1.9.7 zabbix_server CPU

[jesmit]

I thought I would ask here before submitting an issue, but I have upgraded to 1.9.7 (beta) from 1.9.6 (beta) and have experienced zabbix_server processes consuming 100% CPU after almost exactly 1 hour, it does not drop down. This is reproducible, once I stop the zabbix_server process and restart, everything is fine (running 95-95% idle) for another hour. Backing out to 1.9.6 has stopped this issue.

Required values per second is 140, monitoring 48 hosts, 28000 items, 15000 triggers, housekeeper set to run ever 12 hours.. mysql 5.1.47.. rhel 6..

Am I missing something obvious that would cause this, or is it potentially a bug?

Thanks,

Jeff

[richlv]

which process is the one using cpu most ? if it is a zabbix_server process, which type (you can see that in the startup messages in the logfile by pid) ?

[jesmit]

I've backed out so I am not 100% sure, but I am pretty confident it was the pollers. I also setup monitoring of zabbix internals (busy pollers) and that value hit 100 around the same time. Increasing the number of pollers didn't help at all. I will try to find an opportunity to re apply 1.9.7 and get the exact type of process. I should also note that all of my items are snmpv2 with the exception of the zabbix server monitor.

I appreciate the response.

Thanks,

J

[masteinhauser]

I can duplicate this issue as it is a large problem in our environment.
I checked the log files and can't find the specific PID process information to determine which it is, only that they are starting up. From what I can see it looks like they are poller processes as well. Mine are not at 100% all the time in the internals data but that my be due to my CPU usage and process load being too high.

Our environment is similar to Jesmit's in that 95% of our resources are monitored via SNMP(this is why we need snmpbulkget) except 4 CentOS servers which are monitored via the zabbix agent. About 85% of the systems we are monitoring are CentOS or Ubuntu linux servers. We are using a custom SNMPv2 template to reduce the number of items checked to less than the default SNMPv2 Device template.

-Myles