- From: Koen Holtman <koen@win.tue.nl>
- Date: Wed, 10 Apr 1996 21:32:05 +0200 (MET DST)
- To: masinter@parc.xerox.com (Larry Masinter)
- Cc: fielding@avron.ics.uci.edu, koen@win.tue.nl, mogul@pa.dec.com, http-caching@pa.dec.com
Larry Masinter:
>
>> must-revalidate is just a max-age=0 with a special warning attached.
>> If you want to trade must-revalidate for a Warning: header code with
>> the same semantics, I have no big problem.
>
>It seems that max-age=0 with a special warning attached is more
>general, if not more useful, than must-revalidate.
>So what's the "special warning"?
If we split the special warning off
Cache-Control: must-revalidate
we would get
Cache-Control: max-age=0
Warning: 42 Revalidation Essential
or, if we rename Jeff's warning header because it can carry
prescriptive codes:
Cache-Control: max-age=0
Cache-Exception: 42 Revalidation Essential
We would then get a definition like this:
42 Revalidation Essential
Indicates that the revalidation of stale responses is essential
for the correct operation of the service offered by the origin
server. Service authors can generate this exception code along
with a "Cache-Control: max-age=0" header to warn that a failure to
revalidate a request on the entity could result in significantly
incorrect operation, such as incorrect reporting on the nature of
a financial transaction about to be executed.
Proxy caches which fail to, or are unwilling to, revalidate a
response with this code MUST NOT return a stale response with a 13
(?) (revalidation failed) exception code, but MUST return a 504
(Gateway Timeout) error response. User agent caches failing to
revalidate a response with a 42 (Revalidation Essential) code MUST
also return a 504 (Gateway Timeout) error response.
User agent caches which are configured to return stale responses
because of severe connectivity constraints SHOULD return a 504
error response instead of a stale response with a 42 (Revalidation
Essential) code, but MAY also, if specifically configured to allow
this, return the stale response accompanied by a clear warning
that the service author cannot guarantee correct operation of the
service under these caching conditions.
> Is there some risk that the warning
>would get ignored when 'must-revalidate' would not?
I don't think so. I'd be happy with this being a special warning.
Koen.
Received on Wednesday, 10 April 1996 19:57:24 UTC