What to do about HPSS protocol backwards compatibility for faster push?

[Andrew Bennetts]

John Arbash Meinel wrote:
[...]
> Andrew Bennetts wrote:
[...]
> 
> 6) Have large requests use a streaming protocol (chunked transfer encoding) so
> they can fail early rather than after streaming all 10 MB across the wire.
> Obviously this requires getting chunked/streaming working.

This isn't as easy as it sounds.  We currently assume that a request won't get a
response until it is fully sent, and changing this will make working over HTTP
more difficult.  We'd could map it onto the “Expect: 100-continue”/“100
Continue” feature of HTTP/1.1, but there are pitfalls there (like the RFC says
that clients SHOULD NOT wait for an indefinite period for a 100 Continue/417
Expectation failed before sending the request body because older servers and
proxies might not transmit them.

Really, this is extremely close to just sending an explicit “is request X
supported?” request just before the actual request.  In terms of round trips on
the wire it's identical.

-Andrew.