Agent Memory leak on Windows ?

[erisan500]

Hi all,

Refering to my other post (http://www.zabbix.com/forum/showthread.php?t=1338), I can't imagine that none of you are seeing this behavior.

Can you all graph 2 items and post your results here after running for 24 hours?

The items to monitor are:

PerfCounter= Zabbix[WorkSet],"\Process(ZabbixW32)\Working Set",30 <-- This should remain stable.
PerfCounter= Zabbix[PoolMem],"\Process(ZabbixW32)\Pool Paged Bytes",30 <-- This is always increasing.

The problem i'm having is that the Pool Paged Bytes of the Zabbix agent is ever increasing and never releasing its allocated memory. After a couple days this render your server inaccesible and when trying to stop the Zabbix service results in a BSOD. This is hapening on all kinds of Windows installations (servers, workstations (all W2K Sp4))

Attached you can find a graph of the Zabbix Pool Paged Bytes

Please everyone, I realy love Zabbix, but i can't use it for the above mentioned problem.

Feedback is highly appreciated.

Thanks in advance,

EriSan

[elkor]

Are those triggers somehow different from

perf_counter[\Process(ZabbixW32)\Pool Paged Bytes]

which I have defined on the server side as an item for one of my hosts.

I'm not using active checks and that syntax looks like it belongs in the agent's conf file (I'm only monitoring the windows machines to keep everything centralized at the moment, so I haven't dug into the syntax or idiosyncracies of the Win32 client, that will come in a future phase; so please forgive my ignorance if I'm incorrect).

anyway, If it is, I cannot reproduce the leak you are seeing. Here is my data:

Windows HSADMIN1 5.0.2195 Windows 2000 Service Pack 4 Intel IA-32
1.1alpha12


can you try defining the server side parameter like I have above and see if you get the same results? that would at least tell us if we are comparing apples to apples so to speak

[erisan500]

Elkor, thanks for pointing this out.

I did the test like you explained and monitoring the zabbix pool bytes for a couple hours resulted in ......YES ...... a flat line.
This ONLY when using the pasive agent (not Zabbix Agent active)

So after a couple hours, i changed 1 item from pasive to active check, and allready after 10 minutes you see the memory going up.

So i think I can conclude that the memory leak happens when using active checks.

Thanks erisan

[elkor]

noproblem. At least there is some light on the mystery now.

[bbrendon]

Great. I posted something regarding this in the beta forum.

Eagerly awaiting patches...

[ian_d]

I have the same problem here. I've moved back to passive mode and waiting for a fix.

[Alexei]

I'm confused. Does this leak exist in 1.1beta1? Who triggers this leak, active or passive checks?

[erisan500]

Active checks will trigger the momory leak. (100% sure)
I have this problem with the 1.1alpha12 and 1.1beta1 agents.

Thanks for looking at this.

EriSan

[ian_d]

I agree. I had the problem with 1.1alpha12 then upgraded to 1.1beta1 hoping it would fix it. Out of 20 monitored servers, I only had 3 doing active checks and only those were exhibiting the problem with nonpaged memory.

On another note, Alexei, can something be done about the doubling backslashes in item keys?
See (http://www.zabbix.com/forum/showthre...=4928#post4928)
Althought Zabbix has tremendous potential, this is seriously limiting it's usefulness for me.

Keep up the good work!
Ian

[bbrendon]

Bump! Anything happening with this? It seems like it was brushed under the rug.

[Tys]

Same here, on one out of four servers, see image.
Version of Zabbix is 1.1beta2, Zabbix agent on Windows boxes also v1.1b2, although the only changes from 1.1b1 seems to be the version string;p
Three of the four computers are running Windows 2000/SP4, current security patches and fixes applied, one of them a dual xeon, one old xeon 400Mhz(..), one PIV w/ Hyperthreading and one Xeon w/ Hyperthreading (This is the one who is constantly paging memory, the red line).
What next? Maybe exchanging Hardwareconfigurations? Something to exclude points of possible causes?
Regards, Tys

[Alexei]

May I ask for a 24h or 7 days graph?

[Tys]

May I ask for a 24h or 7 days graph?

I would like to help you this way, but this is a production server... Anyhow, yesterday I had alot of entries in the syslog telling me that a performance counter regarding Exchange would be defective. After solving that and restarting the zabbix service, the poolmem stays constant - well, actually, it wobbles about 4.5 MB +/-200 k. I will try to reproduce that on saturday or sunday.
Regards, Tys

[Alexei]

Serious testing has been performed this week. The following issues are believed to be fixed:

- memory leak
- slow startup
- random hangup

Also, log rotation has been implemented.

Please, download the very latest ZABBIX Win32 agent from www.zabbix.com/donwload.php.

It throws some debug info to the trace file, just ignore it.

Let me know if it works as expected!

[Tys]

Working so far

Hi,
just to give you a feedback, the above mentioned four servers are working for six days in a row now, no apparant memory leak so far, no client has stopped sending data.
Thank you very much for the updated Win32 client!
Regards, Tys