Socket: Detect connection close

[Vladimir Panteleev]

On Monday, 24 March 2014 at 18:06:56 UTC, nrgyzer wrote:
> Alright, that world work. But what about sending 10 Bytes, 
> waiting some minutes (probably some hours) and sending 10 Bytes 
> again. Is it possible to do this on the same Socket?

Yes. Unless your socket is marked as non-blocking, the read 
operation will block until more data is available (or the 
connection gets closed).

> My scenario is: I've one central server and multiple clients. 
> The clients are connecting to the server and send (or request) 
> some information. After that the clients are simply "sleeping" 
> for an unknown time. For instance client 1 for 1 seconds, 
> client 2 for 1 minute, client 3 for 1 hours and client 4 for 1 
> day. After that time they resend/request some more information 
> from the server... my questions regarding this are:
>
> - When I client connects, the server creates a thread. How can 
> I determine that I can terminate the thread because the 
> connection is inactive (for instance because the client closed 
> the connection or some connecting issues)?

As mentioned previously, you know that the connection was closed 
cleanly on the remote side when receive returns an empty buffer.

> - How can the client check if the connection is alive? (If the 
> connection isn't alive, I need to reconnect)

You generally have to implement this yourself. Silence on the 
wire can mean two things:

1. No data is being sent;
2. Something is preventing data from being sent (lost packets, a 
peer was shut down or restarted, Internet routing changes, 
squirrel chewed through a wire).

Generally network software implements "ping" packets that are 
sent occasionally (once every few seconds or minutes). If no 
"ping" packets are received throughout a certain interval, the 
connection is considered dead.

TCP also has a "keepalive" option which does this for you, but it 
is an optional TCP feature and thus may or may not work.