Re: Squid-2.5, what is left

On Fri, 2001-10-19 at 18:35, Henrik Nordstrom wrote:
> Adrian Chadd wrote:
>
> > Other than that, uhm.. florin has sent me an initial patch which
> > I've got working against squid-2.5 that makes the internal
> > resolver work with search/domain paths in resolv.conf .
> > Let me go and create a branch and stuff it in there.
>
> I'd say this also is 2.6 material.. domain search lists is not something
> we have had, and should not be taken lighly in proxies. If something is
> done then it should behave similar to how most browsers behave without a
> proxy.. (i.e. "smart" lookups to expand the host name according to a
> list of templates, and send a browser redirect when a match is found).
> Simply adding a "search list" capability to the DNS client is not
> sufficient for proper operation, and only covers a small subset of the
> "DNS" search issues people are having..

Well, the DNS patch is working reliably under high load here, with 2.4

The reason to create it in the first place was because people start
using the search paths as something which is "going without saying" -
they just put that path into resolv.conf and expect every software to
work with it; sometimes a search path is a company standard or "legacy
setting" (which is the case here, at SGI). All daemons do that (heck,
all _clients_ do that), but not Squid.
Using the external dnsservers to achieve the search path resolution is
ugly to say the least, so let's ignore them for the moment. I agree the
redirectors are a clever tool, but isn't nicer to have this capability
_included_ in Squid? Otherwise, we come to the same issue as with
dnsservers: use external processes to do a simple and common feature
that's otherwise achievable internally (if you use my patch).

I'll say, yeah, let's tell people to use redirectors for
_clever_and_nonstandard_ things. That makes sense. But simple
resolv.conf search paths are quite taken for granted by almost anyone,
and the average Joe Sysadmin will expect the "right" thing from every
daemon (i.e.: follow the search path); so, when he's got some list of
domains there, let's make Squid use them. And if they want anything
above that (non-standard, non-common stuff), go ahead and use
redirectors, because it's not a common thing.
I assume here that most of the people will want either no search path at
all, or the classic search path as most daemons do. Anything more
complex than that is surely doable by redirectors, i agree, but i'm not
sure how many people will require that thing. And the common features
it's nice to be included in Squid (as opposed to non-common ones, which
are redirectors' job).

Just my couple of cents...

-- 
Florin Andrei
"Making sure the kernel is highly stable even under extreme load (and
longer uptimes) takes time." - bero@redhat.com