etcd: ssh: Reconnect on SSH failures

If the SSH connection dies, the dialer can now reconnect that portion.

etcd/ssh/ssh.go

@@ -254,14 +254,23 @@ func (obj *World) Connect(ctx context.Context, init *engine.WorldInit) error {
 		return e
 	})
 
-	tunnels := make(map[string]net.Conn)
+	// This runs repeatedly when etcd tries to reconnect.
-	for _, seed := range obj.Seeds {
+	grpcWithContextDialerFunc := func(ctx context.Context, addr string) (net.Conn, error) {
-		addr := seedSSH[seed]
+		var reterr error
+		for _, seed := range obj.Seeds { // first successful connect wins
+			if addr != seedSSH[seed] {
+				continue // not what we're expecting
+			}
 			obj.init.Logf("tunnel: %s", addr)
+
 			tunnel, err := obj.sshClient.Dial("tcp", addr)
 			if err != nil {
-			return errwrap.Append(obj.cleanup(), err)
+				reterr = err
+				obj.init.Logf("ssh dial error: %v", err)
+				continue
 			}
+
+			// TODO: do we need a mutex around adding these?
 			obj.cleanups = append(obj.cleanups, func() error {
 				e := tunnel.Close()
 				if e == io.EOF { // XXX: why does this happen?
@@ -272,22 +281,20 @@ func (obj *World) Connect(ctx context.Context, init *engine.WorldInit) error {
 				}
 				return e
 			})
-		tunnels[addr] = tunnel
+
+			return tunnel, nil // connected successfully
+		}
+
+		if reterr != nil {
+			return nil, reterr
+		}
+		return nil, fmt.Errorf("no ssh tunnels available") // TODO: better error message?
 	}
 
 	etcdClient, err := clientv3.New(clientv3.Config{
 		Endpoints: obj.Seeds,
 		DialOptions: []grpc.DialOption{
-			grpc.WithContextDialer(func(ctx context.Context, addr string) (net.Conn, error) {
+			grpc.WithContextDialer(grpcWithContextDialerFunc),
-				tunnel, exists := tunnels[addr]
-				if !exists {
-					obj.init.Logf("can't find tunnel: %s", addr) // tell user early...
-					return nil, fmt.Errorf("can't find tunnel: %s", addr)
-				}
-				// TODO: print the scheme here on this log msg
-				obj.init.Logf("etcd dial: %s", addr)
-				return tunnel, nil
-			}),
 		},
 	})
 	if err != nil {