The new webserver has become unresponsive episodically, under loads
virtually
identical to those the previous server handled without serious
incident. The
server seems to run along smoothly, all indicators well within modest
expections, then suddenly starts adding socket connections and CPU
usage
soars.
Initially, the httpd-error.log showed MaxClients had been reached
(originally 150, now raised to 225) but now more often the CPU's seem
to be
pegged at ~99%. I'm wondering whether this could be related to
eAccelerator,
PHP5-5.3.1 or to other PHP-related issues?
Both virtual servers are configured with 3Gb (3107660k) RAM, 2
processors
allocated, 200 Gb+ storage on a SAN. One, the webserver, runs CentOS
5.4,
Apache 2.2.14, PHP 5.31, eaccelerator 0.9.6, Suhosin Extension 0.9.29,
etc.
The other is a database server running PostgreSQL 8.4.1.
Thinking it may be memory related, I changed the
eaccelerator.shm_size
setting from "16" to "0". As I understand it, this makes eAccelerator
use
the OS's default shared memory setting. This resulted in
eAccelerator's
"Memory Size" doubling from 16Mb to 32Mb.
While tweaking this, I noticed that the kernel settings appeared to
set the
maximum shared memory (shmmax) to 4Gb, which is greater than the
actual
memory of 3Gb. Is this a possible source of the problems we've had,
or is
shmmax not relevant here?
I've naturally Googled 'til my eyes bled and have solicited help from
CentOS & other Linux groups but so far, no joy. Any thoughts and/or
suggestions would be greatly appreciated!
http://brian.moonspot.net/2008/02/13/apache-worker-and-php/
Sounds like a forking stampede.
Brian Moon
> --
> You received this message because you are subscribed to the Google
> Groups "Atlanta PHP" group.
> To post to this group, send email to atlan...@googlegroups.com.
> To unsubscribe from this group, send email to atlantaphp+...@googlegroups.com
> .
> For more options, visit this group at http://groups.google.com/group/atlantaphp?hl=en
> .
>