> On Dec 16, 2020, at 2:52 AM, Graham Leggett <minf...@sharp.fm> wrote: > > On 12 Dec 2020, at 01:59, Roy T. Fielding <field...@gbiv.com > <mailto:field...@gbiv.com>> wrote: > >> That is too many questions. The purpose of the cache requirement is so that >> the cache >> does not deliver a non-validated entry after receiving a failed validation. >> It doesn't really >> matter what code is returned so long as it is 5xx, so the specs are >> over-constraining here. >> >> The CoAdvisor test suite is overly pedantic. >> >> And, as stated, this only applies when the request contains max-revalidate >> and the >> action being done is a cache revalidation. No other code should behave this >> way. > > To clarify, the change under discussion covers behaviour when a proxy of any > kind (including an ftp proxy) suffers a low level network error of any kind, > from DNS errors to low level network errors.
Yes, and as I said, that is wrong. > Whether ultimately correct or not, CoAdvisor was very specific as to the > codes to be returned in these cases. I am very happy for their financial success. However, they are frequently wrong, even though they are supposed to be based on the standard. >> Reverting the change is the correct call, regardless, but it is also the >> right choice. >> I have filed a bug on the Cache spec to change that MUST send 504 to a MUST >> send 5xx. >> >> https://github.com/httpwg/http-core/issues/608 >> <https://github.com/httpwg/http-core/issues/608> >> >> If you think we need to change other things in the specs, please file bugs >> now. > > The above issue relates to must-revalidate only, while the wider issue is for > which 5xx error to be returned in which situation. Can you clarify? > > The patch in this thread covers each case, and basically boils down to a > choice between “Bad Gateway” or “Timed Out”. The patch changed BAD_GATEWAY to Timed Out for no reason whatsoever. The spec text that you pointed to is about cache and cache control when the cache is disconnected from the network. The spec says that the cache must generate a 504 when it is disconnected, meaning it has NO response other than the entry in its cache. I fully understand why CoAdvisor is confused and unable to properly test for that condition, given that it isn't an internal cache testing suite, but this has nothing to do with a gateway receiving or generating a 5xx code upon gateway fail. And, in any case, the requirement in the spec is overly constraining when it is supposed to be minimally preventing the cache from reusing its cached response. IOW, gateway != cache. > Given the long list here http://coad.measurement-factory.com/clients.html > <http://coad.measurement-factory.com/clients.html> I would be keen to make > sure httpd worked the same way as all those other servers. That is not an interop concern. ....Roy
