More tweaks to TCP failover logic. Such edge case.

3 files changed, 31 insertions, 22 deletions

diff --git a/node/Node.cpp b/node/Node.cpp
index a0dd14c1..051cb317 100644
--- a/node/Node.cpp
+++ b/node/Node.cpp
@@ -543,6 +543,8 @@ Node::ReasonForTermination Node::run()
 		_r->timeOfLastResynchronize = Utils::now();
 
 		while (impl->reasonForTermination == NODE_RUNNING) {
+			//TRACE("--- BEGIN main I/O loop");
+
 			if (Utils::fileExists(shutdownIfUnreadablePath.c_str(),false)) {
 				FILE *tmpf = fopen(shutdownIfUnreadablePath.c_str(),"r");
 				if (!tmpf)
diff --git a/node/Peer.cpp b/node/Peer.cpp
index 3fad7c05..f8ff4082 100644
--- a/node/Peer.cpp
+++ b/node/Peer.cpp
@@ -176,9 +176,9 @@ bool Peer::sendPing(const RuntimeEnvironment *_r,uint64_t now)
 	Mutex::Lock _l(_lock);
 	for(std::vector<Path>::iterator p(_paths.begin());p!=_paths.end();++p) {
 		if ((useTcp)||(!p->tcp())) {
+			p->pinged(now); // we log pings sent even if the send "fails", since what we want to track is when we last tried to ping
 			if (_r->sw->sendHELLO(self,*p)) {
 				p->sent(now);
-				p->pinged(now);
 				sent = true;
 			}
 		}
@@ -187,6 +187,32 @@ bool Peer::sendPing(const RuntimeEnvironment *_r,uint64_t now)
 	return sent;
 }
 
+bool Peer::isTcpFailoverTime(const RuntimeEnvironment *_r,uint64_t now) const
+	throw()
+{
+	uint64_t lastResync = _r->timeOfLastResynchronize;
+	if ((now - lastResync) >= ZT_TCP_TUNNEL_FAILOVER_TIMEOUT) {
+		if ((now - _r->timeOfLastPacketReceived) >= ZT_TCP_TUNNEL_FAILOVER_TIMEOUT)
+			return true;
+
+		uint64_t lastUdpPingSent = 0;
+		uint64_t lastUdpReceive = 0;
+
+		{
+			Mutex::Lock _l(_lock);
+			for(std::vector<Path>::const_iterator p(_paths.begin());p!=_paths.end();++p) {
+				if (p->type() == Path::PATH_TYPE_UDP) {
+					lastUdpPingSent = std::max(lastUdpPingSent,p->lastPing());
+					lastUdpReceive = std::max(lastUdpReceive,p->lastReceived());
+				}
+			}
+		}
+
+		return ( (lastUdpPingSent > lastResync) && (lastUdpPingSent > lastUdpReceive) && ((now - lastUdpPingSent) >= ZT_TCP_TUNNEL_FAILOVER_TIMEOUT) );
+	}
+	return false;
+}
+
 void Peer::clean(uint64_t now)
 {
 	Mutex::Lock _l(_lock);
diff --git a/node/Peer.hpp b/node/Peer.hpp
index 8b8bf578..6a57e14a 100644
--- a/node/Peer.hpp
+++ b/node/Peer.hpp
@@ -267,27 +267,8 @@ public:
 	 * @param now Current time
 	 * @return True if it's time to attempt TCP failover (if we have TCP_OUT paths)
 	 */
-	inline bool isTcpFailoverTime(const RuntimeEnvironment *_r,uint64_t now) const
-		throw()
-	{
-		if ((now - _r->timeOfLastResynchronize) >= ZT_TCP_TUNNEL_FAILOVER_TIMEOUT) {
-			uint64_t lastUdpPingSent = 0;
-			uint64_t lastUdpReceive = 0;
-
-			{
-				Mutex::Lock _l(_lock);
-				for(std::vector<Path>::const_iterator p(_paths.begin());p!=_paths.end();++p) {
-					if (p->type() == Path::PATH_TYPE_UDP) {
-						lastUdpPingSent = std::max(lastUdpPingSent,p->lastPing());
-						lastUdpReceive = std::max(lastUdpReceive,p->lastReceived());
-					}
-				}
-			}
-
-			return ( (lastUdpPingSent > lastUdpReceive) && ((now - lastUdpPingSent) >= ZT_TCP_TUNNEL_FAILOVER_TIMEOUT) );
-		}
-		return false;
-	}
+	bool isTcpFailoverTime(const RuntimeEnvironment *_r,uint64_t now) const
+		throw();
 
 	/**
 	 * @return Current latency or 0 if unknown (max: 65535)