Re: Retry safety of HTTP requests

On Tue, Mar 22, 2016 at 1:55 AM, Subodh Iyengar <subodh@fb.com> wrote:

> Would the HTTP-wg be interested in a document describing retry-safety and
> its limitations?



It's not a passion of mine to work on, mostly because I think a lot of this
stuff will inherently be kicked to application space, but if you have an
idea for a draft that accurately describes what is going on and what needs
to go on I would certainly think that has potential as a valuable source
for interoperability. What are you thinking?

One example is pretty simple - POST on a persistent h1 connection. every
browser I'm aware of when POST-ing data to a previously used connection
will replay that POST if they get a RST or FIN before receiving any data
bytes. That's wrong from an idempotency pov of course, but dealing with the
reality of the race condition of the peer having closed the connection is
more important. Sometimes browsers will try not to do some forms of POST on
reused persistent connections for this reason, but that can't really be
generalized to higher transaction rate things like XHR.