Segmentation fault in php caused by svn change 3310

reub · Joined: 11 Nov 2005 Posts: 8 Location: Sydney, Australia

Hi,

For the last few days I've been unable to run cacti from cvs (branches/BRANCH_0_8_6/cacti/), it is segfaulting out when the poller is invoked from cron or command line, and also killing php on my web server when a cacti file is run.

I tracked it down to a change that went into svn revision 3310 - r3309 is ok but 3310 is busted. Specifically it's some change in the file database.php that is causing it.
If I revert to r 3309 (or just revert the changes to database.php) the web interface works OK now and the poller works ok when invoked from command line.

I'm running a FC4-devel Linux system ie php 5.0.5 with mysql 5.0.14 however the same problem also existed with mysql 4.1, so it's not a mysql-5 specific issue as such.

What else can I provide to help narrow down the problem?

TheWitness · Posted: Sat Nov 12, 2005 9:31 pm Post subject:

Did you update your config.php from config.php.dist when you made that change? 3311 is the version ot use, but you must update your config.php.

Please respond.

TheWitness

reub · Joined: 11 Nov 2005 Posts: 8 Location: Sydney, Australia

Yes.

The change made in my config.php was to add the statement

$database_port = "3306";

A diff of config.php vs. config.php.dist now shows only the actual database_password being different.

TheWitness · Posted: Sun Nov 13, 2005 6:32 am Post subject:

And still it does not function with the 3311 version?

TheWitness

reub · Joined: 11 Nov 2005 Posts: 8 Location: Sydney, Australia

Correct. With specifically 3311 I still get this:

[root@tornado include]# php /var/www/html/cacti.reub.net/poller.php
Segmentation fault
[root@tornado include]#

Anything newer than 3309 exhibits the same problem.

TheWitness · Posted: Sun Nov 13, 2005 1:34 pm Post subject:

I would like to debug this issue with you on Monday, do you have time? What is your Time Zone, etc...

TheWitness

reub · Joined: 11 Nov 2005 Posts: 8 Location: Sydney, Australia

Sure. I am in NZ which at the moment is GMT+13.

I'll try to log onto msn messenger during the day as well as evening and catch you there.

reub · Joined: 11 Nov 2005 Posts: 8 Location: Sydney, Australia

FWIW here are the versions of code I am running on that box (it's not a critical system):

Fedora Core 4 rawhide (which will become FC5)
rrdtool: 1.2.11
mysql: 5.0.15-2
cacti: svn branch BRANCH_0_8_6
php: 5.0.5-6
net-snmp-5.2.2-0.rc4.2
apache: 2.0.54-6

cianhughes · Joined: 13 Nov 2005 Posts: 2

I have the same problem, with poller from svn yesterday.
If you wish to debug I am in GMT.
What may complicate the issue, is that i'm running MySQL 5 & PHP 5, but it all used to work, so i don't think that's the problem.

Here is my output:
OK u:0.00 s:0.01 r:4.51
OK u:0.00 s:0.01 r:4.52
OK u:0.00 s:0.01 r:4.52
OK u:0.00 s:0.01 r:4.53
OK u:0.00 s:0.01 r:5.53
OK u:0.00 s:0.02 r:5.54
OK u:0.00 s:0.02 r:5.54
OK u:0.00 s:0.02 r:5.54
OK u:0.00 s:0.02 r:6.55
OK u:0.00 s:0.02 r:6.56
OK u:0.00 s:0.02 r:6.56
OK u:0.00 s:0.02 r:6.57
OK u:0.00 s:0.02 r:7.58
OK u:0.00 s:0.02 r:7.59
OK u:0.00 s:0.02 r:7.59
OK u:0.00 s:0.02 r:7.60
OK u:0.01 s:0.02 r:7.60
OK u:0.01 s:0.02 r:8.62
OK u:0.01 s:0.02 r:8.63
OK u:0.01 s:0.02 r:9.64
OK u:0.01 s:0.02 r:9.64
OK u:0.01 s:0.02 r:10.67
OK u:0.01 s:0.02 r:10.68
OK u:0.01 s:0.02 r:10.68
OK u:0.01 s:0.02 r:13.70
OK u:0.01 s:0.02 r:14.71
12/01/2005 10:37:57 PM - SYSTEM STATS: Time:25.4223 Method:cmd.php Processes:20 Threads:N/A Hosts:5 HostsPerProcess:1 DataSources:94 RRDsProcessed:30
OK u:0.01 s:0.02 r:15.73
zsh: segmentation fault (core dumped) php /usr/local/www/data/cacti/poller.php

reub · Joined: 11 Nov 2005 Posts: 8 Location: Sydney, Australia

The bug causing the segfault for me was fixed a while ago, some code that needed to be committed to cvs wasn't, and php mysteriously dumped core on it.

I now have a bug that I am slowly working on with regards to threading on FC5-test/FC4 rawhide...apparently a glibc/popen problem.

Do you find it works ok if you restrict your poller to do only 1 poll at a time and not 3 in parallel? [look in your settings]

TheWitness · Posted: Fri Dec 02, 2005 11:51 am Post subject:

Yes, the latest GLIBC from FC4 Development is buggy and will cause scripts to fail.

TheWitness

TheWitness · Posted: Fri Dec 02, 2005 11:52 am Post subject:

MySQL 5.x will not be supported until the 0.8.6h release. Depending on testing this weekend, it may be released early next week.

TheWitness

reub · Joined: 11 Nov 2005 Posts: 8 Location: Sydney, Australia

Hi,

Following up, I have just opened this bug report on the erratic behaviour I am seeing when running the poller with more than one thread under Fedora Core/Rawhide (soon to be FC5)

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=175167

The glibc guy there is pretty good so I hope he'll have some ideas soon