E accelerator Segmentation Error

Posted by shyamala on March 27, 2009 at 5:54am

Eaccelerator is causing a Segmentation fault on our server. When the segmentation fault occurs the site is accessible, only the node pages are not. There is no error logged in the eaccelerator log. Only error logged is a segmentation error in apache:

child pid <22103> exit signal Segmentation fault (11)

On restart the site starts functioning and error is fixed.

The server conf details:
Hardware Component:
2 Seagate \ 146GB:SAS:15K \ ST3146855SS
4 Generic \ 2048 MB \ DDR2 667 ECC
1 Intel \ 2.13 GHz 1066FSB - Kentsfield \ Xeon 3210 (Quad Core)
1 Dell \ Single socket 1067FSB - Quad Core Capable \ PowerEdge 840
1 Adaptec \ 4-channel SAS/SATA RAID PCI Express \ 3405

Software:
Operating System: Redhat Enterprise Linux - OS ES 5.0
Apache: 2.2.3
PHP: 5.1.6
Mysql: 5.0.45
eAccelerator: 0.9.5.2

User Access and load stats per day:
201912 Pageviews, 40000 visits

E accelerator is definitely giving us a memory and CPU load benefit.

Possible causes:

Some PHP functions could trigger this
Mismatch between e accelerator version and PHP version
E accelerator settings
E- accelerator settings:

zend_extension="/usr/lib/php/modules/eaccelerator.so"
eaccelerator.shm_size="16"
eaccelerator.cache_dir="/var/eaccelerator"
eaccelerator.enable="1"
eaccelerator.optimizer="0"
eaccelerator.check_mtime="1"
eaccelerator.debug="0"
eaccelerator.filter=""
eaccelerator.shm_max="0"
eaccelerator.shm_ttl="0"
eaccelerator.shm_prune_period="0"
eaccelerator.shm_only="0"
eaccelerator.compress="1"
eaccelerator.compress_level="9"
eaccelerator.log_file = "/var/log/httpd/eaccelerator_log"

Immediate measure:

Alerts when segmentation error occurs
Automatic restart of the server

Using code recommended by 2bits refer: "Restart Apache after a segmentation fault"

Please provide any suggestions to work toward a NON segmentation fault state!

Comments

eAccelerator bug

Posted by markus_petrux on March 27, 2009 at 9:05am

http://eaccelerator.net/ticket/155

I was also suffering from this bug and finally opted to install APC, which seems to be more stable and actively developed.

ugh! After all these answers...

Posted by markus_petrux on April 6, 2009 at 6:12pm

...it seems that eAccelerator is a dead project. Well, in fact not only from the look of the answers here, but also from looking at their issues queue.

Sad news again, turk mmcache died, now this one. :-(

I wonder if it would be reasonable to edit references in d.o. to eAccelerator to point the status of the project seems to be dead, so that no one else loose their time with it.

I ran into same issues last

Posted by Alexander Langer@drupal.org (not verified) on March 27, 2009 at 9:43am

I ran into same issues last year with eAccelerator and D5. After I switched to APC I never ever had this problem again and therefore migrated all my other machines to APC, too.

Alex

Use APC

Posted by kbahey on March 27, 2009 at 11:16am

We used the restart php script you link to on our site when we were on 4.7 and things worked reasonably well (one restart a week). Once we moved to Drupal 5, the restarts were once every few hours.

We switched to APC and never had a restart, and don't even run the restart script running.

The reasons for this is that eAccelerator development has all but stopped. No activity in the project at all. Pity, because it is the fastest one available and the most memory efficient too.

APC on the other hand is developed by the core PHP team, and therefore is always in sync with new releases that come out.

Drupal performance tuning, development, customization and consulting: 2bits.com, Inc..
Personal blog: Baheyeldin.com.

Another +1 For APC

Posted by Jamie Holly on March 27, 2009 at 11:56am

I had the same problem on a site I managed when they were on Wordpress. Once we switched to APC the problem was history.

HollyIT - Grab the Netbeans Drupal Development Tool at GitHub.

Thanks for the support! WIll

Posted by shyamala on March 28, 2009 at 4:10am

Thanks for the support! Will switch to APC, and keep you updated.

Shyamala

Netlink Technologies Ltd
I blog and Twitter :)

Shyamala
Unimity Solutions

I can confirm this problem;

Posted by vovaodei on March 30, 2009 at 2:48pm

I can confirm this problem; on a heavy-loaded server with about 230 D6 installations
(~ 60K visits per day) it happened every two hours; the only solution until last week was to restart the apache.

Now i switched to APC - and the problem disappeared in no time,,,
although the cpu-load became around 5-10% greater.

Hardware:
IBM 345 two Dual-Core Xeon 3 GHz 32bit
4 GB RAM
RAID1 U320 SCSI dedicated raid for WWW-stuff
OS: Debian sarge
PHP Version: 5.2.0( 5.2.0-8+etch11)
eaccelerator: 0.9.5.3
kernel 2.6.26
apache2: 2.2.3-4 (2.2.3-4+etch5)
+ identical machine as DB-Server.

By the way - i even do not got the "Segmentation faults" in syslog - simply no nodes on the website/all nodes
got "white page". (And that isn't issue of php_memory_limit - i even tried with 128MB... - no effect).

When I suffered this problem

Posted by Jamie Holly on March 30, 2009 at 2:59pm

When I suffered this problem on Wordpress, I ended up tracking it down to requires and includes. Doing a require on the database layer would break it after awhile. Changing that to a require_once fixed it, even though that require never got hit more than once in the entire system load.

True when PHP6 comes out, APC will be integrated with it. It will be disabled by default, but using it will just be a quick setting change instead of installing a pear package.

HollyIT - Grab the Netbeans Drupal Development Tool at GitHub.

"instead of installing a pear package"? :-)

Posted by vovaodei on March 30, 2009 at 3:09pm

phpize
make
make install
apache2ctl restart

Come on, that was not really an effort. ;-)

Hi All, Our problem seems to

Posted by shyamala on March 31, 2009 at 7:23am

Hi All,

Our problem seems to have temporarily vanished with the eaccelerator optimizer set to 0, eaccelerator.optimizer="0". It was earlier eaccelerator.optimizer="1".

As e-accelerator development is not in progress, we are working towards moving to APC.

Some other unanswered questions:
1) What is the effect of Autologout on high configuration site?
2) Any alternate logging system to reduce the load from a statistics module?
3) The effect of the session settings in php.ini, session.gc_maxlifetime = 1440

Netlink Technologies Ltd
I blog and Twitter :)

Shyamala
Unimity Solutions

Logging system

Posted by andribas on March 31, 2009 at 10:12am

If your server has high load you may consider following approach:
1. Use logs of server (i.e. Apache) + awstats updated by cron.
2. run awstats to parse log on another server or nightly when no load.
3. Apache work slower with logging. I prefer to log only php,html - content, no jpg,css,js etc.
4. Why Apache? nginx works much faster. nginx + php-fpm + e-accelerator 2-3 times faster and don't use so much memory like apache.

eaccelerator.shm_size="16" - you may consider to add some memory here
eaccelerator.compress="1" - there is no need if you use gzip/deflate in apache.

One cause of seg faults

Posted by Netzarim on March 31, 2009 at 4:48pm

Check to see if you have xcache.ini, apc.ini, or eaccelerator.ini in the/etc/php5/conf.d folder. I had a problem with xcache being declared in conf.d and eaccelerator being declared in php.ini. This caused all kinds of problems including seg faults.

Eric

Or php-fcgid

Posted by kbahey on March 31, 2009 at 5:42pm

Using Apache with PHP fcgid provides similar benefits: it is very stable, works well with APC, and provides far better resource utilization than APache with mod_php.

It is available in stable version of distros (e.g. Debian's lenny and Ubuntu server 8.04 LTS).

Do you have benchmarks with Drupal on Apache + fcgid vs. ngnix + fpm to back up the 2-3X speed difference? If so, please post them in a new discussion.

Drupal performance tuning, development, customization and consulting: 2bits.com, Inc..
Personal blog: Baheyeldin.com.