Relay Server

Suppose you have two programs that can talk to a relay computer via TCP, but can not reach each other (because of NAT or firewall issues). What do you do? You come to this wonderful repo and use our wonderful RelayServer!

Our relayServer accepts requests from servers that want all their traffic sent through it. To use the relay server:

./relayServer <port>
./serverPorgram <relayHost> <relayPort>

This will show a message like the following:

2017/10/14 11:23:17 Serving relay requests on port 62514

At this point, all your clients can connect to the relayPort on port 62514.

Please see section [connecting your server](Connecting your server) for instructions on how to connect servers to relayServer.

Architecture

The following diagram shows what happens when a new relayRequest is made:

The dotted lines represent TCP connections. The solid ones, actions. Clouds represent tcp servers.

1. A server sends the relayServer a request to have its traffic relayed. This connection is kept open to notify the server when a new client has connected.
2. The relayServer receives the request and creates 2 tcp Servers: one to receive connections from clients and one to receive connections from servers. The relay server advertises the port that clients of the server can connect to.
3. A client connects to the relayServer. The relayServer uses the connection established in 1 to notify the server that a new connection has been made.
4. The server connects to the relayServer.
5. As there is now a connection with both the client and the server, the relayServer synchronizes I/O between them, making sure every byte that is written on any one end is also passed to the other.

As mentioned above, when a client connects to the relayServer, we notify the server by sending a string. This string has the prefix [NEW] followed by the the port and is ended with the \n character. servers using relayServer read this message from the connection stream established when they first connected to the relayServer.

Connecting your server

As we don't want our relayServer to initiate any connections, but to receive them, we start sockets and wait for all interested parties to connect to us instead of trying to connect to them.

In the client/server architecture, clients connect to servers, hence servers have incoming connections from the clients. Accepting a client that wants to connect using the relayServer requires servers themselves to initiate the connection.

In order for servers to keep their logic unmodified, we suggest the following procedure when receiving clients from the relayServer:

Start the server as you would normally (it could be a local interface). Connect to the relayServer. When notified of a new client connection, create 2 tcp sockets: One to the relayServer on the host/port specified in the message and one to the server. Then synchronize these 2 connections.

Note: When closing the connection to the relay server, make sure you only close it for reading (i.e, send EOF without closing the entire connection). This allows all data to be flushed properly and not linger in the connection.

We have come up with 2 examples of a simple echo server and a simple web server that use this method:

echoServer.go

webServer.go