Collapsed_forwarding -- current state?

From: Kale Stedman <kale@dont-contact.us>
Date: Thu, 26 Jan 2006 10:34:27 -0500

Hey all,

It was recommended to me that I post here looking for assistance
regarding this patch. I'm wondering what the current state of
collapsed_forwarding is in Squid 2.5; inasmuch as while we're running
2.5STABLE12, which has the code in place and configured per the
documentation in the source (the dev page lists an incorrect config
variable), we're not seeing it working as we perceive it should be.
For the same uncached object, we're seeing multiple concurrent
requests from our parent cache server to our back-end content
servers, whereas we believe that, with collapsed_forwarding, theses
requests should all be, well, collapsed into a single request, no?

Looking into our cache.log, every once in a while a
"clientProcessExpired: collapsed request STALE!" will pop up, but no
other related collapsed_forwarding messages, such as the desired
"HIT" response. I can easily generate a ton of requests for the same
uncached object (which has no cache-control or no-cache headers) and
kick them through concurrently -- each will generate a MISS and be
forwarded on to the backend server, as opposed to being served a HIT
when the first (collapsed) request returns and the object is cached.

Unfortunately I could find very little traces of information on this
feature on the mailing lists, save for one feller in Aug 2003 who
reported it not working as expected and receiving no reply that I
could find.

If I'm reading into the patch incorrectly, please let me know. I
want to reduce the load on the backend servers and this looks like a
key component in doing so. I'm not on the list; please cc: my
address in the response.

Thank you,
Kale Stedman
Systems Administrator, Zope Corporation
kale@zope.com
Received on Thu Jan 26 2006 - 15:18:21 MST

This archive was generated by hypermail pre-2.1.9 : Fri Jan 27 2006 - 12:00:02 MST