Java中可靠的TCP / IP

Question

Is there a way to have reliable communications (the sender get informed that the message it sent is already received by the receiver) using Java TCP/IP library in java.net.*? I understand that one of the advantages of TCP over UDP is its reliability. Yet, I couldn't get that assurance in the experiment below:

I created two classes:

1) echo server => always sending back the data it received.

2) client => periodically send "Hello world" message to the echo server.

They were run on different computers (and worked perfectly). During the middle of the execution, I disconnected the network (unplugged the LAN cable). After disconnected, the server still keep waiting for a data until a few seconds passed (it eventually raised an exception). Similarly, the client also keep sending a data until a few seconds passed (an exception is raised).

The problem is, objectOutputStream.writeObject(message) doesn't guarantee the delivery status of the message (I expect it to block the thread, keep resending the data until delivered). Or at least I get informed, which messages are missing.

Server Code:

import java.net.*;
import java.io.*;

import java.io.Serializable;

public class SimpleServer {
    public static void main(String args[]) {
        try {
            ServerSocket serverSocket = new ServerSocket(2002);
            Socket socket = new Socket();
            socket = serverSocket.accept();
            InputStream inputStream = socket.getInputStream();
            ObjectInputStream objectInputStream = new ObjectInputStream(
                    inputStream);

            while (true) {
                try {
                    String message = (String) objectInputStream.readObject();
                    System.out.println(message);
                    Thread.sleep(1000);
                } catch (Exception ex) {
                    ex.printStackTrace();
                }
            }
        } catch (Exception ex) {
            ex.printStackTrace();
        }
    }
}

Client code:

import java.net.*;
import java.io.*;

public class SimpleClient {
    public static void main(String args[]) {
        try {
            String serverIpAddress = "localhost"; //change this

            Socket socket = new Socket(serverIpAddress, 2002);
            OutputStream outputStream = socket.getOutputStream();
            ObjectOutputStream objectOutputStream = new ObjectOutputStream(
                    outputStream);

            while (true) {
                String message = "Hello world!";
                objectOutputStream.writeObject(message);

                System.out.println(message);
                Thread.sleep(1000);
            }
        } catch (Exception e) {
            e.printStackTrace();
        }
    }
}

Answer 1

如果您需要知道哪些消息已到达对等应用程序，则对等应用程序必须发送确认。

Answer 2

If you want this level of guarantees it sounds like you really want JMS. This can ensure not only that messages have been delivered but also have been processed correctly. ie there is no point having very reliable delivery if it can be discarded due to a bug.

You can monitor which messages are waiting and which consumers are falling behind. Watch a producer to see what messages it is sending, and have messages saved when it is down and are available when it restarts. ie reliable delivery even if the consumer is restarted.

Answer 3

TCP is always reliable. You don't need confirmations. However, to check that a client is up, you might also want to use a UDP stream with confirmations. Like a PING? PONG! system. Might also be TCP settings you can adjust.

Answer 4

Your base assumption (and understanding of TCP) here is wrong. If you unplug and then re-plug, the message most likely will not be lost.
It boils down on how long to you want the sender to wait. One hour, one day? If you'd make the timeout one day, you would unplug for two days and still say "does not work".

So the guaranteed delivery is that "either data is delivered - or you get informed". In the second case you need to solve it on application level.

Answer 5

You could consider using the SO_KEEPALIVE socket option which will cause the connection to be closed if no data is transmitted over the socket for 2 hours. However, obviously in many cases this doesn't offer the level of control typically needed by applications.

A second problem is that some TCP/IP stack implementations are poor and can leave your server with dangling open connections in the event of a network outage.

Therefore, I'd advise adding application level heartbeating between your client and server to ensure that both parties are still alive. This also offers the advantage of severing the connection if, for example a 3rd party client remains alive but becomes unresponsive and hence stops sending heartbeats.