zabbix_server service process occupies memory

[damon2L]

Zabbix monitors 100+ hosts and 20000+item, and the data collection cycle is 1 minute.

Virtual environment: memory: 32G; disk: 700G
zabbix version: 6.0.1

Cause of failure: After starting zabbix-server, the buff/cache occupies 50% of the memory; the used memory will occupy 80% of the memory space within 72 hours and will continue to increase, causing the zabbix service to die.

How to solve this problem。

[cyber]

How much memory have you defined for different caches in server config? Running webserver and DB on same host? How much have you reserved for those? Mysql seems to be top consumer here...
Have you thought of implementing proxy(ies)?

[damon2L]

Hi cyber ，

1. vfs_cache_pressure is different (100)
2. Running webserver and DB on same host? Yes，Because the detected hosts are relatively small, I did not use webserver and mysql to deploy separately, and did not use poroxy .
3. zabbix_server.conf :Cache Size=2G,ValueCacheSize =512M,
4. I found that the process that takes up a lot of memory is zabbix_server:ODBC poller and it continues to grow. There are 14000+items in my environment for odbc, StartODBCPoller: 5 .

[tim.mooney]

In addition to all of the very relevant suggestions that cyber made, I would also suggest you consider updating to a later version in the 6.0.x series. 6.0.1 was very early after a major new release; many, many bugs have been fixed by intervening point releases. I'm not suggesting that you go to 6.2.x or 6.4.x, just that you look at the most recent bug fix version in the 6.0.x series.

[damon2L]

Hi tim ,

There is data connection between zabbix and the third-party display platform. I don’t know how to obtain the data. Will my upgrade to 6.0.13 affect the operation of the third-party display platform? I will check the possibility of upgrading

[damon2L]

zabbix_server:ODBC poller

[cyber]

Now it looks like you run also a grafana server there... if that is your "3rd party display platform", then upgrading zabbix should not change anything for grafana and zabbix plugin. just check, if that plugin is latest... update if necessary

Your odbc pollers do not seem to be much busy based on last pic... If you say you have 14k odbc items there... those numbers should be higher there (got 0 values.... )

Also whole load on server s small (less than 1), so it may be somekind of mem leak somewhere...
I would also start with upgrading to latest...

[sahine]

Hi damon,

we have the same Problem. Did you resolve it?