httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andrew Wilson <and...@tees.elsevier.co.uk>
Subject Re: big problem on SunOS
Date Mon, 04 Dec 1995 22:37:50 GMT
> 	Hey ya know I saw this problem too. Last night I had 100
> servers spawned (MaxClients is set way high though..) and i had to
> kernel panic the machine to get it to reboot. This morning it spawned
> a ton of processes but I managed to kill it off before it killed the
> machine. I had figured that it was related to the modules that I added
> which haven't been tested well...
> 	(SunOS 4.1.3_U1)
> 
> > 
> > Actually, I don't know yet whether this is restricted to SunOS or not.
> > The problems I reported earlier with Apache 0.8.16 are still in 1.0.0 --
> > my server went to hell three times this weekend and it is directly
> > caused by Apache spawning to death.  It is not obeying MaxClients at all.
> > 
> > I am trying to find a solution, but I think the problem is somehow
> > related to the process of terminating a child when it has finished
> > its last allowed request.  Naturally, being away at conferences for
> > two weeks is making this difficult.
> > 
> > Note: this problem physically kills the server -- the only way to
> > recover is to power the machine off.  It did not occur before 0.8.15,
> > so I am going to remove that scoreboard tweak from http_main.c next.
> > 
> > ......Roy


Boys, now listen.

1)	Apache's /tmp/htstatus.pid file *has* to exist all the time
	otherwise the server will go into meltdown just as you have
	described.

2)	/tmp is volatile space and is almost certainly being trawled by
	scripts that look for old files to throw away.

3)	Most ppl start httpd from SunOS's rc.local, and on my new December 94
	box /tmp is cleared *after* the httpd is started in rc.local following
	a reboot, or a reboot after a panic or some non-webserver related
	foolishness.

SOME THINGS TO LOOK OUT FOR

1)	Reboot yer box, what happens first /tmp getting cleared or httpd
	getting started.  Are you starting httpd from rc.local?

2)	Play god, type 'rm /tmp/htstatus.*' (make a note of the boss
	process ID before you do this, and have a window ready to
	'kill -TERM logs/httpd.pid'.  Watch your machine die by typing
	'ps -ax | grep httpd'.

3)	All bets are off if you've been smart enough to put htstatus.* in
	/var

> sameer

Cheers,
Ay.

Mime
View raw message