defmodule Towerops.Workers.StaleAgentWorker do
  @moduledoc """
  Oban worker that periodically checks for stale agents.

  Runs every minute via Oban.Plugins.Cron to:
  1. Find agents that haven't sent a heartbeat in over 10 minutes
  2. Log warnings for operators
  3. Broadcast alerts via PubSub for UI updates

  Agents are considered stale if:
  - They are enabled
  - They have checked in at least once (last_seen_at is not nil)
  - They haven't checked in for more than 10 minutes
  """
  use Oban.Worker, queue: :maintenance

  import Ecto.Query

  alias Towerops.Agents.AgentToken
  alias Towerops.Repo

  require Logger

  # Consider agents stale if not seen in 10 minutes
  @stale_threshold_minutes 10

  # Consider agents "newly stale" if not seen in 10-15 minutes (for warning logs)
  @newly_stale_threshold_minutes 15

  @impl Oban.Worker
  def perform(%Oban.Job{}) do
    stale_agents = find_stale_agents()

    _ =
      if !Enum.empty?(stale_agents) do
        process_stale_agents(stale_agents)
      end

    :ok
  end

  defp process_stale_agents(stale_agents) do
    # Separate newly stale (10-15 min) from long-term stale (>15 min)
    {newly_stale, long_term_stale} =
      Enum.split_with(stale_agents, &newly_stale?/1)

    # Log warnings for newly stale agents
    log_newly_stale_agents(newly_stale)

    # Log debug for long-term stale agents (reduce log noise)
    Enum.each(long_term_stale, fn agent -> log_stale_agent(agent, :debug) end)

    # Broadcast all stale agents for UI updates
    _ = Phoenix.PubSub.broadcast(Towerops.PubSub, "agents:health", {:agents_stale, stale_agents})
  end

  defp log_newly_stale_agents([]), do: :ok

  defp log_newly_stale_agents(newly_stale) do
    Enum.each(newly_stale, fn agent -> log_stale_agent(agent, :warning) end)
    Logger.warning("Detected #{length(newly_stale)} newly stale agent(s)")
  end

  @doc """
  Returns a list of stale agent tokens.

  Public for testing and manual inspection.
  """
  @spec find_stale_agents() :: [AgentToken.t()]
  def find_stale_agents do
    cutoff = DateTime.add(DateTime.utc_now(), -@stale_threshold_minutes, :minute)

    AgentToken
    |> where([t], t.enabled == true)
    |> where([t], not is_nil(t.last_seen_at))
    |> where([t], t.last_seen_at < ^cutoff)
    |> preload(:organization)
    |> Repo.all()
  end

  defp newly_stale?(agent) do
    newly_stale_cutoff = DateTime.add(DateTime.utc_now(), -@newly_stale_threshold_minutes, :minute)
    DateTime.after?(agent.last_seen_at, newly_stale_cutoff)
  end

  defp log_stale_agent(agent, level) do
    minutes_since_seen =
      DateTime.utc_now()
      |> DateTime.diff(agent.last_seen_at, :second)
      |> div(60)

    message = "Agent '#{agent.name}' is stale - last seen #{minutes_since_seen} minutes ago"

    metadata = [
      agent_token_id: agent.id,
      agent_name: agent.name,
      organization_id: agent.organization_id,
      last_seen_at: agent.last_seen_at,
      last_ip: agent.last_ip,
      minutes_since_seen: minutes_since_seen
    ]

    case level do
      :warning -> Logger.warning(message, metadata)
      :debug -> Logger.debug(message, metadata)
    end
  end
end