From 5dde733d0c1d620a29197b58b312eaf5ace7ef3e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?P=C3=A9ter=20Szil=C3=A1gyi?= <peterke@gmail.com>
Date: Wed, 29 Apr 2015 18:42:43 +0300
Subject: [PATCH] p2p: add a dialer cooldown to prevent reattempting failures

---
 p2p/peer_error.go |  2 ++
 p2p/server.go     | 26 +++++++++++++++++++++++---
 2 files changed, 25 insertions(+), 3 deletions(-)

diff --git a/p2p/peer_error.go b/p2p/peer_error.go
index a912f60644..d9c09a9c6c 100644
--- a/p2p/peer_error.go
+++ b/p2p/peer_error.go
@@ -73,6 +73,7 @@ const (
 	DiscSelf
 	DiscReadTimeout
 	DiscSubprotocolError
+	DiscFailureCooldown
 )
 
 var discReasonToString = [...]string{
@@ -89,6 +90,7 @@ var discReasonToString = [...]string{
 	DiscSelf:                "Connected to self",
 	DiscReadTimeout:         "Read timeout",
 	DiscSubprotocolError:    "Subprotocol error",
+	DiscFailureCooldown:     "Failure cooldown",
 }
 
 func (d DiscReason) String() string {
diff --git a/p2p/server.go b/p2p/server.go
index 794c361256..1f15ca2011 100644
--- a/p2p/server.go
+++ b/p2p/server.go
@@ -21,6 +21,7 @@ const (
 	defaultDialTimeout       = 10 * time.Second
 	refreshPeersInterval     = 30 * time.Second
 	trustedPeerCheckInterval = 15 * time.Second
+	dialFailureCooldown      = 30 * time.Second
 
 	// This is the maximum number of inbound connection
 	// that are allowed to linger between 'accepted' and
@@ -100,9 +101,11 @@ type Server struct {
 
 	ourHandshake *protoHandshake
 
-	lock    sync.RWMutex // protects running and peers
+	lock sync.RWMutex // protects running and peers
+
 	running bool
-	peers   map[discover.NodeID]*Peer
+	peers   map[discover.NodeID]*Peer // Currently active peer pool
+	fails   map[discover.NodeID]bool  // List of recently failed connections
 
 	trusts    map[discover.NodeID]*discover.Node // Map of currently trusted remote nodes
 	trustDial chan *discover.Node                // Dial request channel reserved for the trusted nodes
@@ -201,6 +204,7 @@ func (srv *Server) Start() (err error) {
 	}
 	srv.quit = make(chan struct{})
 	srv.peers = make(map[discover.NodeID]*Peer)
+	srv.fails = make(map[discover.NodeID]bool)
 
 	// Create the current trust map, and the associated dialing channel
 	srv.trusts = make(map[discover.NodeID]*discover.Node)
@@ -456,7 +460,20 @@ func (srv *Server) dialNode(dest *discover.Node) {
 		// dialNode, so we need to count it down again. startPeer also
 		// does that when an error occurs.
 		srv.peerWG.Done()
-		glog.V(logger.Detail).Infof("dial error: %v", err)
+
+		// Mark the dialing as failed so prevent too fast redials, and set a
+		// timer for re-allowing dialing
+		srv.lock.Lock()
+		srv.fails[dest.ID] = true
+		srv.lock.Unlock()
+		go func() {
+			glog.V(logger.Detail).Infof("Dialing %v failed: %v. Locked until %v\n", dest, err, time.Now().Add(dialFailureCooldown))
+			<-time.After(dialFailureCooldown)
+
+			srv.lock.Lock()
+			delete(srv.fails, dest.ID)
+			srv.lock.Unlock()
+		}()
 		return
 	}
 	srv.startPeer(conn, dest)
@@ -546,6 +563,9 @@ func (srv *Server) checkPeer(id discover.NodeID) (bool, DiscReason) {
 	case id == srv.ntab.Self().ID:
 		return false, DiscSelf
 
+	case srv.fails[id]:
+		return false, DiscFailureCooldown
+
 	default:
 		return true, 0
 	}