HDFS-10310. hdfsConnect hangs when given bad host or port. Contributed by James Clampffer.

hadoop-hdfs-project/hadoop-hdfs-native-client/src/main/native/libhdfspp/lib/common/util.h

@@ -82,6 +82,34 @@ bool lock_held(T & mutex) {
 // Returns a string containing error message on failure, otherwise an empty string.
 std::string SafeDisconnect(asio::ip::tcp::socket *sock);
 
+
+
+// The following helper function is used for classes that look like the following:
+//
+// template <typename socket_like_object>
+// class ObjectThatHoldsSocket {
+//   socket_like_object sock_;
+//   void DoSomethingWithAsioTcpSocket();
+// }
+//
+// The trick here is that ObjectThatHoldsSocket may be templated on a mock socket
+// in mock tests.  If you have a method that explicitly needs to call some asio
+// method unrelated to the mock test you need a way of making sure socket_like_object
+// is, in fact, an asio::ip::tcp::socket.  Otherwise the mocks need to implement
+// lots of no-op boilerplate.  This will return the value of the input param if
+// it's a asio socket, and nullptr if it's anything else.
+
+template <typename sock_t>
+inline asio::ip::tcp::socket *get_asio_socket_ptr(sock_t *s) {
+  (void)s;
+  return nullptr;
+}
+template<>
+inline asio::ip::tcp::socket *get_asio_socket_ptr<asio::ip::tcp::socket>
+                                            (asio::ip::tcp::socket *s) {
+  return s;
+}
+
 }
 
 #endif

hadoop-hdfs-project/hadoop-hdfs-native-client/src/main/native/libhdfspp/lib/connection/datanodeconnection.cc

@@ -56,8 +56,13 @@ void DataNodeConnectionImpl::Connect(
 }
 
 void DataNodeConnectionImpl::Cancel() {
-  mutex_guard state_lock(state_lock_);
+  std::string err;
-  std::string err = SafeDisconnect(conn_.get());
+
+  { // scope the lock for disconnect only, log has it's own lock
+    mutex_guard state_lock(state_lock_);
+    err = SafeDisconnect(conn_.get());
+  }
+
   if(!err.empty()) {
     LOG_WARN(kBlockReader, << "Error disconnecting socket in DataNodeConnectionImpl::Cancel, " << err);
   }

hadoop-hdfs-project/hadoop-hdfs-native-client/src/main/native/libhdfspp/lib/rpc/rpc_connection.h

@@ -151,7 +151,11 @@ void RpcConnectionImpl<NextLayer>::ConnectComplete(const ::asio::error_code &ec)
       HandshakeComplete(s);
     });
   } else {
-    next_layer_.close();
+    std::string err = SafeDisconnect(get_asio_socket_ptr(&next_layer_));
+    if(!err.empty()) {
+      LOG_INFO(kRPC, << "Rpc connection failed to connect to endpoint, error closing connection: " << err);
+    }
+
     if (!additional_endpoints_.empty()) {
       // If we have additional endpoints, keep trying until we either run out or
       //    hit one
@@ -355,8 +359,8 @@ void RpcConnectionImpl<NextLayer>::Disconnect() {
 
   request_over_the_wire_.reset();
   if (connected_ == kConnecting || connected_ == kConnected) {
-    next_layer_.cancel();
+    // Don't print out errors, we were expecting a disconnect here
-    next_layer_.close();
+    SafeDisconnect(get_asio_socket_ptr(&next_layer_));
   }
   connected_ = kDisconnected;
 }