prop/lib/microwaveprop/workers/gefs_fetch_worker.ex

defmodule Microwaveprop.Workers.GefsFetchWorker do
  @moduledoc """
  Fetches one GEFS ensemble-mean forecast hour, extracts the CONUS
  grid, and upserts into `gefs_profiles`.

  Each `perform/1` handles a single `(run_time, forecast_hour)` pair so
  a failure at one hour doesn't block the others — the cron seeder
  enqueues the full `f024..f168` set per run. Per-hour independence
  also keeps memory bounded: the wgrib2 extraction step for a GEFS
  0.5° file is small (<1 MB after byte-range slicing) but the target
  0.125° grid still carries ~200k cells.
  """

  use Oban.Pro.Worker,
    queue: :gefs,
    max_attempts: 5,
    unique: [period: 3600, states: :incomplete]

  alias Microwaveprop.Propagation
  alias Microwaveprop.Weather
  alias Microwaveprop.Weather.GefsClient
  alias Microwaveprop.Weather.SoundingParams

  require Logger

  @valid_run_hours [0, 6, 12, 18]

  @impl Oban.Pro.Worker
  def process(%Oban.Job{args: %{"run_time" => run_time_iso, "forecast_hour" => fh}}) when is_integer(fh) do
    with {:ok, run_time, _} <- DateTime.from_iso8601(run_time_iso),
         :ok <- validate_run_hour(run_time.hour) do
      run_date = DateTime.to_date(run_time)
      run_hour = run_time.hour

      Logger.info("GefsFetch: run_time=#{run_time_iso} fh=#{fh}")

      case GefsClient.fetch_grid_profiles(run_date, run_hour, fh) do
        {:ok, profiles} ->
          valid_time = run_time |> DateTime.add(fh * 3600, :second) |> DateTime.truncate(:second)
          score_and_persist(profiles, run_time, fh, valid_time)
          :ok

        {:error, :wgrib2_not_available} ->
          Logger.warning("GefsFetch: wgrib2 missing on host — cancelling job")
          {:cancel, :wgrib2_not_available}

        {:error, reason} ->
          handle_error(reason, "fh=#{fh} @ #{run_time_iso}")
      end
    else
      _ -> {:cancel, :invalid_args}
    end
  end

  def process(%Oban.Job{args: args}) when args == %{} do
    seed_extended_horizon()
  end

  def process(%Oban.Job{args: _args}), do: {:cancel, :invalid_args}

  @doc """
  Forecast hours enqueued per GEFS run: f024 through f168 at the
  6-hour cadence GEFS publishes at. Covers Day 2-7 — the window HRRR
  can't reach. Starts at f024 so the near-term (f00-f18) stays
  HRRR-owned and the handoff lines up on the f018 boundary.
  """
  @spec extended_horizon_hours() :: [non_neg_integer()]
  def extended_horizon_hours, do: Enum.to_list(24..168//6)

  @doc """
  Enqueues one fetch job per extended-horizon forecast hour for the
  most recent GEFS run that should already be published on NOMADS
  (run publication lags ~3-4 hours behind the run time; the seeder
  targets runs 5+ hours in the past for safety).
  """
  @spec seed_extended_horizon() :: :ok
  def seed_extended_horizon do
    run_time = most_recent_available_run(DateTime.utc_now())

    Logger.info("GefsFetch: seeding extended-horizon chain for run_time=#{run_time}")

    Enum.each(extended_horizon_hours(), fn fh ->
      %{"run_time" => DateTime.to_iso8601(run_time), "forecast_hour" => fh}
      |> new()
      |> Oban.insert()
    end)

    :ok
  end

  @doc """
  Picks the most-recent 00/06/12/18Z GEFS run whose file set should be
  fully published given an ~5-hour publication lag. Exposed so tests
  can pin a deterministic "now".
  """
  @spec most_recent_available_run(DateTime.t()) :: DateTime.t()
  def most_recent_available_run(%DateTime{} = now) do
    now
    |> DateTime.add(-5 * 3600, :second)
    |> GefsClient.nearest_run()
  end

  @doc """
  Converts a `GefsClient.build_profile/1` result (already tagged with
  `lat` and `lon`) into a full `gefs_profiles` row attr map. Exposed
  for unit testing.
  """
  @spec build_profile_attrs(DateTime.t(), non_neg_integer(), map()) :: map()
  def build_profile_attrs(%DateTime{} = run_time, forecast_hour, profile) when is_integer(forecast_hour) do
    valid_time = DateTime.add(run_time, forecast_hour * 3600, :second)
    derived = SoundingParams.derive(profile.profile || [])

    base = %{
      run_time: DateTime.truncate(run_time, :second),
      forecast_hour: forecast_hour,
      valid_time: DateTime.truncate(valid_time, :second),
      lat: profile.lat,
      lon: profile.lon,
      profile: profile.profile,
      surface_temp_c: profile.surface_temp_c,
      surface_dewpoint_c: profile.surface_dewpoint_c,
      surface_pressure_mb: profile.surface_pressure_mb,
      pwat_mm: profile[:pwat_mm],
      wind_u_mps: profile[:wind_u] || profile[:wind_u_mps],
      wind_v_mps: profile[:wind_v] || profile[:wind_v_mps],
      cloud_cover_pct: profile[:cloud_cover_pct],
      precip_mm: profile[:precip_mm]
    }

    case derived do
      nil ->
        base

      params ->
        Map.merge(base, %{
          surface_refractivity: params.surface_refractivity,
          min_refractivity_gradient: params.min_refractivity_gradient,
          ducting_detected: params.ducting_detected,
          duct_characteristics: params.duct_characteristics
        })
    end
  end

  # Runs the propagation scorer across the fetched GEFS grid and
  # persists scores to the `.prop` file for this `valid_time` so the
  # extended-horizon outlook picks them up via the usual
  # `Propagation.scores_at/3` path. Also upserts raw profiles to
  # `gefs_profiles` for future per-contact extended-horizon lookups.
  defp score_and_persist(profiles, run_time, fh, valid_time) do
    grid_data = profiles_to_grid(profiles)
    supervisor = {:via, PartitionSupervisor, {Microwaveprop.TaskSupervisor, self()}}

    scores =
      supervisor
      |> Task.Supervisor.async_stream_nolink(
        grid_data,
        fn {{lat, lon}, profile} ->
          profile
          |> Propagation.score_grid_point(valid_time, lat, lon)
          |> Enum.map(fn r ->
            %{
              lat: lat,
              lon: lon,
              valid_time: valid_time,
              band_mhz: r.band_mhz,
              score: r.score,
              factors: nil
            }
          end)
        end,
        max_concurrency: System.schedulers_online() * 2,
        timeout: 30_000
      )
      |> Stream.flat_map(fn
        {:ok, results} ->
          results

        {:exit, reason} ->
          Logger.error(
            "GefsFetch scoring task crashed: fh=#{fh} valid_time=#{inspect(valid_time)} reason=#{inspect(reason)}"
          )

          []
      end)
      |> Enum.to_list()

    _ =
      case Propagation.replace_scores(scores, valid_time) do
        {:ok, count} ->
          Logger.info("GefsFetch: wrote #{count} scores for fh=#{fh} @ #{valid_time}")
          broadcast_updated(valid_time)

        error ->
          Logger.error("GefsFetch: replace_scores failed fh=#{fh}: #{inspect(error)}")
      end

    attrs = Enum.map(profiles, &build_profile_attrs(run_time, fh, &1))
    {count, _} = Weather.upsert_gefs_profiles_batch(attrs)
    Logger.info("GefsFetch: upserted #{count}/#{length(attrs)} profile rows for fh=#{fh}")
  end

  defp profiles_to_grid(profiles) do
    Map.new(profiles, fn profile ->
      {{profile.lat, profile.lon}, profile}
    end)
  end

  defp broadcast_updated(valid_time) do
    Phoenix.PubSub.broadcast(
      Microwaveprop.PubSub,
      "propagation:updated",
      {:propagation_updated, [valid_time]}
    )
  end

  defp validate_run_hour(hour) when hour in @valid_run_hours, do: :ok
  defp validate_run_hour(_), do: {:error, :invalid_run_hour}

  defp handle_error(reason, label) do
    if transient?(reason) do
      Logger.error("GefsFetch transient error for #{label}: #{inspect(reason)}")
      {:error, reason}
    else
      Logger.warning("GefsFetch permanent failure for #{label}: #{inspect(reason)}")
      {:cancel, reason}
    end
  end

  defp transient?(%{__exception__: true}), do: true
  defp transient?("GEFS idx HTTP " <> status), do: server_error?(status)
  defp transient?("GEFS grib HTTP " <> status), do: server_error?(status)
  defp transient?(_), do: false

  defp server_error?(status) do
    case Integer.parse(status) do
      {code, _} when code in [429, 500, 502, 503, 504] -> true
      _ -> false
    end
  end
end