On Tue, Apr 14, 2020 at 4:15 AM Stuart Douglas <sdouglas(a)redhat.com> wrote:
On Mon, 13 Apr 2020 at 01:34, Ivan Ristic <ivan.ristic(a)gmail.com> wrote:
> Hello list,
> Is there any interest in getting HTTP CONNECT over HTTP/2 to work? I
> spent some time on it, but although I fixed two small issues, I hit a third
> that was beyond my undertow knowledge. I was left feeling that getting this
> to work only requires a few connecting pieces, although naturally I could
> be wrong.
It's a cool idea, but it's unlikely we will have time to look at it in the
short term. If you wanted to look into it we should be able to give you
Understood. This is for a project that we may undertake in the second half
of this year, so we may be able to allocate some time then.
That said, a lot of our time would have to be spent on learning undertow.
If there's someone on this list who might be interested in implementing
this feature on a commercial basis, could you please get in touch with me?
> Here's what I discovered:
> - In Http2ReceiveListener, there's code that refuses requests that don't
> have path set (with "No :path header sent in HTTP/2 request"). However,
> path must be omitted with CONNECT per
> - Later on, ConnectHandler doesn't expect the path to be null.
> - The acceptConnectRequest handler from
> is not getting called and the connection is thus not setup. This is because
> Http2ServerConnection#setConnectListener doesn't do anything, unlike its
> counterpart HttpServerConnection.
> For context, I was playing with the idea of building a general-purpose
> forward TCP proxy based on undertow. HTTP/2 is of interest here because of
> multiplexing, which would eliminate connection setup costs [of both TCP and
> TLS] between the client and proxy and substantially improve performance
> when many connections are needed.
> If it matters, I used Jetty's HTTP/2 client library to connect to
> Undertow, although I had to patch it to not send path and scheme when
> method is CONNECT.
> undertow-dev mailing list