[plt-scheme] Limiting Concurrent Connections to Web Server

From: Jay McCarthy (jay.mccarthy at gmail.com)
Date: Fri Jan 16 08:47:09 EST 2009

On Thu, Jan 15, 2009 at 11:23 PM, Henk Boom <lunarc.lists at gmail.com> wrote:
> I tried it in one of the nightly builds, and it seems to maintain the
> concurrency limit properly. I've noticed, though, that threads
> belonging to cancelled connections are no longer dropped, though, so
> if I hit refresh 10 times in firefox with a concurrency limit of 1, it
> actually does the full calculations for all of the responses one at a
> time before I get the final response.
>
> I can't tell why this happens though, from looking at the above code.
> Maybe it's because channel-put/get share the property of 'sleep' that
> prevents the thread from being terminated? (What is that reason, by
> the way?)

The 'lock' will be released when it is yield or when the thread dies.
The thread will die in a few different ways: completion (but this will
cause a yield), exceptions (but this will cause a yield, from
dynamic-wind), calling `exit' (no yield), or being shutdown by its
custodian (no yield).

It is unlikely that your code is calling `exit', so the most common
way for the thread to die unnaturally is when its custodian is
shutdown. There are a few conditions when the custodian for a
connection (and thus its handler) is shut-down:

1. Timeout
2. Connection is HTTP/1.0 and first request is received
3. Error outputting the response
4. Error reading the request
5. The request handler is the servlet dispatcher and the servlet calls exit

If the client closes the connection, then there is probably going to
be an error in #3 or #4, because the port will be written/read and it
will fail. You'll have to see what your browser is actually doing
though, because it could be starting a new connection without killing
the old (leading to #1); it could be using an HTTP/1.1 connection and
sending a new request along in sequence with the first request, so
they are all serviced; etc.

You could do a wireshark and see what actually happens to the TCP
connections. (Or just email a dump to me.)

Jay


-- 
Jay McCarthy <jay at cs.byu.edu>
Assistant Professor / Brigham Young University
http://teammccarthy.org/jay

"The glory of God is Intelligence" - D&C 93


Posted on the users mailing list.