What happens to select() when socket is set to handle timeout

Question

According to the manual page for recv(), errno is set to EAGAIN or EWOULDBLOCK if a timeout has been set for receive using setsockopt(SO_RCVTIMEO).

My question is what happens if multiple such sockets are used with select(). Would select return if one of the socket times out due to inactivity. What would be returned by select().

I am trying to implement a tftp server with feature to detect timeouts. One way could be to use a timeout with select() but then I would have to use a different value of timeout for each socket and keep updating the timer to the minimum value, and then do some more juggling.... etc.. etc... Just feels like a lot of unnecessary work.

PS: The tftp server is a concurrent server with multiple clients being handled using I/O Multiplexing.

And implement the timeout using alarms?? what other way is there? — Parth K, Apr 11 '19 at 18:30
But every socket in select has a different value for timeout — Parth K, Apr 11 '19 at 18:32
And according to the first comment on [this answer](https://stackoverflow.com/a/2878982/7906473), using alarms is not a good idea. — Parth K, Apr 11 '19 at 18:35
No, using alarms is a bad idea. That's why I suggest select timeouts. You need to buffer the data for every socket so you might as well have a *structure* for every one of them. — Antti Haapala -- Слава Україні, Apr 11 '19 at 18:36
But then I have to update the timeval structure for each client every time select returns and I would have to keep checking the list to check the minimum value that would be passed to `select()` — Parth K, Apr 11 '19 at 18:38
Keep the structs, or pointers to them, in a queue. Struct with the nearest timeout will always be at the head, so timeout the head.. Add new timeouts at the tail end. — Martin James, Apr 11 '19 at 18:51

score 1 · Accepted Answer · answered Apr 11 '19 at 19:53

The timeout parameter of select() determines the maximum time that the select() call itself will wait for something to happen before the call returns, not how long individual sockets will wait before returning a timeout error.

It sounds like you are wanting to declare some kind of an error condition if you don't hear from a client for some period of time. With UDP, you will have to keep track of that yourself. For each client, keep a record of the last time you heard from it. Put select() in a loop with a timeout of something like 1 second, then every time it returns check the difference between the current time and the last time you heard from each client. When that difference exceeds whatever threshold you want, you have your error condition.

That is why I am thinking of setting the timeout parameter to each socket instead of `select()`. My understanding is if I set individual sockets to time out, and then call `select()`, if the sockets do time out, I will simply return from select with the socket fd, and attempt a read on the socket. I imagine the read would fail with EAGAIN and that's my error condition. — Parth K, Apr 11 '19 at 19:59
UDP sockets do not timeout. TCP sockets will only timeout if you enable keepalive. — Seth Noble, Apr 12 '19 at 20:15

What happens to select() when socket is set to handle timeout

1 Answers1