David,

[replying to the list, for others' edutainment]

There are 3 things the backend webserver can instruct Perlbal to do in
its HTTP response headers, reproxy-wise:

X-Reproxy-URL:  go serve a URL from elsewhere.  (actually a list of URLs,
and Perlbal will find one that's alive)

X-Reproxy-File:  go serve a file from the Perlbal machine's filesystem

X-Reproxy-Service:  replay the request to a different named SERVICE in
Perlbal's config file


That last one was designed for TypePad.com's migration (still in progress)
... a request to TypePad.com first goes to Perlbal, then to a "redirector"
pool of HTTP servers, which looks at the request and decides, "Has this
user's account and/or this feature of the service been migrated to the new
architecture?  If so, X-Reproxy-Service: NEW_STACK, else
X-Reproxy-Service: OLD_STACK"  Or, the decider HTTP servers can just
return the new-stack contents or old-stack contents themselves.  In this
example, {NEW,OLD}_STACK are a Perlbal SERVICEs with their own
options/pool/etc.

The reason TypePad couldn't just use X-Reproxy-URL, is because we didn't
want to pick a SPECIFIC URL (specific host) to internally redirect the
user... we still wanted Perlbal's smart load balancing to do that.  So
instead we just tell Perlbal to switch pools, and user gets back in line
for the fastest connection to another pool.  (which again may bounce them
around elsewhere with X-Reproxy-Whatever...)

Hope that clarifies.

- Brad


On Wed, 2 May 2007, David Weekly wrote:

> Brad,
>
> Hi! Nate and I are both fans of the reproxying mechanism you baked
> into perlbal and have made a very rough alpha patch for pound to
> perform the same mechanism. It surprised me, however, to note the lack
> of clear documentation I was able to find on the definition of
> X-Reproxy-Service in particular.
>
> I think Nathan had read it as a way for a reverse proxy to announce to
> its downstream users its ability to reproxy, but in reading over the
> patches for perlbal / mogile, this doesn't seem to be the case. Could
> you point at us some better documentation for what, exactly, the
> header X-Reproxy-Service means?
>
> Cheers,
> -David
>
>

[ Post a follow-up to this message ]