Rapid test the local endpoints as well

peers/manager.go

@@ -196,7 +196,7 @@ func (pm *PeerManager) AddPeer(siteConfig SiteConfig) error {
 
 	// Perform rapid initial holepunch test (outside of lock to avoid blocking)
 	// This quickly determines if holepunch is viable and triggers relay if not
-	go pm.performRapidInitialTest(siteConfig.SiteId, siteConfig.Endpoint)
+	go pm.performRapidInitialTest(siteConfig.SiteId, siteConfig.Endpoint, siteConfig.LocalEndpoints)
 
 	return nil
 }
@@ -868,15 +868,43 @@ endpoint=%s:%d`, util.FixKey(peer.PublicKey), formattedEndpoint, relayPort)
 }
 
 // performRapidInitialTest performs a rapid holepunch test for a newly added peer.
-// If the test fails, it immediately requests relay to minimize connection delay.
+// It races a test of the public endpoint against a test of the local candidate endpoints
-// This runs in a goroutine to avoid blocking AddPeer.
+// (if any) and waits for both to finish before acting, so we never request relay only to
-func (pm *PeerManager) performRapidInitialTest(siteId int, endpoint string) {
+// have it immediately superseded by a local connection (or vice versa). Local wins if it's
+// viable at all; otherwise relay is requested only if the public endpoint isn't viable.
+// This runs in a goroutine to avoid blocking AddPeer - the peer already starts out pointed
+// at the public endpoint (set synchronously in AddPeer), so this just settles the peer onto
+// its steady-state connection within ~1-2 seconds.
+func (pm *PeerManager) performRapidInitialTest(siteId int, endpoint string, localEndpoints []string) {
 	if pm.peerMonitor == nil {
 		return
 	}
 
-	// Perform rapid test - this takes ~1-2 seconds max
+	var wg sync.WaitGroup
-	holepunchViable := pm.peerMonitor.RapidTestPeer(siteId, endpoint)
+	var localWinner string
+	var holepunchViable bool
+
+	if len(localEndpoints) > 0 {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			localWinner = pm.peerMonitor.RapidTestLocalEndpoints(siteId, localEndpoints)
+		}()
+	}
+
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		holepunchViable = pm.peerMonitor.RapidTestPeer(siteId, endpoint)
+	}()
+
+	wg.Wait()
+
+	if localWinner != "" {
+		logger.Info("Rapid test: local connection viable for site %d, switching to %s", siteId, localWinner)
+		pm.LocalPeer(siteId, localWinner)
+		return
+	}
 
 	if !holepunchViable {
 		// Holepunch failed rapid test, request relay immediately

peers/monitor/monitor.go

@@ -370,6 +370,49 @@ func (pm *PeerMonitor) RapidTestPeer(siteID int, endpoint string) bool {
 	return false
 }
 
+// RapidTestLocalEndpoints performs a rapid connectivity test of local candidate endpoints
+// for a newly added peer, so local viability is known within the same ~1-2 second window as
+// RapidTestPeer's public-endpoint test (rather than waiting for the next backoff-loop tick,
+// which could be tens of seconds away). Candidates are tried in order (best-to-worst) and
+// the first reachable one wins. Returns the winning endpoint, or "" if none are reachable.
+func (pm *PeerMonitor) RapidTestLocalEndpoints(siteID int, endpoints []string) string {
+	if pm.holepunchTester == nil || len(endpoints) == 0 {
+		return ""
+	}
+
+	pm.mutex.Lock()
+	timeout := pm.rapidTestTimeout
+	pm.mutex.Unlock()
+
+	for _, endpoint := range endpoints {
+		result := pm.holepunchTester.TestEndpoint(endpoint, timeout)
+		if !result.Success {
+			continue
+		}
+
+		logger.Info("Rapid test: local endpoint %s for site %d SUCCEEDED (RTT: %v)", endpoint, siteID, result.RTT)
+
+		pm.mutex.Lock()
+		// Peer may have been removed while we were testing.
+		stillTracked := false
+		if _, tracked := pm.localEndpoints[siteID]; tracked {
+			stillTracked = true
+			pm.localActiveEndpoint[siteID] = endpoint
+			pm.localFailures[siteID] = 0
+		}
+		pm.mutex.Unlock()
+
+		if stillTracked {
+			pm.sendLocal(siteID, endpoint)
+		}
+
+		return endpoint
+	}
+
+	logger.Info("Rapid test: no local endpoint reachable for site %d", siteID)
+	return ""
+}
+
 // UpdatePeerEndpoint updates the monitor endpoint for a peer
 func (pm *PeerMonitor) UpdatePeerEndpoint(siteID int, monitorPeer string) {
 	pm.mutex.Lock()