Cacti 1.2.24 - Poller.php starves all memory when MariaDB is stopped

[mbordoni]

To make a backup of my Cacti raspi server I decided to stop mariadb and I found that the server hanged up very soon.
I am using cmd.php as a poller.
After some debugging I found that Poller.php (the second process run from cron) once the mariadb is stopped gets all the cpu and started looping till it consumes all the available memory.

I went to php.in and set a memory_limit but it was no worth.
I went to poller.php itself and I found the following lines of code:
"
/* let PHP only run 1 second longer than the max runtime,
* plus the poller needs lot's of memory
*/
ini_set('max_execution_time', MAX_POLLER_RUNTIME + 1);
ini_set('memory_limit', '-1');
"
and set a limit again
"
/* let PHP only run 1 second longer than the max runtime,
* plus the poller needs lot's of memory
*/
ini_set('max_execution_time', MAX_POLLER_RUNTIME + 1);
ini_set('memory_limit', '1500M');
"

This is a bad design (never allow "no limits" it is not safe). Either set a limit, or remove it and let to use default settings, or, finally, let a config inside cacti configuration to set memory limit accordingly.
After this if someone is able to find the root cause of the loop and memory leak in that situation would be perfect.

Bests,

[TheWitness]

Sounds like you've got a php version with a memory leak. It used to be real bad in the php4.x days. Not so much any more.

[mbordoni]

BTW: I am using PHP 8.2.18 on a raspi with Debian bookworm.

The design flaw is really in poller.php that has the setting to allow infinite memory consumption.
NEVER DO THIS, IT IS IT HARAKIRI!

For the root cause I think it is due to something about the permanent db connection end the related pooling that cannot understand if and when db goes down.

If you put a big at will amount of memory in the setting at most the process is killed and this prevents the server from hanging up.

In my case when I start the backup procedure I stop cron, apache and mariadb processes and in case this process ends is a good way to proceed since at the end of backup by starting the 3 services I will get it back on via cron scheduling.

[TheWitness]

Upgrade to 1.2.27 as soon as it's out. That may not be the solution, but there were some light design changes between releases. Also, check for a memory leak in PHP. Check the php-fpm and apache processes over time. You may need to reduce the default servers and threads in the apache config on small machines like a Rasberry Pi. Take a look at this article, and then update for your version of Apache:

https://serverfault.com/questions/77585 ... oncurrency

Tips:

Minimize Servers
Reduce Lifetime of any process (as they will accumulate memory in cache)
Minimize Threads

But, keep an eye on the Cactid service (if you are running it).

I hope this helps.