fix(p2p): give each underlay address fresh 15s connection timeout

pkg/p2p/libp2p/libp2p.go

@@ -997,6 +997,7 @@ func (s *Service) Connect(ctx context.Context, addrs []ma.Multiaddr) (address *b
 	// addresses for the same peer is passed to the host.Host.Connect function
 	// and reachabiltiy Private is emitted on libp2p EventBus(), which results
 	// in weaker connectivity and failures in some integration tests.
+
 	for _, addr := range addrs {
 		// Extract the peer ID from the multiaddr.
 		ai, err := libp2ppeer.AddrInfoFromP2pAddr(addr)
@@ -1029,7 +1030,11 @@ func (s *Service) Connect(ctx context.Context, addrs []ma.Multiaddr) (address *b
 			return address, p2p.ErrAlreadyConnected
 		}
 
-		if err := s.connectionBreaker.Execute(func() error { return s.host.Connect(ctx, *info) }); err != nil {
+		connectCtx, cancel := context.WithTimeout(ctx, 15*time.Second)
+		err = s.connectionBreaker.Execute(func() error { return s.host.Connect(connectCtx, *info) })
+		cancel()
+
+		if err != nil {
 			if errors.Is(err, breaker.ErrClosed) {
 				s.metrics.ConnectBreakerCount.Inc()
 				return nil, p2p.NewConnectionBackoffError(err, s.connectionBreaker.ClosedUntil())

pkg/topology/kademlia/kademlia.go

@@ -42,10 +42,9 @@ const (
 
 	addPeerBatchSize = 500
 
-	// To avoid context.Timeout errors during network failure, the value of
-	// the peerConnectionAttemptTimeout constant must be equal to or greater
-	// than 5 seconds (empirically verified).
-	peerConnectionAttemptTimeout = 15 * time.Second // timeout for establishing a new connection with peer.
+	// Each underlay address gets up to 15s for connection (in libp2p.Connect).
+	// This budget allows multiple addresses to be tried sequentially per peer.
+	peerConnectionAttemptTimeout = 45 * time.Second // timeout for establishing a new connection with peer.
 )
 
 // Default option values
@@ -807,9 +806,6 @@ func (k *Kad) connectBootNodes(ctx context.Context) {
 	var attempts, connected int
 	totalAttempts := maxBootNodeAttempts * len(k.opt.Bootnodes)
 
-	ctx, cancel := context.WithTimeout(ctx, 15*time.Second)
-	defer cancel()
-
 	for _, addr := range k.opt.Bootnodes {
 		if attempts >= totalAttempts || connected >= 3 {
 			return
@@ -820,6 +816,10 @@ func (k *Kad) connectBootNodes(ctx context.Context) {
 			if attempts >= maxBootNodeAttempts {
 				return true, nil
 			}
+
+			ctx, cancel := context.WithTimeout(ctx, peerConnectionAttemptTimeout)
+			defer cancel()
+
 			bzzAddress, err := k.p2p.Connect(ctx, []ma.Multiaddr{addr})
 
 			attempts++