raven_integration_xmpp

defmodule Integrations.Xmpp do
  @moduledoc """
  XMPP server monitor.

  Opens a TCP connection to an XMPP server and sends an XML stream opening.
  A healthy server responds with its own `<stream:stream ...>` header and
  advertises its supported features. The check confirms the server is alive
  and speaking the XMPP protocol without performing authentication.

  Uses raw `:gen_tcp` — no external dependencies required.

  Collection only — see `Integrations.Xmpp.Display` (same package) for the
  dashboard panel. `Display.BundledDefault` auto-hooks it whenever this
  monitor starts.

  ## Params

    * `:host`          — XMPP server hostname or IP. Required.
    * `:port`          — Port number. Defaults to `5222` (client-to-server).
                         Use `5269` for server-to-server federation checks.
    * `:domain`        — XMPP domain to use in the stream `to` attribute.
                         Defaults to `:host`. Set this when the domain differs
                         from the server address (virtual hosting).
    * `:timeout_ms`    — TCP connect and receive timeout in milliseconds.
                         Defaults to `5000`.
    * `:latency_degraded_ms` — Time-to-first-byte threshold for `:degraded`.
                               Defaults to `1000`.

  ## Health signal

    * `:up`       — Server responded with a valid XMPP stream header and
                    latency is within threshold.
    * `:degraded` — Server responded but latency exceeded threshold, or
                    the response was not a well-formed stream header.
    * `:down`     — Connection refused, timeout, or no response.
  """

  use CodeNameRaven.Monitor

  @default_port            5222
  @default_timeout_ms      5_000
  @default_lat_degraded_ms 1_000

  @stream_ns "jabber:client"

  @impl true
  def params_template do
    %{host: "", port: "5222", domain: "", timeout_ms: "5000"}
  end

  @impl true
  def params_schema do
    [
      host:                [type: :string,          required: true,  doc: "XMPP server hostname or IP"],
      port:                [type: :non_neg_integer, default: 5222,   doc: "Port (5222 client, 5269 server federation)"],
      domain:              [type: :string,          required: false,  doc: "XMPP domain for stream header (defaults to host)"],
      timeout_ms:          [type: :non_neg_integer, default: 5_000,  doc: "TCP connect and receive timeout in milliseconds"],
      latency_degraded_ms: [type: :non_neg_integer, default: 1_000,  doc: "TTFB threshold for degraded"]
    ]
  end

  @impl true
  def target_uri(params) do
    host = get_param(params, :host)
    port = parse_int(get_param(params, :port), @default_port)
    if host, do: {:ok, "xmpp://#{host}:#{port}"}, else: :none
  end

  @impl true
  def identity_params(params) do
    %{
      host:   get_param(params, :host),
      port:   parse_int(get_param(params, :port), @default_port)
    }
  end

  # ---------------------------------------------------------------------------
  # Collect
  # ---------------------------------------------------------------------------

  @impl true
  def collect(params, state) do
    host = get_param(params, :host)

    if is_nil(host) do
      {:error, "missing required param :host", state}
    else
      port       = parse_int(get_param(params, :port), @default_port)
      timeout_ms = parse_int(get_param(params, :timeout_ms), @default_timeout_ms)
      domain     = get_param(params, :domain) || host

      tcp_opts = [:binary, packet: :raw, active: false, send_timeout: timeout_ms]

      started_at = System.monotonic_time(:millisecond)

      case :gen_tcp.connect(to_charlist(host), port, tcp_opts, timeout_ms) do
        {:ok, socket} ->
          result = do_xmpp_check(socket, domain, timeout_ms, started_at)
          :gen_tcp.close(socket)
          case result do
            {:ok, data}      -> {:ok, data, state}
            {:error, reason} -> {:error, reason, state}
          end

        {:error, :econnrefused} ->
          {:error, "connection refused on port #{port}", state}

        {:error, :timeout} ->
          {:error, "connection timed out", state}

        {:error, :nxdomain} ->
          {:error, "hostname not found: #{host}", state}

        {:error, reason} ->
          {:error, inspect(reason), state}
      end
    end
  end

  # ---------------------------------------------------------------------------
  # Healthy?
  # ---------------------------------------------------------------------------

  @impl true
  def healthy?(result) do
    lat_deg = result[:latency_degraded_ms] || @default_lat_degraded_ms

    cond do
      not result.stream_ok        -> :degraded
      result.latency_ms >= lat_deg -> :degraded
      true                         -> :up
    end
  end

  # ---------------------------------------------------------------------------
  # Metrics
  # ---------------------------------------------------------------------------

  @impl true
  def metrics(result) do
    %{latency_ms: result.latency_ms}
  end

  # ---------------------------------------------------------------------------
  # XMPP protocol
  # ---------------------------------------------------------------------------

  defp do_xmpp_check(socket, domain, timeout_ms, started_at) do
    stream_open = "<stream:stream xmlns='#{@stream_ns}' xmlns:stream='http://etherx.jabber.org/streams' to='#{domain}' version='1.0'>"

    :gen_tcp.send(socket, stream_open)

    case recv_until_stream_header(socket, timeout_ms, "") do
      {:ok, data} ->
        ttfb_ms   = System.monotonic_time(:millisecond) - started_at
        stream_ok = String.contains?(data, "<stream:stream") or
                    String.contains?(data, "<stream:error")

        server_id = extract_attribute(data, "id")
        from      = extract_attribute(data, "from")
        features  = extract_features(data)

        {:ok, %{
          latency_ms: ttfb_ms,
          stream_ok:  stream_ok,
          server_id:  server_id || from,
          features:   features
        }}

      {:error, reason} ->
        {:error, reason}
    end
  end

  # Read until we have seen the stream:stream opening tag close
  # (it's not a complete XML doc, so we just collect until we see '>')
  defp recv_until_stream_header(socket, timeout_ms, acc) do
    case :gen_tcp.recv(socket, 0, timeout_ms) do
      {:ok, data} ->
        acc = acc <> data
        # Stream header is complete once we see the closing '>' of the opening tag
        # We also accept partial data that has the stream:stream element
        if String.contains?(acc, "<stream:stream") and String.contains?(acc, ">") do
          {:ok, acc}
        else
          recv_until_stream_header(socket, timeout_ms, acc)
        end

      {:error, :timeout} -> {:error, "timed out waiting for stream header"}
      {:error, reason}   -> {:error, inspect(reason)}
    end
  end

  defp extract_attribute(xml, attr) do
    case Regex.run(~r/\b#{attr}="([^"]*)"/, xml) do
      [_, value] -> value
      _          -> nil
    end
  end

  defp extract_features(xml) do
    # Extract feature names from <stream:features>...</stream:features>
    case Regex.run(~r|<stream:features>(.*?)</stream:features>|s, xml) do
      [_, features_xml] ->
        Regex.scan(~r/<(\w+[:\w]*)[^>]*\/>|<(\w+[:\w]*)[^>]*>/, features_xml)
        |> Enum.flat_map(fn
          [_, feat, ""] -> [feat]
          [_, "", feat] -> [feat]
          _             -> []
        end)
        |> Enum.reject(&(&1 in ["stream:features", ""]))
        |> Enum.uniq()

      _ -> []
    end
  end

  # ---------------------------------------------------------------------------
  # Helpers
  # ---------------------------------------------------------------------------

  defp get_param(params, key) when is_atom(key) do
    v = params[key] || params[to_string(key)]
    if is_binary(v) and String.trim(v) == "", do: nil, else: v
  end

  defp parse_int(nil, default), do: default
  defp parse_int(v, _) when is_integer(v), do: v
  defp parse_int(v, default) when is_binary(v) do
    case Integer.parse(v) do
      {n, _} -> n
      :error -> default
    end
  end
  defp parse_int(_, default), do: default
end