diff --git a/router/fastdp.go b/router/fastdp.go
index 3de8eee2..84f5098b 100644
--- a/router/fastdp.go
+++ b/router/fastdp.go
@@ -582,9 +582,11 @@ type fastDatapathForwarder struct {
 	ackedHeartbeat    bool
 	stopChan          chan struct{}
 	stopped           bool
+	healthy           bool
 
 	establishedChan chan struct{}
 	errorChan       chan error
+	healthChan      chan bool
 }
 
 func (fastdp fastDatapathOverlay) PrepareConnection(params mesh.OverlayConnectionParams) (mesh.OverlayConnection, error) {
@@ -622,6 +624,7 @@ func (fastdp fastDatapathOverlay) PrepareConnection(params mesh.OverlayConnectio
 		connUID:        params.ConnUID,
 		vxlanVportID:   vxlanVportID,
 		sessionKey:     params.SessionKey,
+		healthy:        true,
 
 		remoteAddr:        remoteAddr,
 		heartbeatInterval: FastHeartbeat,
@@ -629,6 +632,7 @@ func (fastdp fastDatapathOverlay) PrepareConnection(params mesh.OverlayConnectio
 
 		establishedChan: make(chan struct{}),
 		errorChan:       make(chan error, 1),
+		healthChan:      make(chan bool),
 	}
 
 	return fwd, nil
@@ -699,9 +703,12 @@ func (fwd *fastDatapathForwarder) ErrorChannel() <-chan error {
 	return fwd.errorChan
 }
 
+func (fwd *fastDatapathForwarder) HealthChannel() <-chan bool {
+	return fwd.healthChan
+}
+
 func (fwd *fastDatapathForwarder) doHeartbeats() {
 	var err error
-
 	for err == nil {
 		select {
 		case <-fwd.heartbeatTimer.C:
@@ -711,7 +718,12 @@ func (fwd *fastDatapathForwarder) doHeartbeats() {
 			fwd.heartbeatTimer.Reset(fwd.heartbeatInterval)
 
 		case <-fwd.heartbeatTimeout.C:
-			err = fmt.Errorf("timed out waiting for vxlan heartbeat")
+			fwd.healthChan <- false
+			if fwd.healthy {
+				fwd.healthy = false
+			} else {
+				err = fmt.Errorf("timed out waiting for vxlan heartbeat")
+			}
 
 		case <-fwd.stopChan:
 			return
@@ -808,6 +820,10 @@ func (fwd *fastDatapathForwarder) handleVxlanSpecialPacket(frame []byte, sender
 	// heartbeatTimeout
 	if fwd.heartbeatTimeout != nil {
 		fwd.heartbeatTimeout.Reset(HeartbeatTimeout)
+		if !fwd.healthy {
+			fwd.healthy = true
+			fwd.healthChan <- true
+		}
 	}
 }
 
diff --git a/router/network_overlay.go b/router/network_overlay.go
index 7ed00e42..8ec2f426 100644
--- a/router/network_overlay.go
+++ b/router/network_overlay.go
@@ -31,6 +31,8 @@ type OverlayForwarder interface {
 	// Confirm().  The return value nil means the key could not be
 	// handled by this forwarder.
 	Forward(ForwardPacketKey) FlowOp
+
+	HealthChannel() <-chan bool
 }
 
 type NullNetworkOverlay struct{ mesh.NullOverlay }
diff --git a/router/overlay_switch.go b/router/overlay_switch.go
index 20f7dcbb..04d7288b 100644
--- a/router/overlay_switch.go
+++ b/router/overlay_switch.go
@@ -142,6 +142,7 @@ type overlaySwitchForwarder struct {
 	alreadyEstablished bool
 	establishedChan    chan struct{}
 	errorChan          chan error
+	healthChan         chan bool
 }
 
 // A subsidiary forwarder
@@ -154,6 +155,9 @@ type subForwarder struct {
 
 	// closed to tell the forwarder monitor goroutine to stop
 	stopChan chan<- struct{}
+
+	// set to true to skip this forwarder in chooseBest()
+	onHold bool
 }
 
 // An event from a subsidiary forwarder
@@ -166,6 +170,9 @@ type subForwarderEvent struct {
 
 	// is this an error event?
 	err error
+
+	// event to indicate if forwarder is in healthy state to be considered for best forwarded
+	healthy bool
 }
 
 func (osw *OverlaySwitch) PrepareConnection(params mesh.OverlayConnectionParams) (mesh.OverlayConnection, error) {
@@ -239,6 +246,7 @@ func (osw *OverlaySwitch) PrepareConnection(params mesh.OverlayConnectionParams)
 
 func monitorForwarder(index int, eventsChan chan<- subForwarderEvent, stopChan <-chan struct{}, fwd OverlayForwarder) {
 	establishedChan := fwd.EstablishedChannel()
+	healthChan := fwd.HealthChannel()
 loop:
 	for {
 		e := subForwarderEvent{index: index}
@@ -248,6 +256,9 @@ loop:
 			e.established = true
 			establishedChan = nil
 
+		case healthy := <-healthChan:
+			e.healthy = healthy
+
 		case err := <-fwd.ErrorChannel():
 			e.err = err
 
@@ -282,6 +293,8 @@ loop:
 				fwd.established(e.index)
 			case e.err != nil:
 				fwd.error(e.index, e.err)
+			default:
+				fwd.healthCheck(e.index, e.healthy)
 			}
 		}
 	}
@@ -318,6 +331,21 @@ func (fwd *overlaySwitchForwarder) error(index int, err error) {
 	fwd.chooseBest()
 }
 
+func (fwd *overlaySwitchForwarder) healthCheck(index int, healthy bool) {
+	fwd.lock.Lock()
+	defer fwd.lock.Unlock()
+
+	if healthy == true && fwd.forwarders[index].onHold {
+		fwd.forwarders[index].onHold = false
+		fwd.chooseBest()
+	}
+
+	if healthy == false && !fwd.forwarders[index].onHold {
+		fwd.forwarders[index].onHold = true
+		fwd.chooseBest()
+	}
+}
+
 func (fwd *overlaySwitchForwarder) stopFrom(index int) {
 	for index < len(fwd.forwarders) {
 		subFwd := &fwd.forwarders[index]
@@ -337,7 +365,7 @@ func (fwd *overlaySwitchForwarder) chooseBest() {
 
 	for i := range fwd.forwarders {
 		subFwd := &fwd.forwarders[i]
-		if subFwd.fwd == nil {
+		if subFwd.fwd == nil || subFwd.onHold {
 			continue
 		}
 
@@ -415,6 +443,10 @@ func (fwd *overlaySwitchForwarder) ErrorChannel() <-chan error {
 	return fwd.errorChan
 }
 
+func (fwd *overlaySwitchForwarder) HealthChannel() <-chan bool {
+	return fwd.healthChan
+}
+
 func (fwd *overlaySwitchForwarder) Stop() {
 	fwd.lock.Lock()
 	defer fwd.lock.Unlock()
diff --git a/router/sleeve.go b/router/sleeve.go
index 2c6e507d..4cda180c 100644
--- a/router/sleeve.go
+++ b/router/sleeve.go
@@ -326,6 +326,7 @@ type sleeveForwarder struct {
 	// listener channels
 	establishedChan chan struct{}
 	errorChan       chan error
+	healthChan      chan bool
 
 	// Explicitly locked state
 	lock       sync.RWMutex
@@ -445,6 +446,10 @@ func (fwd *sleeveForwarder) ErrorChannel() <-chan error {
 	return fwd.errorChan
 }
 
+func (fwd *sleeveForwarder) HealthChannel() <-chan bool {
+	return fwd.healthChan
+}
+
 type curriedForward struct {
 	NonDiscardingFlowOp
 	fwd *sleeveForwarder