httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chuck Murcko <ch...@topsail.org>
Subject Re: [PATCH] Re: general/1555: excessive error log messages ...
Date Sun, 25 Jan 1998 22:08:04 GMT
Dean Gaudet wrote:
> 
> On Sun, 25 Jan 1998, Chuck Murcko wrote:
> 
> > I agree with Roy. The reasons for his -1 are the reasons we added
> > 'warning' level logging to Apache in the first place. They replaced
> > timeout errors, if I remember correctly.
> 
> Neither of you have yet to explain how this "warning" is of any use
> though.  To me it represents the normal situation -- browsers close
> connections half way through a response all the time; and users get
> kicked off their modems all the time (call waiting, whatever), and are
> replaced with another user at the same ip -- leading to RST.
> 
>     if (sig == SIGPIPE) {
>         ap_snprintf(errstr, sizeof(errstr), "%s lost connection to %s",
>                     timeout_name ? timeout_name : "request",
>                     get_remote_host(current_conn, dirconf, REMOTE_NAME));
>     }
>     else {
>         ap_snprintf(errstr, sizeof(errstr), "%s timed out for %s",
>                     timeout_name ? timeout_name : "request",
>                     get_remote_host(current_conn, dirconf, REMOTE_NAME));
>     }
> 
> So these errors occur whenever there's an RST.
> 
> Oh yeah and win95 boxes never FIN connections, they always RST them.
> Leading to races which can generate SIGPIPE.
> 
> Doesn't acroread also cause this due to its moving around within a
> byte-range served file?
> 
> Maybe I'm totally wrong.  But if it's confusing to me then it's probably
> confusing to our users.
> 
I believe we were getting swamped with bug reports about the timeout
errors. The warning was an attempt to downgrade them in apparent
severity. While most of the time they're normal, they may also point to
excessive packet loss or lack of capacity on the outbound link. That's
why I'd be reticent to just lose them, unless we wanted to replace them
with some kind of configurable timeout rate trigger mechanism that would
scream when too many timeouts happened in a givin sample period.
-- 
chuck
Chuck Murcko            The Topsail Group             West Chester PA
USA
chuck@topsail.org

Mime
View raw message