- From: Nic Ferrier <nferrier@tapsellferrier.co.uk>
- Date: Sat, 02 Sep 2000 19:02:19 +0100
- To: www-talk@w3.org
Another reason why the UA should pay more attention to the "close" value of the Connection header than the "keep-alive" value is that if the client keeps the persitent connection open there will be a delay whilst the server waits for timeout before bringing the connection down. Even if the server *does* mean to keep the connection alive it can always resend responses. But if the UA keeps the connection open the user is left waiting for the timeout period. That's not very user friendly... it was also what prompted Jim to ask the question in the first place (so in his situation it would obviously have been the "correct" thing to do). One more reason for closing the connection in Jims case is that "Connection: keep-alive" is not a defined part of the HTTP/1.1 spec. The token keep-alive has no meaning. See section 14.10 of RFC2616. And the last reason I can think of is that the "Connection: keep-alive" header:token value is the system used by some HTTP/1.0 servers which presumes that the connection will be closed unless otherwise specified. It could be that what is being talked to by Jim is an HTTP/1.0 webserver connected to an HTTP/1.1 servlet engine (or possibly vice versa). Imagine it: the servlet engine is behaving with what it understands of it's protocol - it sets the response to HTTP/1.1 and puts in the "Connection: close" header because it no longer needs the connection. But then the slightly stupid webserver takes over and adds it's own concept of the "Connection" header - thus confusion. Remember that 1.0 relied upon the client to close the connection if it wished, essentially saying this: If there is a UA at the end of this connection who understands this persistence protocol and wishes the connection to remain open it will hold it open, otherwise it'll get closed down Therefore, thinking pragmatically, HTTP/1.1 UAs (and proxies on the server side) should always prioritize a "close" higher than anything else. Nic
Received on Saturday, 2 September 2000 13:57:52 UTC