Re: Re: apache reload and child fault
I have a similar story.
On a Debian/stable, apache (1.3.26-0woody3) with mod_ssl, mod_php4,mod_dav and
auth_mysql on everything from Debian stock except 2.4.24 kernel without
Debian patches.
Reasonnably loaded server (around 10 millions hits / 80 GB of data pushed by
apache every month), several hundreds of MB of log a week spanned over 6
files.
I am using keepalive, and my number of children oscilates between 40 and 105
( I hit the MaxClients once in a while).
During the latest log rotation, I got the "[alert] Child XXXX returned a Fatal
error...".
Right before the log rotation, there were approx. 45 children.
My monitoring shows that apache was completely unable to answer for an hour or
so, then answered during 15 minutes, and then back to nothing (thx mrtg).
When I logged in, I had a bunch a Apache "parents" (parent PID was 1) and very
few children.
"killall -9 apache" got rid of these folks, and "/etc/init.d/apache restart"
restarted the machinery.
mod_ssl is used only for administration purposes. Most recent time it had been
before the log rotation / apache crash was 1 hour and 40 minutes earlier.
Before I fine tuned the server a bit several weeks ago, I had once in a while
those
[warn] child process XXXX still did not exit, sending a SIGTERM
[error] child process XXXX still did not exit, sending a SIGKILL
on several children (usually 2-3, sometimes 20 or more).
Sometimes also, these errors had a
[notice] child pid XXXX exit signal Segmentation fault (11)
right before it.
I've done some serious cleaning (remove useless modules) to try and reduce the
size of my Apache children, and since then I almost did not have anything
wrong in the logs.
Without counting this "Child XXXX returned a fatal error", of course.
The ssl logs that follow the log rotation / crash are:
[11/Jan/2004 aa:bb:33 00590] [info] Init: 5nd restart round (already
detached)
[11/Jan/2004 aa:bb:33 00590] [info] Init: Reinitializing OpenSSL library
[11/Jan/2004 aa:bb:33 00590] [info] Init: Seeding PRNG with 23177 bytes of
entropy
[11/Jan/2004 aa:bb:33 00590] [info] Init: Configuring temporary RSA private
keys (512/1024 bits)
[11/Jan/2004 aa:bb:33 00590] [info] Init: Configuring temporary DH parameters
(512/1024 bits)
[11/Jan/2004 aa:bb:33 00590] [info] Init: Initializing (virtual) servers for
SSL
[11/Jan/2004 aa:bb:33 00590] [info] Init: Configuring server myserver.com:443
for SSL protocol
[11/Jan/2004 cc:dd:19 00566] [info] Connection to child 6 established (server
myserver.com:443, client a.b.c.d)
It is exacly the same restart as when Apache restarts "normally" after the log
rotation (i.e. without a child crashing the entire Apache), except that it
happened three minutes later (sure Apache was already in a bad shape).
Earlier in the thread, somebody was suspricious about mod_ssl: did someone try
not to use mod_ssl and see a difference?
BTW, I have another server on Debian/stable running apache-ssl (Apache
Ben-SSL/1.48 and not apache with mod_ssl) for several years, it never had
such problem. Note also that it has far less traffic and is not running
mod_php4 but modperl.
--
Guillaume Tamboise
"First they ignore you, then they laugh at you, then they fight you,
then you win." -- Gandhi
Reply to: