Hi,

Expanded to add support for mpm worker and prefork.

PATCH: https://github.com/machine-moon/httpd/commit/07edab0c0bb314e140117fb357b24b9f622c0357

Any concerns?


On 2026-07-17 17:49, Tarek wrote:

Hi,

This patch adds optional hooks so modules that manage connections outside the event MPM accept loop (UDP connections) can be accounted for.

Patch against trunk for feedback: https://github.com/machine-moon/httpd/commit/9ae19a4b0af00973d867f0740a709e059b4e6c7c

This patch is critical for mod_http3.

You can find the project here: https://github.com/machine-moon/mod_http3


On 2026-05-12 14:55, Pierre Brochard wrote:

A syntax to specify udp listening sockets should be found for virtualhosts. nginx add quic and reuseport keyword to their listen directive in config files.

Pierre

Le 12/05/2026 à 19:11, Pierre a écrit :
I was thinking of adding the udp ports to the list of ports to listen in the listening thread of thé event pool.

Le 12 mai 2026 18:48, tarek i <[email protected]> a écrit :

    Hi,

    I’m new here and participating in GSoC 2026, currently working
    on the mod_http3 project. I’ve made some cleanups to the
    codebase here: https://github.com/machine-moon/mod_http3/

    Could someone point me toward beginner documentation on the
    Worker MPM?

    Also I’d appreciate any help with testing my changes using
    (curl, chrome, etc).

    Cheers,
    Tarek


    On Tue, May 12, 2026 at 11:53 AM Eric Covener
    <[email protected]> wrote:

        On Tue, May 12, 2026 at 11:24 AM jean-frederic clere
        <[email protected]> wrote:
        >
        > On 5/10/26 6:53 PM, Pierre Brochard wrote:
        > > Hi,
        > >
        > > What about the Event MPM which has workers threads, a
        thread for waiting
        > > sockets and others (for each child) ?
        > >
        > > It may be more adequate than the Worker MPM ?
        >
        > If we want to use the openssl streams like sockets we need
        to write a
        > specialized worker, for the moment the prototype use
        threads that aren't
        > in the MPM logic.
        >
        > I did a bunch of tries and on linux the upd data seem to
        stick on the
        > process/thread that was used to "accept" the new connection.

        Did you see PT_USER / ap_mpm_register_poll_callback_timeout
        in trunk?


--
Cheers,
Tarek

GPG Fingerprint: B9BD295C26A1ED077599810CA284507F96DB79EF

--
Cheers,
Tarek

GPG Fingerprint: B9BD295C26A1ED077599810CA284507F96DB79EF

Reply via email to