgres"
      ),
      distribution("syntropy.event_recorder.flush.duration_ms",
        reporter_options: [buckets: [1, 5, 10, 25, 50, 100, 250, 500, 1_000]]
      )
    ]
  end
end