The following reply was made to PR general/3046; it has been noted by GNATS.
From: [EMAIL PROTECTED] (Miles O'Neal)
To: [EMAIL PROTECTED] (Marc Slemko)
Cc: [EMAIL PROTECTED], [EMAIL PROTECTED]
Subject: Re: general/3046: Apache letting child processes run too long
Date: Wed, 23 Sep 1998 01:35:27 -0500 (CDT)
Marc Slemko said...
|
|> Linux 1.2.8+, gcc 2.6.3, cmopiled with -O2 on everything (I think).
|
|Linux 1.2.8!?!? Erm... that is ancient and could have countless bugs.
Nope. That's where the "+" comes in. It's patched a lot,
and runs beautifully. It ran NCSA 1.[45].x for 3 years without
a hitch, and runs all sorts of complex software, including the
GIMP, some hefty CGI scripts, studly versions of bind and sendmail,
and Netscape Navigator. It wa sone of the first CDDB servers, and
is still going strong. I really think it's capable of basic
counting! 8^)
This is a production system, and not easily upgraded to a whole
new OS. Maybe if Limnux ever gets sensible about upgrades...
|> >Description:
|> I was monitoring my web server, and noticed that Apache
|> seemed to be killing child processes off after about 1355
|> accesses.
|
|What makes you say that? Give exact example output from the status page
|that you say shows this.
Here ya go. Latest figures.
-----------------------------------------------------
Server 0 (6774): 0|475|12260 [Ready] u3.93 s21.56 cu4.33 cs16.9 343 59 (0
B|2.4 MB|88.9 MB)
1cust85.tnt2.tampa.fl.gt.uu.net {GET /~meo/Places/fr/sofitel.html HTTP/1.0}
Server 1 (29964): 0|11737|11737 [Ready] u107.28 s545.04 cu188.65 cs836.8 9 825
(0 B|83.6 MB|83.6 MB)
localhost {GET /status¬able HTTP/1.0}
Server 2 (29966): 1|9181|9181 [Keepalive] u82.37 s422.44 cu145.79 cs660.19 10
97 (151 B|63.3 MB|63.3
MB) localhost {GET /icons/warning.gif HTTP/1.0}
Server 3 (29969): 1|7046|7046 [Write] u62.36 s332.03 cu99.06 cs419.14 0 0 (267
B|44.0 MB|44.0 MB)
localhost {GET /status?notable HTTP/1.0}
Server 4 (29971): 0|5317|5317 [Ready] u46.21 s246.84 cu59.73 cs259.82 148 1370
(0 B|27.7 MB|27.7 MB)
pdx-94.budget.net {POST /~cddb/cddb.cgi HTTP/1.0}
-----------------------------------------------------
Supposedly the access numbers are:
this connection / this child / this slot
Server 1 has a child with well over 10,000 accesses (the current
max). I got my "kill the child" number by subtracting Server 0's
"this child" number from its "this slot" number - that gave me
the same number before that I observed as the "kill the child"
number.
-Miles
