phk at phk.freebsd.dk
Sat Jul 8 22:36:58 CEST 2006
In message <ujrodvzyhvt.fsf at cat.linpro.no>, Dag-Erling =?iso-8859-1?Q?Sm=F8rgra
>Yep. I did some more testing (fixed the backend.port bug!) and came
>across at least one important issue which we discussed earlier this
>week but didn't fix: hashing does not take the host into account, so
>you can't stick Varnish in front of a server with multiple name-based
>I believe that the way we want to fix this is by URL normalization:
>the URL we store in struct http should be a cleaned-up absolute URI.
I don't want to get into URI cleanup at all. My preference would
be to do the opposite: hash on $host+" "+$url_without_host because
trimming the host from absolute URIs is practically free whereas
normalizing them is text-processing.
If for some reason people need to do normalization, we can add
it as a VCL operation.
>(BTW, I don't like using the rr argument to http_Dissect() to
>distinguish between request and response; I'd rather have separate
>functions for each, with a common helper for the headers; but that's
A lot of the HTTP functions evolved and some amount of cleanup
is in order, but in this particular case I'm not sure if I agree.
Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG | TCP/IP since RFC 956
FreeBSD committer | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.
More information about the varnish-dev