Measuring uptime

We determine the SLA of a host by using the 'SLA Report' of Zabbix services. This report bases the measured downtime on triggers with a matching label.

The following must be taken into account:

• Because the calculation is based on the duration of triggers, the triggers can 'filter' the raw data. For example, if the trigger fires when the last three measurements were bad (bad meaning they indicate something is not working correctly, such as the HTTPS service being down), the underlying data could have occasional single/double failures, but the trigger won't fire. To that end, ensure to configure the 'SLA report' to use 1:1 triggers. For example, a trigger that fires as soon as the last measurement was bad.
• Fired triggers can be excluded from the report, but this should only be done for 'false positives' (This means only for triggers that fired while they should not have).

A warning:

When viewing an item graph, Zabbix provides an average value of the observed period. This average value may not always be correct. The average value will be incorrect when viewing trend data of an item with 'integer' as an item type (In zabbix, this is a 'Numeric (unsigned)'). This is because Zabbix calculates the average of the item in a certain time bucket to create the trend data, and turns this average into an integer if the item type is an integer. This means that the average as calculated in the graph is not an 'average of averages', but an average of floored averages, which is not the same.