[client] Add tri-state connection status to guard for smarter ICE retry (#5828)

* [client] Add tri-state connection status to guard for smarter ICE retry

Refactor isConnectedOnAllWay to return a ConnStatus enum (Connected,
Disconnected, PartiallyConnected) instead of a boolean. When relay is
up but ICE is not (PartiallyConnected), limit ICE offers to 3 retries
with exponential backoff then fall back to hourly attempts, reducing
unnecessary signaling traffic. Fully disconnected peers continue to
retry aggressively. External events (relay/ICE disconnect, signal/relay
reconnect) reset retry state to give ICE a fresh chance.

client/internal/peer/guard/guard.go

@@ -8,7 +8,19 @@ import (
 	log "github.com/sirupsen/logrus"
 )
 
-type isConnectedFunc func() bool
+// ConnStatus represents the connection state as seen by the guard.
+type ConnStatus int
+
+const (
+	// ConnStatusDisconnected means neither ICE nor Relay is connected.
+	ConnStatusDisconnected ConnStatus = iota
+	// ConnStatusPartiallyConnected means Relay is connected but ICE is not.
+	ConnStatusPartiallyConnected
+	// ConnStatusConnected means all required connections are established.
+	ConnStatusConnected
+)
+
+type connStatusFunc func() ConnStatus
 
 // Guard is responsible for the reconnection logic.
 // It will trigger to send an offer to the peer then has connection issues.
@@ -20,14 +32,14 @@ type isConnectedFunc func() bool
 // - ICE candidate changes
 type Guard struct {
 	log                     *log.Entry
-	isConnectedOnAllWay     isConnectedFunc
+	isConnectedOnAllWay     connStatusFunc
 	timeout                 time.Duration
 	srWatcher               *SRWatcher
 	relayedConnDisconnected chan struct{}
 	iCEConnDisconnected     chan struct{}
 }
 
-func NewGuard(log *log.Entry, isConnectedFn isConnectedFunc, timeout time.Duration, srWatcher *SRWatcher) *Guard {
+func NewGuard(log *log.Entry, isConnectedFn connStatusFunc, timeout time.Duration, srWatcher *SRWatcher) *Guard {
 	return &Guard{
 		log:                     log,
 		isConnectedOnAllWay:     isConnectedFn,
@@ -57,8 +69,17 @@ func (g *Guard) SetICEConnDisconnected() {
 	}
 }
 
-// reconnectLoopWithRetry periodically check the connection status.
-// Try to send offer while the P2P is not established or while the Relay is not connected if is it supported
+// reconnectLoopWithRetry periodically checks the connection status and sends offers to re-establish connectivity.
+//
+// Behavior depends on the connection state reported by isConnectedOnAllWay:
+//   - Connected: no action, the peer is fully reachable.
+//   - Disconnected (neither ICE nor Relay): retries aggressively with exponential backoff (800ms doubling
+//     up to timeout), never gives up. This ensures rapid recovery when the peer has no connectivity at all.
+//   - PartiallyConnected (Relay up, ICE not): retries up to 3 times with exponential backoff, then switches
+//     to one attempt per hour. This limits signaling traffic when relay already provides connectivity.
+//
+// External events (relay/ICE disconnect, signal/relay reconnect, candidate changes) reset the retry
+// counter and backoff ticker, giving ICE a fresh chance after network conditions change.
 func (g *Guard) reconnectLoopWithRetry(ctx context.Context, callback func()) {
 	srReconnectedChan := g.srWatcher.NewListener()
 	defer g.srWatcher.RemoveListener(srReconnectedChan)
@@ -68,36 +89,46 @@ func (g *Guard) reconnectLoopWithRetry(ctx context.Context, callback func()) {
 
 	tickerChannel := ticker.C
 
+	iceState := &iceRetryState{log: g.log}
+	defer iceState.reset()
+
 	for {
 		select {
-		case t := <-tickerChannel:
-			if t.IsZero() {
-				g.log.Infof("retry timed out, stop periodic offer sending")
-				// after backoff timeout the ticker.C will be closed. We need to a dummy channel to avoid loop
-				tickerChannel = make(<-chan time.Time)
-				continue
+		case <-tickerChannel:
+			switch g.isConnectedOnAllWay() {
+			case ConnStatusConnected:
+				// all good, nothing to do
+			case ConnStatusDisconnected:
+				callback()
+			case ConnStatusPartiallyConnected:
+				if iceState.attempt() {
+					callback()
+				} else {
+					ticker.Stop()
+					tickerChannel = iceState.hourlyC()
+				}
 			}
 
-			if !g.isConnectedOnAllWay() {
-				callback()
-			}
 		case <-g.relayedConnDisconnected:
 			g.log.Debugf("Relay connection changed, reset reconnection ticker")
 			ticker.Stop()
-			ticker = g.prepareExponentTicker(ctx)
+			ticker = g.newReconnectTicker(ctx)
 			tickerChannel = ticker.C
+			iceState.reset()
 
 		case <-g.iCEConnDisconnected:
 			g.log.Debugf("ICE connection changed, reset reconnection ticker")
 			ticker.Stop()
-			ticker = g.prepareExponentTicker(ctx)
+			ticker = g.newReconnectTicker(ctx)
 			tickerChannel = ticker.C
+			iceState.reset()
 
 		case <-srReconnectedChan:
 			g.log.Debugf("has network changes, reset reconnection ticker")
 			ticker.Stop()
-			ticker = g.prepareExponentTicker(ctx)
+			ticker = g.newReconnectTicker(ctx)
 			tickerChannel = ticker.C
+			iceState.reset()
 
 		case <-ctx.Done():
 			g.log.Debugf("context is done, stop reconnect loop")
@@ -120,7 +151,7 @@ func (g *Guard) initialTicker(ctx context.Context) *backoff.Ticker {
 	return backoff.NewTicker(bo)
 }
 
-func (g *Guard) prepareExponentTicker(ctx context.Context) *backoff.Ticker {
+func (g *Guard) newReconnectTicker(ctx context.Context) *backoff.Ticker {
 	bo := backoff.WithContext(&backoff.ExponentialBackOff{
 		InitialInterval:     800 * time.Millisecond,
 		RandomizationFactor: 0.1,

client/internal/peer/guard/ice_retry_state.go

@@ -0,0 +1,57 @@
+package guard
+
+import (
+	"time"
+
+	log "github.com/sirupsen/logrus"
+)
+
+const (
+	// maxICERetries is the maximum number of ICE offer attempts when relay is connected
+	maxICERetries = 3
+	// iceRetryInterval is the periodic retry interval after ICE retries are exhausted
+	iceRetryInterval = 1 * time.Hour
+)
+
+// iceRetryState tracks the limited ICE retry attempts when relay is already connected.
+// After maxICERetries attempts it switches to a periodic hourly retry.
+type iceRetryState struct {
+	log     *log.Entry
+	retries int
+	hourly  *time.Ticker
+}
+
+func (s *iceRetryState) reset() {
+	s.retries = 0
+	if s.hourly != nil {
+		s.hourly.Stop()
+		s.hourly = nil
+	}
+}
+
+// attempt processes a single ICE retry tick. It returns true if the caller should send an offer.
+// When retries are exhausted it starts the hourly ticker and returns false once to signal the caller
+// to swap the tick channel. Subsequent calls (from the hourly ticker) return true.
+func (s *iceRetryState) attempt() bool {
+	if s.hourly != nil {
+		s.log.Debugf("hourly ICE retry attempt")
+		return true
+	}
+
+	s.retries++
+	if s.retries <= maxICERetries {
+		s.log.Debugf("ICE retry attempt %d/%d", s.retries, maxICERetries)
+		return true
+	}
+
+	s.log.Infof("ICE retries exhausted (%d/%d), switching to hourly retry", maxICERetries, maxICERetries)
+	s.hourly = time.NewTicker(iceRetryInterval)
+	return false
+}
+
+func (s *iceRetryState) hourlyC() <-chan time.Time {
+	if s.hourly == nil {
+		return nil
+	}
+	return s.hourly.C
+}