httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dean Gaudet <dgau...@arctic.org>
Subject Re: [PATCH] Take III, final version of request-time
Date Mon, 07 Jul 1997 02:56:54 GMT
+1 in principle.

I managed to do something like this in hotwired's code just by mimicing
the setting of r->request_time.  However I chose to move it up to the
connection record ('cause I never feel confident about getting all the
sub request cases properly :).  Then a special module logs the data... to
get a good wall-clock rating I chose to call the times logger specially
in the child_main loop *after* the socket had been closed.  This gets
closer to what the end-user experienced for real-time download time.

The difference in the approaches (other than the use of the data) is that
with yours it requires a little more knowledge of how the core code changes
the child status.  Whereas if you do it right in child_main (or a related
routine) it might be less likely to break in the future.

Dean

On Fri, 4 Jul 1997, Jim Jagielski wrote:

> Aha... I was wondering why my servers were showing almost no time
> at all when processing requests... Seems that process_request()
> would set the child status to BUSY_LOG and then back again to
> BUSY_WRITE. Well, the code triggers on BUSY_WRITE as the start of
> the request... To fix, I added another child stage, DONE_REQ which
> is basically set when the child is done processing the request.


Mime
View raw message