[RTC-474] Expand room metrics (#160)

lib/jellyfish/room.ex

@@ -216,6 +216,7 @@ defmodule Jellyfish.Room do
           :ok = Engine.add_endpoint(state.engine_pid, peer.engine_endpoint, id: peer_id)
 
           Logger.info("Peer #{inspect(peer_id)} connected")
+          :telemetry.execute([:jellyfish, :room], %{peer_connects: 1}, %{room_id: state.id})
 
           {:ok, state}
 
@@ -433,6 +434,7 @@ defmodule Jellyfish.Room do
         {peer_id, peer} ->
           :ok = Engine.remove_endpoint(state.engine_pid, peer_id)
           Event.broadcast_server_notification({:peer_disconnected, state.id, peer_id})
+          :telemetry.execute([:jellyfish, :room], %{peer_disconnects: 1}, %{room_id: state.id})
           peer = %{peer | status: :disconnected, socket_pid: nil}
 
           put_in(state, [:peers, peer_id], peer)
@@ -737,11 +739,14 @@ defmodule Jellyfish.Room do
 
     Logger.info("Removed peer #{inspect(peer_id)} from room #{inspect(state.id)}")
 
-    if peer.status == :connected and reason == :peer_removed,
-      do: Event.broadcast_server_notification({:peer_disconnected, state.id, peer_id})
+    if peer.status == :connected and reason == :peer_removed do
+      Event.broadcast_server_notification({:peer_disconnected, state.id, peer_id})
+      :telemetry.execute([:jellyfish, :room], %{peer_disconnects: 1}, %{room_id: state.id})
+    end
 
     with {:peer_crashed, crash_reason} <- reason do
       Event.broadcast_server_notification({:peer_crashed, state.id, peer_id, crash_reason})
+      :telemetry.execute([:jellyfish, :room], %{peer_crashes: 1}, %{room_id: state.id})
     end
 
     state

lib/jellyfish/room_service.ex

@@ -186,7 +186,8 @@ defmodule Jellyfish.RoomService do
         [:jellyfish, :room],
         %{
           peers: peer_count,
-          peer_time_total: peer_count * @metric_interval_in_seconds
+          peer_time: peer_count * @metric_interval_in_seconds,
+          duration: @metric_interval_in_seconds
         },
         %{room_id: room.id}
       )
@@ -223,14 +224,7 @@ defmodule Jellyfish.RoomService do
   end
 
   defp clear_room_metrics(room_id) do
-    :telemetry.execute(
-      [:jellyfish, :room],
-      %{
-        peers: 0,
-        peer_time_total: 0
-      },
-      %{room_id: room_id}
-    )
+    :telemetry.execute([:jellyfish, :room], %{peers: 0}, %{room_id: room_id})
   end
 
   defp find_best_node(node_resources) do

lib/jellyfish_web/telemetry.ex

@@ -125,37 +125,52 @@ defmodule JellyfishWeb.Telemetry do
           description: "Total HTTP traffic sent (bytes)"
         ),
         last_value("jellyfish.rooms",
-          description: "Amount of rooms currently present in Jellyfish"
+          description: "Number of rooms currently present in Jellyfish"
         ),
 
         # FIXME: Prometheus warns about using labels to store dimensions with high cardinality,
         # such as UUIDs. For more information refer here: https://prometheus.io/docs/practices/naming/#labels
         last_value("jellyfish.room.peers",
           tags: [:room_id],
-          description: "Amount of peers currently present in a given room"
+          description: "Number of peers currently present in a given room"
         ),
         sum("jellyfish.room.peer_time.total.seconds",
           event_name: [:jellyfish, :room],
-          measurement: :peer_time_total,
+          measurement: :peer_time,
           tags: [:room_id],
           description: "Total peer time accumulated for a given room (seconds)"
+        ),
+        sum("jellyfish.room.duration.seconds",
+          event_name: [:jellyfish, :room],
+          measurement: :duration,
+          tags: [:room_id],
+          description: "Duration of a given room (seconds)"
+        ),
+        sum("jellyfish.room.peer_connects.total",
+          event_name: [:jellyfish, :room],
+          measurement: :peer_connects,
+          tags: [:room_id],
+          description:
+            "Number of PeerConnected events emitted during the lifetime of a given room"
+        ),
+        sum("jellyfish.room.peer_disconnects.total",
+          event_name: [:jellyfish, :room],
+          measurement: :peer_disconnects,
+          tags: [:room_id],
+          description:
+            "Number of PeerDisconnected events emitted during the lifetime of a given room"
+        ),
+        sum("jellyfish.room.peer_crashes.total",
+          event_name: [:jellyfish, :room],
+          measurement: :peer_crashes,
+          tags: [:room_id],
+          description: "Number of PeerCrashed events emitted during the lifetime of a given room"
         )
       ]
   end
 
   def default_webrtc_metrics() do
-    :telemetry.execute(
-      [Membrane.ICE, :ice, :payload, :sent],
-      %{
-        bytes: 0
-      }
-    )
-
-    :telemetry.execute(
-      [Membrane.ICE, :ice, :payload, :received],
-      %{
-        bytes: 0
-      }
-    )
+    :telemetry.execute(@ice_sent_event, %{bytes: 0})
+    :telemetry.execute(@ice_received_event, %{bytes: 0})
   end
 end