[racket] WebSocket server dropping connections

From: Jakub Piotr Cłapa (jpc-ml at zenburn.net)
Date: Wed Feb 23 21:15:47 EST 2011

On 09.02.11 21:42, Jay McCarthy wrote:
> 2011/2/8 Jakub Piotr Cłapa<jpc-ml at zenburn.net>:
>> On 07.02.11 16:19, Jay McCarthy wrote:
>>> I don't know much about Chrome 10, but is it possible that it has
>>> dropped support for the old framing mode? The WebSocket implementation
>>> supports both old and new, but defaults to old because that's what a
>>> lot of browsers (used to) default to.
>>> Look at framing-mode in the documentation.
>> The connections work fine most of the time so I do not think this is a
>> problem with handling of the protocol itself. Sorry, I should have been more
>> explicit about this in my e-mail.
>>> It is unlikely that they are being garbage collected, because it uses
>>> the same "connection manager" machinery as the Web server, that
>>> retains references to the ports.
>> Does this machinery kill its threads without warning?
> Yes, they have timeouts.
>> Like shutting down the
>> custodian after the tcp sockets get closed by the other end?
> Yes.

I guess I will just catch the exceptions and clean everything up in the 
handler. OTOH WebSockets should be able to be really long-lived. I guess 
will look into it.

>> After inspecting the source code for ws-closed? it also seams to me there is
>> no way to find out if a websocket was closed but without proper handshake.
>> Maybe ws-closed? should also check whether the underlaying sockets are open
>> or the ws-send and ws-receive calls could change the ws-conn state on any
>> tcp error?
>> The problem can get magnified by the fact that AFAICT ws-close! will not be
>> able to set ws-conn-closed? to #f if there are errors in the shutdown
>> sequence (i.e. if write-byte throws an exception).
> That sounds right... do you want to do patches/test cases?

I finally found time to try to do it but it looks like that although 
there is a difference in the external text representation 
(display/write), there is no tcp-closed? predicate. So no luck on this one.

Jakub Piotr Cłapa

Posted on the users mailing list.