Re: digest only peering?

On Nov 20, 6:11am, Bruce Campbell (possibly) wrote:
> On Fri, 20 Nov 1998, Alex Rousskov wrote:
>
> > On Fri, 20 Nov 1998, Andrew Smith wrote:
> >
> > > So long as you are aware that the siblings wind up downloading
> > > your false digest hits on them, and then giving the objects to
> > > you. The owners of the siblings may not like that, or it may
> > > result in IP based access control problems at remote sites. We
> > > had to go back to ICP for that reason. (This is on Squid2.0
> > > patch 2 anyway).
> >
> > How can IP based access control on remote sites distinguish false
> > hit from a true hit or ICP hit from a digest hit?
>
> They can't, unless you want to start checking false hits in the last
> digest (more cpu work) and supply the object if you did indeed advertise
> it in a previous cache-digest.

This isn't at all difficult to do (see the patch below), and if cache
digest hits did take that much time to calculate they wouldn't be
worth very much...

The patch (which I'd appreciate if somebody with siblings with
miss_access and cache-digests would check out - we don't yet have the
resources here for it...) is:

diff -Naurd squid-2.1.PATCH1.old/src/forward.c squid-2.1.PATCH1/src/forward.c
--- squid-2.1.PATCH1.old/src/forward.c Wed Nov 18 12:09:46 1998
+++ squid-2.1.PATCH1/src/forward.c Fri Nov 20 21:25:48 1998
@@ -335,6 +335,11 @@
         ch.src_addr = peer_addr;
         ch.request = r;
         answer = aclCheckFast(Config.accessList.miss, &ch);
+#if USE_CACHE_DIGESTS
+ if (answer == 0) {
+ answer = cacheDigestTest(store_digest, r->canonical);
+ }
+#endif
         if (answer == 0) {
             err = errorCon(ERR_FORWARDING_DENIED, HTTP_FORBIDDEN);
             err->request = requestLink(r);

> Personally, patching squid to check the X-Squid-Error field and retry the
> request at another sibling/parent/direct seems appropriate. (on a failed
> request)

Sure, better error handling is a definite need... but decreasing the
number of errors will also help, and is easier to do (in this case).

        -Allen

-- 
Allen Smith				easmith@beatrice.rutgers.edu