We are observing an odd, but repeating trend in zabbix with regards to "100% busy pollers" every two hours, exactly on the hour. During this period, Postgresql is very busy, IOWait is high and same with CPU Load. It seems that when the load increases on the database, the pollers can not keep up and we end up with areas of no or missing data in our monitoring data which creates a highly undesirable situation for monitoring our large enterprise network. Additionally, We observe this trend on three separate instances of zabbix, one instance is using MySQL, and the other two are using postgresql. We observe this trend even when housekeeping is turned off. Zabbix is the only software application that talks to each database, so there is not any contention from any other application.
I am wondering if anyone in the community is knowledgeable enough about the zabbix internals that can tell me why we are seeing this trend exactly every two hours.
We see these trends starting at 2:00 A.M, 4:00 A.M. 6:00 A.M. etc.
Thanks,
Randall
I am wondering if anyone in the community is knowledgeable enough about the zabbix internals that can tell me why we are seeing this trend exactly every two hours.
We see these trends starting at 2:00 A.M, 4:00 A.M. 6:00 A.M. etc.
Thanks,
Randall

Comment