fix keepalive issue when there is network/cpu bottleneck

agent/agent-priv.h

@@ -112,6 +112,7 @@ struct _NiceAgent
   GStaticRecMutex mutex;                 /* Mutex used for thread-safe lib */
   NiceCompatibility compatibility; /* property: Compatibility mode */
   StunAgent stun_agent;            /* STUN agent */
+  gboolean media_after_tick;       /* Received media after keepalive tick */
 #ifdef HAVE_GUPNP
   GUPnPSimpleIgdThread* upnp;	   /* GUPnP Single IGD agent */
   gboolean upnp_enabled;           /* whether UPnP discovery is enabled */

agent/agent.c

@@ -537,6 +537,7 @@ nice_agent_init (NiceAgent *agent)
   agent->conncheck_timer_source = NULL;
   agent->keepalive_timer_source = NULL;
   agent->refresh_list = NULL;
+  agent->media_after_tick = FALSE;
 
   agent->compatibility = NICE_COMPATIBILITY_DRAFT19;
 
@@ -1790,6 +1791,8 @@ _nice_agent_recv (
     }
   }
 
+  agent->media_after_tick = TRUE;
+
   if (stun_message_validate_buffer_length ((uint8_t *) buf, (size_t) len) != len)
     /* If the retval is no 0, its not a valid stun packet, probably data */
     return len;

agent/conncheck.c

@@ -489,11 +489,25 @@ static gboolean priv_conn_keepalive_retransmissions_tick (gpointer pointer)
         stun_message_id (&pair->keepalive.stun_message, id);
         stun_agent_forget_transaction (&pair->keepalive.agent->stun_agent, id);
 
-        nice_debug ("Agent %p : Keepalive conncheck timed out!! "
-            "peer probably lost connection", pair->keepalive.agent);
-        agent_signal_component_state_change (pair->keepalive.agent,
-            pair->keepalive.stream_id, pair->keepalive.component_id,
-            NICE_COMPONENT_STATE_FAILED);
+        if (pair->keepalive.agent->media_after_tick) {
+          nice_debug ("Agent %p : Keepalive conncheck timed out!! "
+              "but media was received. Suspecting keepalive lost because of "
+              "network bottleneck", pair->keepalive.agent);
+
+          if (pair->keepalive.tick_source) {
+            g_source_destroy (pair->keepalive.tick_source);
+            g_source_unref (pair->keepalive.tick_source);
+            pair->keepalive.tick_source = NULL;
+          }
+          pair->keepalive.stun_message.buffer = NULL;
+
+        } else {
+          nice_debug ("Agent %p : Keepalive conncheck timed out!! "
+              "peer probably lost connection", pair->keepalive.agent);
+          agent_signal_component_state_change (pair->keepalive.agent,
+              pair->keepalive.stream_id, pair->keepalive.component_id,
+              NICE_COMPONENT_STATE_FAILED);
+        }
         break;
       }
     case STUN_USAGE_TIMER_RETURN_RETRANSMIT:
@@ -589,6 +603,8 @@ static gboolean priv_conn_keepalive_tick_unlocked (NiceAgent *agent)
             if (buf_len > 0) {
               stun_timer_start (&p->keepalive.timer);
 
+              agent->media_after_tick = FALSE;
+
               /* send the conncheck */
               nice_socket_send (p->local->sockptr, &p->remote->addr,
                   buf_len, (gchar *)p->keepalive.stun_buffer);