> Here's the specific issue I'm trying to address with SingleListen, and
> the entire mutex-at-runtime... I build one version of Apache and
> implement it on my suite of servers, say all are running OS Foo 3.4.
> Now some of those servers are single CPU, some are multiCPU. I'd like
> to avoid having separate builds for individual machines. Now say that
> 3.5 comes out, and it supports S_L_U_A... I install it on some
> machines, but not all. It would be nice to be able to continue to use
> the same build of Apache for those as well. I want to give sys-admins
> finer control and avoid the need for recompiles as much as possible.
>
> True, the probability (and usefullness) of the latter is a bit less
> than the former, but it would be nice, and it's a logical extension
> of the concept.
All you gain by introducing SingleListen is the ability to serve requests slightly
faster
on OS Foo 3.5 using a binary compiled under OS Foo 3.4. That's it. IMHO, this small
gain
is more than offset by the additional complexity of the server (the code), additional
complexity of learning and understanding an obscure configuration directive (do you
want
to teach every sysadmin the socket API and how Apache accepts connections? That
knowledge
is prereq to understanding how the SingleListen config directive works), and the real
possibility of enabling a sysadmin to configure a system that fails intermittently in
really hard to debug ways.
Bill
