Dont request keyframes if the stream is inactive or if we are currently receiving a keyframe.

webrtc/video/video_receive_stream.cc

Index: webrtc/video/video_receive_stream.cc
diff --git a/webrtc/video/video_receive_stream.cc b/webrtc/video/video_receive_stream.cc
index cee62d297c796c4fec7d42ddfb7e9ff4418103fe..e48143cf64862a73651856e251af3aec27ef0978 100644
--- a/webrtc/video/video_receive_stream.cc
+++ b/webrtc/video/video_receive_stream.cc
@@ -482,9 +482,28 @@ bool VideoReceiveStream::Decode() {
     if (video_receiver_.Decode(frame.get()) == VCM_OK)
       rtp_stream_receiver_.FrameDecoded(frame->picture_id);
   } else {
-    LOG(LS_WARNING) << "No decodable frame in " << kMaxWaitForFrameMs
-                    << " ms, requesting keyframe.";
-    RequestKeyFrame();
+    int64_t now_ms = clock_->TimeInMilliseconds();
+    rtc::Optional<int64_t> last_packet_ms =
+        rtp_stream_receiver_.LastReceivedPacketMs();
+    rtc::Optional<int64_t> last_keyframe_packet_ms =
+        rtp_stream_receiver_.LastReceivedKeyframePacketMs();
+
+    // To avoid spamming keyframe requests for a stream that is not active we
+    // check if we have received a packet within the last 5 seconds.
+    bool stream_is_active = last_packet_ms && now_ms - *last_packet_ms < 5000;
+
+    // If we recently (within |kMaxWaitForFrameMs|) have been receiving packets
+    // belonging to a keyframe then we assume a keyframe is being received right
+    // now.
+    bool receiving_keyframe =
+        last_keyframe_packet_ms &&
+        now_ms - *last_keyframe_packet_ms < kMaxWaitForFrameMs;
+
+    if (stream_is_active && !receiving_keyframe) {

[philipel, 2017/04/28 12:11:29]

I know we talked about this as a potential problem when it comes to
screensharing (spamming keyframe requests and in turn never receive a full
keyframe), but I'm not sure if this is a good idea.

Since we resend packets instead of padding we might end up in a state where
packets belonging to a keyframe is received, which would cause us to not send a
keyframe request even if we might need it.

[ilnik, 2017/04/28 12:40:55]

How about a following workaround: prohibit keyframe packets to be used instead
of padding packets?

[philipel, 2017/04/28 12:58:24]

I just realized that we can update the timestamps after that packet has been
actually inserted.

+      LOG(LS_WARNING) << "No decodable frame in " << kMaxWaitForFrameMs
+                      << " ms, requesting keyframe.";
+      RequestKeyFrame();
+    }
   }
   return true;
 }