Zabbix server 'out of memory' errors

[yvesd]

Hi,

I run Zabbix server (1.4.1) on a Linux box (RedHat, with 512M or RAM and some paging space).

I receive on regular base 'out of memory' errors, and zabbix_server died.

We monitor 2 Linux boxes, and 6 AIX boxes.

The agent is running on Linux and AIX system, but on AIX we are also using trapper to populate Zabbix with parameters not supported by the agent on AIX (in fact all parameters exepting filesystems...).

Does anyone already saw that? Any help?

thanks in advance.

[bbrendon]

You may want to review the memory leak threads. I think my memory problems went away with the 1.4.2 release. Not 100% sure yet though.

[yvesd]

I did the upgrade to 1.4.2, and the problem is still there.

When not sending trap, the memory usage of zabbix server is quite stable. When I start my 'collectors' on the client and send trap to server, the memory usage for zabbix server start to increase, using all the memory, ant after using all the paging space until the message 'out of memory'.

[Alexei]

I did the upgrade to 1.4.2, and the problem is still there.

When not sending trap, the memory usage of zabbix server is quite stable. When I start my 'collectors' on the client and send trap to server, the memory usage for zabbix server start to increase, using all the memory, ant after using all the paging space until the message 'out of memory'.

I started zabbix_sender in a "while true" loop, I cannot notice growing memory usage of ZABBIX server.

[yvesd]

Hi,
we don't start the zabbix_sender because it's not working on AIX, we implemented the protocol into a C program (to send the base64 encoded value).

Today we crached the machine (512M of ram with 1.8G of paging) after starting zabbix for only 7hours.

Is is possible that if the protocol is not correctly implemented in our 'agent' there is a leak in the memory of the server?

I will try to reproduce the problem with a simple perl script and keep you informed.

[Alexei]

Is is possible that if the protocol is not correctly implemented in our 'agent' there is a leak in the memory of the server?

I have absolutely no idea! You should have mentioned that you use your own extenstions before...

[yvesd]

Sorry about that... but as I mentionned in an other post, zabbix_sender is not working for AIX...

I'm running now a test with a perl script and seems not to have any problem...

I'm now looking to the IPtrace between the client and the server...

Do you have any idea of the 'maximum rate' of trap that the server can receive withour any problem?

[Alexei]

Sorry about that... but as I mentionned in an other post, zabbix_sender is not working for AIX...

ZABBIX 1.4.2 sender should work on any platform. There was a problem with different byte orders on different hardware architectures which is fixed now.

[yvesd]

It seems that when the server receive a key that is not defined into de 'items' the memory used by the server is increasing, and the server need to be restarted for returning to a normal situation...

About the compilation of 1.4.2, I still have problems: link

[Alexei]

It seems that when the server receive a key that is not defined into de 'items' the memory used by the server is increasing, and the server need to be restarted for returning to a normal situation...

Thanks for the hint! Indeed, ZABBIX server didn't free memory when processing non-existant trapper item. It is fixed!

[Palmertree]

The link is broken in the developers download for 1.4.x.

[Palmertree]

It's working now. Thank you.

[Alexei]

The link is broken in the developers download for 1.4.x.

Sorry for this, fixed.

[dsipe]

Hye yvesd,

I'm also monitoring AIX boxes from my Suse Linux Zabbix server.

I've compiled the agent on one of our 5.2 aix server.
All the agent's binaries are working fine on AIX !

I also used items of type trapper, and i have developped some kornshell to retrieve some aix values [ cpu, swapspaces, .. ] that are sent back to zabbix server with the zabbix_sender command....

[yvesd]

We are in fact using nmon, interfaced with a C program that trap the usefull parameters every 15seconds, and send them to the server.

We tried before using script and the configuration into the agent, but that was asking too many resources on the machine.