zabbix alert: "Utilization of poller processes over 75%" -

[meitaltal]

Hi everyone,

I'm getting a Zabbix alert saying: "Utilization of poller processes over 75%", and I’ve noticed that this issue happens specifically on my Lenovo servers (SNMPv3 via XCC2) . The servers model: SR665 V3

Has anyone else experienced this behavior with Lenovo hardware?

It seems strange that only Lenovo servers would consistently show high poller utilization or respond more slowly compared to others in the same environment.

What could be causing this, and what steps would you recommend to troubleshoot or resolve it?

Thanks in advance!

[cyber]

increase number of pollers. Try to recreate your snmp templates to utilze snmp_pollers instead.
How did you identify, that those get utilized heavily only in case they poll your Lenovo boxes?

[meitaltal]

increase number of pollers. Try to recreate your snmp templates to utilze snmp_pollers instead.
How did you identify, that those get utilized heavily only in case they poll your Lenovo boxes?

Thanks.
I have something like 2000+ servers.

My settings:
CacheSize=8G
ValueCacheSize=4G
StartPollers=200
StartSNMPPollers=800
StartPreprocessors=64
StartPollersUnreachable=100


It's ok?
Now I don't get the error message.

[cyber]

you do not need that much ..:P
one SNMPPoller is good for 1000 async checks... so starting 800 of them is real overkill. and they are specially for new get and walk items, so if you dont have those in your templates, they do not get used at all...
200 pollers seems also a bit excessive... But you have metrics for it, you can see how much utilized they get now, you can decrease accordingly.

[meitaltal]

you do not need that much ..:P
one SNMPPoller is good for 1000 async checks... so starting 800 of them is real overkill. and they are specially for new get and walk items, so if you dont have those in your templates, they do not get used at all...
200 pollers seems also a bit excessive... But you have metrics for it, you can see how much utilized they get now, you can decrease accordingly.

At the beginning, I configured only 64 pollers, but I kept getting an alert saying:
"Utilization of poller processes over 75%."

Because of that, I increased the number of pollers to handle the load and avoid the alert.

Do you think there’s a better way to tune this without over-provisioning?