Re: Connection fatal HTTP/2 stream errors

On 18 June 2014 02:32, Greg Wilkins <gregw@intalio.com> wrote:
> we are currently leaning towards treating all stream errors as fatal to the
> connection.

This fits with our current error handling philosophy.

The general sense here has been that failing hard is preferred at this
stage.  Postel's principle leads to the sorts of sloppiness that
HTTP/1.1 suffers from in excess, and the more we can do to expose
errors now, the less we will have to deal with corner cases later.

All of the errors you cite can, and probably should, be fatal to the
extent that is reasonable.  Obviously, you can't catch everything, and
things like bad paths might simply be forwarded on from HTTP/1.1,
where that sort of crap appears all the time.

On the other hand, if you are operating a server, you are free to be
more strict.  Don't abuse it, but you can enforce a fairly arbitrary
dress code.

Received on Wednesday, 18 June 2014 19:13:26 UTC