Re: p1: generating "internal" errors

[Willy Tarreau <[e-mail]>]

On Sat, Apr 20, 2013 at 05:06:08PM +1000, Mark Nottingham wrote:
> 
> On 20/04/2013, at 5:03 PM, Willy Tarreau <[e-mail]> wrote:
> 
> > On Sat, Apr 20, 2013 at 02:07:52PM +1000, Mark Nottingham wrote:
> >> p1 3.2.4 requires that a syntax violation in a received response be turned
> >> into a 502 (Bad Gateway) status code.
> >> 
> >> I'm not necessarily against it, but I think if we're going to take this
> >> approach to errors in received responses, it should be systematic, and we
> >> should recommend that others do it too. Currently, a lot of people are
> >> inventing new pseudo status codes to fill this role.
> >> 
> >> What do people think?
> > 
> > haproxy does exactly this right now (502) and I was not aware that people
> > invent their own code, this is pretty bad :-(
> 
> I'm thinking more about client libraries than intermediaries.

OK. As was once discussed here, if we insist on no status code in the range
100-599 to be randomly picked by a developer, we're leaving enough room for
libraries to do what they want without risk of interference.

> >> This might not result in any changes in our specs beyond adjusting language
> >> in a few other places to do the same thing. I could see writing a separate
> >> spec for a header that described the type of error, though.
> > 
> > Good idea. Alternatively the reason code after the 502 could be modulated too.
> 
> That is discarded in some circumstances, and in any case we shouldn't
> encourage people to start using it for semantically significant things...

Agreed.

Willy