I was trying to accept some data from client, send it to the server and also send feedback from server to the client again. When server only gets some data, everething works fine. While trying to implement two-way communication, server just hungs.
Server's code:
var tcpListener = new TcpListener(IPAddress.Any, 8888);
try
{
tcpListener.Start();
Console.WriteLine("Server is currently run, waiting for incoming connections");
while (true)
{
using var tcpClient = await tcpListener.AcceptTcpClientAsync();
await using var stream = tcpClient.GetStream();
var incomeData = await DeserializeHelper.ConverteToBytes(stream);
var incomePerson = MessagePackSerializer.Deserialize<Person>(incomeData);
Console.WriteLine($"Income message from {incomePerson.Name}: '{incomePerson.Message}'");
var responseMessage = $"Thank you, {incomePerson.Name} for the message";
var responceData = MessagePackSerializer.Serialize(responseMessage);
await stream.WriteAsync(responceData);
}
}
finally
{
tcpListener.Stop();
}
Client's code:
using TcpClient tcpClient = new TcpClient();
await tcpClient.ConnectAsync("127.1.199.250",8888);
Console.WriteLine("Hello, client!");
try
{
var stream = tcpClient.GetStream();
Console.WriteLine("Enter your name");
var name = Console.ReadLine();
Console.WriteLine("Enter message");
var message = Console.ReadLine();
Person tom = new Person(name, message);
var sentData = MessagePackSerializer.Serialize(tom);
await stream.WriteAsync(sentData);
Console.WriteLine("Message was sent to the server");
var incomeResponce = await DeserializeHelper.ConverteToBytes(stream);
var incomeResponceMessage = MessagePackSerializer.Deserialize<string>(incomeResponce);
Console.WriteLine($"Server's message: {incomeResponce}");
}
catch (Exception e)
{
Console.WriteLine(e.Message);
}
Also all of them using my own static method ConverteToBytes, here it's code:
public static async Task<byte[]> ConverteToBytes(NetworkStream stream)
{
await using var ms = new MemoryStream();
byte[] bytesData = null;
int count = 0;
do
{
byte[] buffer = new byte[1024];
count = await stream.ReadAsync(buffer,0,1024);
ms.Write(buffer,0,count);
} while (stream.CanRead && count > 0);
return bytesData = ms.ToArray();
}
I guess the problem is with not correct stream using, but don't know what exactly is wrong. Thank's everyone!
I strongly recommend not using TCP/IP in the first place. If possible, use a well-established protocol with good library support, e.g., HTTP or WebSockets. Writing a correct TCP/IP application is extremely difficult.
If you must use TCP/IP, then I recommend watching my video series on the subject and reading my TCP/IP .NET Sockets FAQ.
Of particular note is that network streams are streams of bytes, not streams of messages. So there's no built-in way to know when a message completes. So you'll have to add that to your protocol yourself, i.e., using message framing. I describe message framing on my blog, and the video series linked above shows a more modern way to do it.
Without message framing, your server is likely just waiting for more bytes to arrive from the client connection.
Related
I am creating a TCP Server/Client to communicate with each other using multithread and I'm having trouble getting the Server to correctly read strings that my Client is sending. For example, I send the string "testing" from the Client to the Server. On the server console, it instead reads "Received System.Net.Sockets.NetworkStream". I've been stuck on it for a couple hours and I am at a loss at how to approach this situation. Any help with be greatly appreciated.
TLDR: Server and client can communicate with each other but server can't correctly read string sent from client.
Server Code:
async Task EchoAsync(TcpClient client, CancellationToken ct)
{
var buf = new byte[4096];
var stream = client.GetStream();
while (!ct.IsCancellationRequested)
{
var amountRead = await stream.ReadAsync(buf, 0, buf.Length, ct);
WriteToConsole("Received " + stream.ToString(), 0);
if (amountRead == 0)
{
break; // End of stream
}
await stream.WriteAsync(buf, 0, amountRead, ct);
}
}
Client Code:
private async void ClientToServerMsg_btn_Click(object sender, RoutedEventArgs e)
{
using (var NetworkStream = clientSocket.GetStream())
using (var writer = new StreamWriter(NetworkStream))
{
//writer.AutoFlush = true;
WriteToConsole("Writing to server...", 1);
await writer.WriteLineAsync(ClientToServerMsg.Text);
}
}
Try
WriteToConsole("Received " + System.Text.Encoding.ASCII.GetString(buf), 0);
Instead
WriteToConsole("Received " + stream.ToString(), 0);
Currently I try to read and write Async to/from a network stream. My software is the Client part and the server can send informations on its own or respond to commands I send him.
So I need a socket which
reads all the time (in case the server sends status informations)
stops reading when I want to send commands (commands can be sequences of data with multible Write and Read operations)
So I thought it would be a good approach to create a Semaphore and a Background Task which handles the server sent messages and in case I want to send a command I block the semaphore and have full access to read/write operations to the socket.
Here is what I do currently.
private TcpClient _tcpClient = new TcpClient();
protected SemaphoreSlim ClientSemaphore { get; } = new SemaphoreSlim(1, 1);
public async Task ConnectAsync()
{
if (_tcpClient.Connected)
{
await DisconnectAsync();
}
await _tcpClient.ConnectAsync(Hostname, RemotePort);
//here the background Task is started
_ = AutoReceiveMessages();
}
private async Task AutoReceiveMessages()
{
while (_tcpClient.Connected)
{
//enter and lock semaphore
await ClientSemaphore.WaitAsync();
try
{
//read from socket until timeout (ms)
var msg = await ReadFromSocket(2000);
foreach (var cmd in SplitMessageInTelegrams(msg))
{
Console.WriteLine("MESSAGE --> " + cmd);
}
}
catch (Exception ex)
{
}
finally
{
//release semaphore
ClientSemaphore.Release();
}
}
}
private async Task<string> ReadFromSocket(double timeout = 0)
{
var buf = new byte[4096];
var stream = _tcpClient.GetStream();
//read from stream or timeout
var amountReadTask = stream.ReadAsync(buf, 0, buf.Length);
var timeoutTask = Task.Delay(TimeSpan.FromMilliseconds(timeout));
await Task.WhenAny(timeoutTask, amountReadTask)
.ConfigureAwait(false);
//timeout
if (!amountReadTask.IsCompleted)
{
throw new TimeoutException("Timeout");
}
//no timeout
return Encoding.ASCII.GetString(buf, 0, amountReadTask.Result);
}
But this do not work as I expected...
I use this methode to send a message to the server and in WireShark I see the server resonds with the same message
protected async Task SendTelegramAsync(ITelegram telegram)
{
await ClientSemaphore.WaitAsync();
try
{
_ = telegram ?? throw new ArgumentException($"{nameof(telegram)}");
if (!_tcpClient.Connected) throw new InvalidOperationException("Socket not connected!");
var buf = new byte[4096];
var stream = _tcpClient.GetStream();
var msg = Encoding.ASCII.GetBytes("\x02" + telegram.GetCommandMessage() + "\x03");
Console.WriteLine("WRITE --> " + msg);
await stream.WriteAsync(msg, 0, msg.Length);
//comment AutoReceiveMessage and remove comment from this
//and I get responses from the server
//var test = await ReadFromSocket(2000);
}
finally
{
ClientSemaphore.Release();
}
}
I know in this case I do not need the semaphore but later I want to create sequences so one command consists of multible writes and reads and as long as the command is executed I do not want to use the AutoReceiveMessages method.
The problem now is
If I use it like this I never get a response the ReadFromSocket method always get the timeout even when wireshark tell me the server has responded
But even better if I disable AutoReceiveMessages (just comment _ = AutoReceiveMessages()) and use ReadFromSocket directly in SendTelegramAsync() everything work as expected.
So I think the problem is something related to the background task and the ReadAsync but I couldnt figure it out...
Got It!
stream.DataAvailable is your friend (or my friend :)).
If I check before the ReadAsync if DataIsAvailable then I have no problem anymore.
if (_tcpClient.GetStream().DataAvailable)
var msg = await ReadFromSocket(DEFAULT_TIMEOUT);
Im building a simple TCP client and server as a basis for my networking project. Im planning to use the async await technique for future proof and scaleable server.
If I put wrong ip address, the client cant connect to my server and throw an exception. I can catch the exception using try/catch but is that the recommended way to do?
What do you guys think of the implementation. Any comments for me to improve?
My server
private void startServer_Click(object sender, RoutedEventArgs e)
{
if (anyIP.IsChecked == true)
{
listener = new TcpListener(IPAddress.Any, Int32.Parse(serverPort.Text));
Logger.Info("Ip Address : " + IPAddress.Any + " Port : " + serverPort.Text);
}
else
{
listener = new TcpListener(IPAddress.Parse(serverIP.Text), Int32.Parse(serverPort.Text));
Logger.Info("Ip Address : " + serverIP.Text + " Port : " + serverPort.Text);
}
try
{
listener.Start();
Logger.Info("Listening");
HandleConnectionAsync(listener, cts.Token);
}
//finally
//{
//cts.Cancel();
//listener.Stop();
//Logger.Info("Stop listening");
//}
//cts.Cancel();
}
async Task HandleConnectionAsync(TcpListener listener, CancellationToken ct)
{
while (!ct.IsCancellationRequested)
{
Logger.Info("Accepting client");
//TcpClient client = await listener.AcceptTcpClientAsync();
TcpClient client = await listener.AcceptTcpClientAsync();
Logger.Info("Client accepted");
EchoAsync(client, ct);
}
}
async Task EchoAsync(TcpClient client, CancellationToken ct)
{
var buf = new byte[4096];
var stream = client.GetStream();
while (!ct.IsCancellationRequested)
{
var amountRead = await stream.ReadAsync(buf, 0, buf.Length, ct);
Logger.Info("Receive " + stream.ToString());
if (amountRead == 0) break; //end of stream.
await stream.WriteAsync(buf, 0, amountRead, ct);
Logger.Info("Echo to client");
}
}
private void stopServer_Click(object sender, RoutedEventArgs e)
{
cts.Cancel();
listener.Stop();
Logger.Info("Stop listening");
}
My client
private void connect_Click(object sender, System.Windows.RoutedEventArgs e)
{
IPAddress ipAddress;
int port;
//TODO Check if ip address is valid
ipAddress = IPAddress.Parse(serverIP.Text);
//TODO port range is 0-65000
port = int.Parse(serverPort.Text);
StartClient(ipAddress, port);
}
private static async void StartClient(IPAddress serverIpAddress, int port)
{
var client = new TcpClient();
//can i try/catch to catch await exception?
try
{
await client.ConnectAsync(serverIpAddress, port);
}
catch (Exception e)
{
Logger.Info(e);
}
Logger.Info("Connected to server");
using (var networkStream = client.GetStream())
using (var writer = new StreamWriter(networkStream))
using (var reader = new StreamReader(networkStream))
{
writer.AutoFlush = true;
for (int i = 0; i < 10; i++)
{
Logger.Info("Writing to server");
await writer.WriteLineAsync(DateTime.Now.ToLongDateString());
Logger.Info("Reading from server");
var dataFromServer = await reader.ReadLineAsync();
if (!string.IsNullOrEmpty(dataFromServer))
{
Logger.Info(dataFromServer);
}
}
}
if (client != null)
{
client.Close();
Logger.Info("Connection closed");
}
}
I have a .NET TCP/IP FAQ that I recommend to get some of the basics down.
After just a brief look at your code, these points stood out to me:
Both your client and server have times when they're only reading (not writing). This means you're subject to the half-open scenario (as I describe in my FAQ). A robust server should be writing periodically even if it has nothing to say.
Both your client and server have times when they're only writing (not reading). This means that you're subject to a deadlock (as I describe in my FAQ) if the other end is not behaving well (e.g., sending lots of data). However, you can't just read indefinitely or you'll open yourself up to a DoS; so you should decide where your limit is and establish read buffer sizes (and write timeouts) that make sense for your application.
Using ReadLineAsync leaves you open to a trivial DoS attack, since you can't specify the maximum allowed size of the line.
Your code must be prepared for an exception at any time (as I describe in my FAQ). Obviously, ReadAsync and WriteAsync may throw. What's less obvious is that any socket method may throw, including AcceptTcpClientAsync.
Your code uses a mixture of exception handling types. The async Task methods are never awaited, so exceptions there just silently end that method. The StartClient method is more problematic, since it is async void. You'll need to think through your application needs for error detection and retry strategies, and apply proper handling at every level.
In conclusion, I reiterate my comment: I strongly recommend just self-hosting SignalR. Sockets should only be used if you have no choice.
I have simple server that gets string from client and prints it on screen.
I also have simple client, sending data and closing:
static void Main()
{
var client = new TcpClient("localhost", 26140);
var stream = client.GetStream();
Byte[] data = System.Text.Encoding.UTF8.GetBytes("CALC qwer");
stream.Write(data, 0, data.Length);
stream.Close();
client.Close();
//Thread.Sleep(100);
}
And with uncommented string 'Thread.Sleep(100)' it works ok.
But when commenting, sometimes ( 1 of 5-10 runs ) client doesn't send the string.
Watching wireshark and netstat I've noticed that client sends SYN,ACK package, establishes connection and exits without sending anything and without closing the socket.
Could anyone explain this behaivor? Why sleep helps? What am I doing wrong?
UPD:
With this sample code adding flush() before closing really works, thanks Fox32.
But after it I returned to my initial code:
var client = new TcpClient("localhost", 26140);
client.NoDelay = true;
var stream = client.GetStream();
var writer = new StreamWriter(stream);
writer.WriteLine("CALC qwer");
writer.Flush();
stream.Flush();
stream.Close();
client.Close();
And it isn't working, even with NoDelay. It's bad - using StreamWriter over network stream?
UPD:
Here is server code:
static void Main(string[] args)
{
(new Server(26140)).Run();
}
In Server class:
public void Run()
{
var listener = new TcpListener(IPAddress.Any, port);
listener.Start();
while (true)
{
try
{
var client = listener.AcceptTcpClient();
Console.WriteLine("Client accepted: " + client.Client.RemoteEndPoint);
var stream = client.GetStream();
stream.ReadTimeout = 2000;
byte[] buffer = new byte[1000];
stream.Read(buffer, 0, 1000);
var s = Encoding.UTF8.GetString(buffer);
Console.WriteLine(s);
}
catch (Exception ex)
{
Console.WriteLine("ERROR! " + ex.Message);
}
}
}
UPD:
Adding even Sleep(1) makes crashes happen in 1 of 30-50 clients running at the same time.
And adding Sleep(10) seems to be solving it totally, I can't catch any crash.
Don't understand, why socket needs this several milliseconds to close correctly.
The TcpClient is using the Nagle's algorithm and waits for more data before sending it over the wire. If you close the socket to fast, no data is trasmitted.
You have multiple ways to solve this problem:
The NetworkStream has a Flush method for flushing the stream content (I'm not sure if this method does anything from the comment on MSDN)
Disable Nagle's algorithm: Set NoDelay of the TcpCLient to true.
The last option is to set the LingerState of the TcpClient. The Close method documentation states, that the LingerState is used while calling Close
In almost all cases you are supposed to call Shutdown on a Socket or TcpClient before disposing it. Disposing rudely kills the connection.
Your code basically contains a race condition with the TCP stack.
Setting NoDelay is also a fix for this but hurts performance. Calling Flush IMHO still results an an disorderly shutdown. Don't do it because they are just hacks which paint over the problem by hiding the symptoms. Call Shutdown.
I want to stress that Shutdown being called on the Socket is the only valid solution that I know of. Even Flush just forces the data onto the network. It can still be lost due to a network hickup. It will not be retransmitted after Close has been called because Close is a rude kill on the socket.
Unfortunately TcpClient has a bug which forces you to go to the underlying Socket to shut it down:
tcpClient.Client.Shutdown();
tcpClient.Close();
According to Reflector, if you have ever accessed GetStream this problem arises and Close does not close the underlying socket. In my estimation this bug was produced because the developer did not really know about the importance of Shutdown. Few people know and many apps are buggy because of it. A related question.
In your server side code you are only calling Read() once, but you can't assume the data will be available when you call read. You have to continue reading in a loop until no more data is available. See the full example below.
I have tried to reproduce your issue with the minimal amount of code and was not able to. The server prints out the clients message everytime. No special settings such as NoDelay and no explicit Close() or Flush(), just Using statements which ensures all resources are properly disposed.
class Program
{
static int port = 123;
static string ip = "1.1.1.1";
static AutoResetEvent waitHandle = new AutoResetEvent(false);
static void Main(string[] args)
{
StartServer();
waitHandle.WaitOne();
for (int x=0; x<1000; x++)
{
StartClient(x);
}
Console.WriteLine("Done starting clients");
Console.ReadLine();
}
static void StartClient(int count)
{
Task.Factory.StartNew((paramCount) =>
{
int myCount = (int)paramCount;
using (TcpClient client = new TcpClient(ip, port))
{
using (NetworkStream networkStream = client.GetStream())
{
using (StreamWriter writer = new StreamWriter(networkStream))
{
writer.WriteLine("hello, tcp world #" + myCount);
}
}
}
}, count);
}
static void StartServer()
{
Task.Factory.StartNew(() =>
{
try
{
TcpListener listener = new TcpListener(port);
listener.Start();
Console.WriteLine("Listening...");
waitHandle.Set();
while (true)
{
TcpClient theClient = listener.AcceptTcpClient();
Task.Factory.StartNew((paramClient) => {
TcpClient client = (TcpClient)paramClient;
byte[] buffer = new byte[32768];
MemoryStream memory = new MemoryStream();
using (NetworkStream networkStream = client.GetStream())
{
do
{
int read = networkStream.Read(buffer, 0, buffer.Length);
memory.Write(buffer, 0, read);
}
while (networkStream.DataAvailable);
}
string text = Encoding.UTF8.GetString(memory.ToArray());
Console.WriteLine("from client: " + text);
}, theClient);
}
}
catch (Exception e)
{
Console.WriteLine(e);
}
}, TaskCreationOptions.LongRunning);
}
}
UPD:
I've tested this bug on several computers and nothing crashed. Seems like it is a local bug on my computer.
ENDOFUPD
So, what I've found about reproducing this bug.
#Despertar - your code works well. But it isn't reproduce conditions of this bug. On client you need to send data and quit after it. And in your code many clients are sending data and after all application is closing.
This is how I'm testing this on my computer:
I have server ( just accepting connection and print incoming data ), client ( just sends data once end exits ) and running utility ( runs client exe several times ).
So, I starts server, copies running utility to the clients folder and runs it.
Running ulility starts 150 clients connecting to server and 5-10 of them dies ( I see error in the server console ). And uncommenting Thread.Sleep() on client works well, no errors.
Can anyone try to reproduce this version of code?
Client code:
private static void Main(string[] args)
{
try
{
using (TcpClient client = new TcpClient(ip, port))
{
using (NetworkStream networkStream = client.GetStream())
{
using (StreamWriter writer = new StreamWriter(networkStream))
{
writer.WriteLine("# hello, tcp world #");
writer.Flush();
}
networkStream.Flush();
networkStream.Close();
}
client.Close();
//Thread.Sleep(10);
}
}
catch (Exception ex)
{
Console.WriteLine(ex.Message);
}
}
Code, running client several times ( compile it in exe file and put near client's exe - this code will run many clients one by one ):
static void Main(string[] args)
{
string path = "YOU_CLIENT_PROJECT_NAME.exe";
for (int i = 0; i < 150; i++ )
{
Console.WriteLine(i);
Process.Start(path);
Thread.Sleep(50);
}
Console.WriteLine("Done");
Console.ReadLine();
}
( don't forget to change path to corrent exe filename )
Server code:
class Program
{
static int port = 26140;
static AutoResetEvent waitHandle = new AutoResetEvent(false);
static void Main(string[] args)
{
StartServer();
waitHandle.WaitOne();
Console.ReadLine();
}
static void StartServer()
{
Task.Factory.StartNew(() =>
{
try
{
TcpListener listener = new TcpListener(port);
listener.Start();
Console.WriteLine("Listening...");
waitHandle.Set();
while (true)
{
TcpClient theClient = listener.AcceptTcpClient();
Task.Factory.StartNew(paramClient =>
{
try
{
TcpClient client = (TcpClient) paramClient;
byte[] buffer = new byte[32768];
MemoryStream memory = new MemoryStream();
using (NetworkStream networkStream = client.GetStream())
{
networkStream.ReadTimeout = 2000;
do
{
int read = networkStream.Read(buffer, 0, buffer.Length);
memory.Write(buffer, 0, read);
} while (networkStream.DataAvailable);
string text = Encoding.UTF8.GetString(memory.ToArray());
}
}
catch (Exception e)
{
Console.WriteLine("ERROR: " + e.Message);
}
}, theClient);
}
}
catch (Exception e)
{
Console.WriteLine(e);
}
}, TaskCreationOptions.LongRunning);
}
}
I've tried code, reproducing this bug on several computers. No one crashes. Seems like it's my local computer bug.
Thanks for everybody for trying to help me.
Anyway, it's so strange. If I'll found out why this bug exists on my computer, I'll write about it.
I have a client application that serializes a object and sends it to a server application. The server should deserialize the object, make changes to it, then serialize it and send it back.
Server Code:
TcpClient client = server.AcceptTcpClient();
using(NetworkStream stream = client.GetStream())
{
using(StreamReader streamReader = new StreamReader(stream))
{
string xmlData = streamReader.ReadToEnd();
}
}
The ReadToEnd doesn't return unless the client closes the stream. But if the client closes the stream, I can't send a response.
Is there a better way to do this?
You can signal "end of data" by closing only your half of the duplex TCP connection. This is accomplished with Socket.Disconnect.
See how it works with this example, which I kept similar to yours. The client sends the data and then calls Disconnect; this allows ReadToEnd to return while still keeping the server's half of the connection open. The server then sends a response and also disconnects, after which both parties can Close their end of the connection to tear it down.
static void Main(string[] args)
{
Action clientCode = () =>
{
var buffer = new byte[100];
var clientSocket = new Socket(AddressFamily.InterNetwork,
SocketType.Stream, ProtocolType.Tcp);
clientSocket.Connect(IPAddress.Loopback, 6690);
clientSocket.Send(buffer);
clientSocket.Disconnect(false);
Console.WriteLine("Client: message sent and socket disconnected.");
while (true) {
var bytesRead = clientSocket.Receive(buffer);
if (bytesRead == 0) {
break;
}
Console.WriteLine("Client: read " + bytesRead + " bytes.");
}
clientSocket.Dispose();
};
var server = new TcpListener(IPAddress.Loopback, 6690);
var thread = new Thread(new ThreadStart(clientCode));
server.Start();
thread.Start();
var client = server.AcceptTcpClient();
using(NetworkStream stream = client.GetStream()) {
using(StreamReader streamReader = new StreamReader(stream))
{
var data = streamReader.ReadToEnd();
Console.WriteLine("Server: read " + data.Length + " bytes.");
// Since we 're here we know that the client has disconnected.
// Send the response before StreamReader is disposed, because
// that will cause the socket itself to be closed as well!
Thread.Sleep(TimeSpan.FromSeconds(1));
Console.WriteLine("Server: sending response.");
stream.Write(new byte[10], 0, 10);
Console.WriteLine("Server: closing socket.");
}
}
server.Stop();
Console.WriteLine("Server: waiting for client thread to complete.");
thread.Join();
return;
}
You could use a higher level framework like WCF, or if you are hell-bent on managing your own streams, then don't use ReadToEnd()- use ReadLine() (and have the client send messages as lines), or use Read() and have a special character (a sentinel) represent the end of a message.