On Mon, 2 Mar 2009 12:45:06 -0800
Ask Bjørn Hansen <a...@develooper.com> wrote:
>
> On Feb 20, 2009, at 8:21, Jared Johnson wrote:
>
> >> The attached diff should fix it for you. I'll commit later if nobody
> >> complains.
> >
> > It appears this still hasn't been applied. I just applied it myself
> > and it works for me; is there anything stopping this from being
> > committed?
>
>
> (Including the patch below for reference).
>
> Could someone summarize the issue for me? It seems a bit heavy
> handed to remove the reuse of the loaded plugins; is that really
> necessary? Or am I misunderstanding what this does?
Well, the plugins are not recompiled or newly initialized (at least
not if you have a queue/ plugin :-)). ->load_plugins() just returns the
loaded and initialized plugins.
When reconnecting to a prefork instance which has auth'd someone before
and you try to auth, you get a "503 but you already said AUTH". This
hasn't popped up until we removed the workaround in plugins/tls (exit
after SSL, see SVN revs 805, 894) for prefork. The "cloning" of the
base instance doesn't do a "deep copy" of the base object, it just
returns a new ref to it. This way the base instance is used for all
connecting clients. We do some cleanup stuff, but probably not
enough... => create new instance instead of reusing a possibly dirty
one.
Mhh, maybe there's an issue with this patch: I think you can't use the
pre-connection hook to check for a fresh config file. That'd be
reloaded on any connect and does not stay shared for all connections
until it changes (see dnswl plugin in SVN's contrib/ for example).
Hanno
