Re: I-D Action:draft-nottingham-http-stale-if-error-01.txt

On 11/05/2008, at 7:41 AM, Brian Smith wrote:

> Mark Nottingham wrote:
>> Yeah. Anybody else want to weigh in pro or con on this?
>
> "In this context, an error is any situation which would result in a
> 500, 502, 503 or 504 HTTP response status code being returned."
>
> It seems wrong to limit this to these four status codes. Why not  
> have the server indicate that stale responses are okay when it  
> returns the error response? In other words, instead of returning  
> stale-if-error=XXX on a successful response, it would return it in  
> the 5xx responses.

Because the origin server isn't always the one that generates a 5xx;  
it may be an intermediary along the way (including one that the cache  
is implemented within).

> Alternatively, allow the stale-if-error subfield to list the status  
> codes for which a stale response could be returned.

That's way too complex, and serving a stale representation for a 4xx  
error doesn't really make any sense.

> Also, I think that the specification needs to explicitly mention  
> that a warning should be added when a cache returns a stale response  
> due to a back-end error.

I can add an e.g. (like in swr).

Thanks,



--
Mark Nottingham     http://www.mnot.net/

Received on Sunday, 11 May 2008 01:06:51 UTC