Recently I've started to receive alerts from Zabbix about high iowait at one of our servers. This event is caused by pg_dump backing up our database and it's perfectly ok because the spike is short and... well, because backing up is a legitimate activity at the server.
Is there a way to teach Zabbix to be smart about such things? I think there is no need to send email alerts about one short (<30secs) spike of iowait per day. On the other hand, if server day is full of 5-secs spikes, this should be investigated.
The most popular and easiest way is to use proper trigger functions. For example, instead of what you most likely use, last()
, go for min(60)
or avg(60)
. Minimum would not detect repeated spiking, average might.
Other potentially useful trigger functions for this purpose:
regexp()
(mostly for text items)str()
(mostly for text items)count()
Note that last()
cannot return or evaluate multiple values, thus last()
= last(0)
= last(300)
.