httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Tom Harrington <...@crosstor.com>
Subject Re: Alarms/timeouts in 1.3.x
Date Fri, 21 Apr 2000 16:35:36 GMT
dean gaudet wrote:
> 
> On Thu, 20 Apr 2000, Tom Harrington wrote:
> 
> > I know everyone's focused on 2.0, but hopefully someone can help out
> > a guy still trying to get 1.3.x running on a new OS...
> >
> > We don't have Unix-style signals on this OS, so the use of SIGALRM and
> > the Unix version of ap_set_callback_and_alarm is right out for us.
> >
> > It looks like what I want to do is #define OPTIMIZE_TIMEOUTS for this
> > platform, and then #ifdef out the child_timeouts-related stuff (since
> > I don't plan on supporting child_timeouts anyway).
> 
> if you don't have signals i don't see how optimize_timeouts would work --
> all that does is send the SIGALRM from the parent rather than from the
> kernel.

I do actually have a system call that'll have more or less the same
effect as explicitly sending a SIGALRM, it's scheduling the signal
via alarm() that's difficult.  So in addition to the above I'd need to
replace the kill(pid, SIGALRM) with the OS-specific equivalent.

> >  But I notice that
> > neither Win32 nor Netware used this approach, even though it seems like
> > it would have worked for them.  Is this approach one that I'm likely to
> > regret?
> 
> they use the same approach we use in 2.0 for all architectures --
> individual timeouts on each i/o operation.  look in main/buff.c.

OIC, I did not realize that this was the 2.0 approach, as I've been
too buried in 1.3.x code to look at 2.0 source.  In that case I
suppose it would be better to use this method.

-- 
Tom Harrington
CrosStor Software, Inc.
tph@crosstor.com

Mime
View raw message