More reliable keep-alive video packets.

remoting/host/video_scheduler.h

Index: remoting/host/video_scheduler.h
diff --git a/remoting/host/video_scheduler.h b/remoting/host/video_scheduler.h
index f2eabd9744bbbcc94b178944ab0f4d48e1c55286..7688bad6dee599309a9fc48aa8914f4e0f3e3d2c 100644
--- a/remoting/host/video_scheduler.h
+++ b/remoting/host/video_scheduler.h
@@ -140,7 +140,13 @@ class VideoScheduler : public base::RefCountedThreadSafe<VideoScheduler>,
 
   // Callback passed to |video_stub_| for the last packet in each frame, to
   // rate-limit frame captures to network throughput.
-  void VideoFrameSentCallback();
+  void OnVideoPacketSent();
+
+  // Called by |keep_alive_timer_|.
+  void SendKeepAlivePacket();
+
+  // Callback for |video_stub_| called after a keep-alive packet is sent.
+  void OnKeepAlivePacketSent();
 
   // Send updated cursor shape to client.
   void SendCursorShape(scoped_ptr<protocol::CursorShapeInfo> cursor_shape);
@@ -173,6 +179,10 @@ class VideoScheduler : public base::RefCountedThreadSafe<VideoScheduler>,
   // Timer used to schedule CaptureNextFrame().
   scoped_ptr<base::OneShotTimer<VideoScheduler> > capture_timer_;
 
+  // Timer used to insure that we occasionally send a frame to the client

[Wez, 2014/05/23 00:45:24]

insure -> ensure

[Wez, 2014/05/23 00:45:24]

nit: suggest "occasionally send a frame" -> "send empty keep-alive frames"

[Sergey Ulanov, 2014/05/23 01:27:13]

Done.

[Sergey Ulanov, 2014/05/23 01:27:13]

Done.

+  // even when the video stream is paused or encoder is busy.

[Wez, 2014/05/23 00:45:24]

Why would we want to send frames when the video stream is paused? The point of
pausing is to stop video packets causing consumption of bandwidth, CPU &
battery.

[Sergey Ulanov, 2014/05/23 01:27:13]

That is to fix crbug.com/376528. Libjingle has to send pings on each transport
channel every 480ms anyway, so I don't think this change will have any
measurable effect on resources consumption. 

Ideally we should be using state of the channel to detect when connection is
broken, but that would require changes in libjingle to get more detailed
notifications about state of the connection. That's why we decided to use frames
stream to monitor status of connection.

We could also change the client to disable reconnect logic in case when video is
paused, but it would be more complicated change, not something we would want to
merge to M36.

+  scoped_ptr<base::DelayTimer<VideoScheduler> > keep_alive_timer_;
+
   // The number of frames being processed, i.e. frames that we are currently
   // capturing, encoding or sending. The value is capped at 2 to minimize
   // latency.