[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Master/slave repositories: RE: [rfc] redirects with neon

From: Sander Striker <striker_at_apache.org>
Date: 2002-03-29 20:39:15 CET

> To: dev@subversion.tigris.org
> Subject: Re: [rfc] redirects with neon
>
> On Thu, Mar 28, 2002 at 08:50:43PM -0500, Mark Benedetto King wrote:
> > On Thu, Mar 28, 2002 at 07:50:25PM -0500, Ben Collins wrote:
> > > On Thu, Mar 28, 2002 at 03:44:48PM -0800, Greg Stein wrote:
> >...
> > > > The tricky part will be telling the app "<that> is now located <here>" and
> > > > having it do something smart. Especially if <that> is a whole directory and
> > > > everything needs to get shifted recursively.
> > >
> > > One other place that Sander pointed out where this would be useful is to
> > > have master read/write repo's, with read-only slaves. Redirect all the
> > > write operations to the master server. Thing is, in this case it is a
> > > "standard" operation (much like what I am doing with the http and https
> > > redirect). Any ideas on how to handle this specially?
>
> If they were temporary redirects, then we wouldn't be pushing URLs back to
> the application.
>
> > I don't think write operations should be redirected.
> >
> > RFC 2616:
> >
> > 10.3.2
> >
> > [...]
> >
> > If the 301 status code is received in response to a request other
> > than GET or HEAD, the user agent MUST NOT automatically redirect the
> > request unless it can be confirmed by the user, since this might
> > change the conditions under which the request was issued.
>
> Yah... good point on the redirect on writes. We probably don't want to allow
> those.

Ah. Hmmm. The reason I pointed Ben Collins to mod_rewrite was to be
able to get users redirected to https on commits, while reads would just
go over http. This apparently isn't allowed :(
 
> > Of course, neon already deviates (intentionally) from this behaviour, but
> > it does it only on "read-only" request types ("HEAD","GET", "PROPFIND",
> > and "OPTIONS").
>
> RFC 2616 was too specific when it mentioned GET/HEAD. The *intent* (and
> confirmed by Roy Fielding) is to allow automatic redirect for any idempotent
> request method.
>
> > If you implement the read-only slaves as write-through caches, you
> > shouldn't need to redirect the writes.
>
> Yup. But recognize: if a read-only slave is simply a caching proxy, then the
> writes will write-through anyways.

Yes, but I wasn't getting at that. The slaves I had in mind aren't just
simple caching proxies (although I agree one could use those to implement
slaves).

I had a chat some time ago with Bill Tutt. Maybe he can summarize what
we said (or even put it in a nice document *prod* *prod*). Bill?

> Cheers,
> -g

Sander

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri Mar 29 20:35:35 2002

This is an archived mail posted to the Subversion Dev mailing list.