On Tue, Apr 19, 2011 at 8:03 PM, Yang Zhang <yanghatespam@xxxxxxxxx> wrote: > On Mon, Apr 18, 2011 at 9:18 PM, Yang Zhang <yanghatespam@xxxxxxxxx> wrote: >> On Mon, Apr 18, 2011 at 9:08 PM, Amos Jeffries <squid3@xxxxxxxxxxxxx> wrote: >>> On Mon, 18 Apr 2011 20:07:33 -0700, Yang Zhang wrote: >>>> >>>> We're using squid (3.0.STABLE19-1ubuntu0.1) as an application cache >>>> and we're trying to cache everything, for a long time, but max-age=0 >>>> is throwing off squid. >>> >>> max-age=0 is called "reload" in HTTP terminology. >>> >>>> >>>> I have this refresh_pattern in squid.conf: >>>> >>>> $ grep ^refresh_pattern /etc/squid3/squid.conf >>>> refresh_pattern ^ftp: 1440 20% 10080 >>>> refresh_pattern ^gopher: 1440 0% 1440 >>>> refresh_pattern . 525600 100% 525600 override-expire ignore-private >>>> ignore-no-cache ignore-no-store >>>> >>> >>> To quote Frankenstein >>> "Why? whyyyy?" >>> >>> >>>> I have the following response prepared to be sent from a dummy web server: >>>> >>>> $ cat response >>>> HTTP/1.1 200 OK >>>> Server: Apache-Coyote/1.1 >>>> Expires: 0 >>>> Pragma: no-cache >>>> Cache-Control: no-cache, must-revalidate, max-age=0 >>>> Content-Type: text/html;charset=UTF-8 >>>> Date: Mon, 18 Apr 2011 10:06:26 GMT >>>> Connection: close >>> >>> Those headers say two things: >>> * that cached content MUST NOT be supplied to the client. EVER. >>> * that any existing cached content also MUST be checked for updates on the >>> next request. >>> >>> Ignoring this requirement *from the server*. Again... why? >>> >>> >>> I can only imagine that you have fallen for the myth that "application >>> caches" must always provide HITs. This is false and following it will break >>> things for you in potentially nasty ways. >>> >>> A cachable application will not send max-age=0 or "Expires: 0", both of >>> which mean the object was obsolete the second it was produced. >>> >>> Squid operating as a reverse-proxy will cache the replies and keep updating >>> fresh copies as long as the server is available to present new ones. If the >>> server dies an 5xx start coming back, the stale versions will start to be >>> used. How long for and whether "must-revalidate" produces 4xx/5xx to the >>> client depends on particular squid versions. >>> I recommend you grab the newest release you can for the best HTTP/1.1 >>> caching behaviour. there are still things we are fixing but work is ongoing >>> so the newer the release the better things work. I maintain a PPA for Ubuntu >>> (https://launchpad.net/~yadi/+archive/ppa) with newer versions. >>> >>> <snip> >>>> >>>> This response prevents anything from being cached; all requests for >>>> http://localhost:8765 result in TCP_MISS. I believe it's because of >>>> the max-age=0 - once I remove that, this request is cached fine. Yet >>>> I had set override-expire in the refresh_pattern. >>>> (Note that I don't think it's a must-revalidate/ignore-revalidate (new >>>> in v3.1) problem - I can leave that in fine.) >>> >>> However "We're using squid (3.0.STABLE19-1ubuntu0.1)" ... not a v3.1 >>> release. >>> >>> "ignore-revalidate" blocks the *client* from requesting a revalidation >>> (max-age=0). But still obeys the server controls. >>> >>>> >>>> Any hints on how to make squid ignore the max-age=0? I've pored over >>>> the refresh_pattern documentation a bunch of times but still can't see >>>> what's wrong. Thanks in advance for any help. >>> >>> If you must; >>> * "ignore-reload" to ignore the max-age=0 from either client or server >>> IIRC. >> >> Just tried this out - it doesn't work. The comments in the squid.conf >> also say that it's for max-age=0 from the client (no mention of >> server). >> >>> * "reload-into-ims" to convert a client max-age=0 ("force a new copy or >>> die") into a soft IMS request ("new content or cached copy"). >> >> The client is not issuing any max-age=0. Tried adding this in anyway >> - didn't work. >> >>> >>> 'must-revalidate' will still have affect until you are using a 3.1 release >>> with the "ignore-must-revalidate". The result of revalidate should only be >>> 304 (TCP_IMS_*) or 200 (TCP_REFRESH_UPDATE), not a "TCP_MISS" since the >>> content being validated *is* in the cache. >> >> My point earlier is that the must-revalidate from the server doesn't >> seem be what's preventing the response from getting cached.... > > Also, FWIW, these -max-age=0 responses are making it *into* the cache > (verified by grepping /var/spool/squid3), for whatever reason (not > sure why Squid is deciding to keep them if only to later not use > them). > > -- > Yang Zhang > http://yz.mit.edu/ > Update: using Squid 3.1.12.1 and ignore-must-revalidate fixed this. Weird since the original problem only arose from the presence of max-age=0; the revalidate itself wasn't preventing caching. -- Yang Zhang http://yz.mit.edu/
