Overtime, the SLO community has created specific terminology. Since not every user coming to project like OpenSLO knows exactly what everything means we have created a glossary. This glossary should help newcomers understand things that we discuss here and make contributions easier.
In addition, I have created a section with literature that I used to write these definitions. There is a need for a place where we can write down definitions to facilitate easier discussions about subjects adn share differing point of view.
A carefully defined quantitative measure of some aspect of the level of service that is provided.
A target value or range of values for a service level that is measured by an SLI.
An explicit or implicit contract with your users that includes consequences of meeting (or missing) the SLOs they contain.
Time in which reliability of service is measured and calculated.
A measurement that allows us to see how much SLO can be violated in a given period of time.
Error budget calculation method. It is calculated by counting the number of good events in the total count of the events. Example: Count if total number of measured latencies, measured in last two weeks, was below 100 ms for 95% of measurements.
Error budget calculation method. This method uses ratio of good slices to total slices in budgeting period. Example: How many minutes, in last two weeks, your service had latency below 100ms in measurements in these minutes.
Error budget calculation method. It is calculated by taking average of good events to total events in all slices. Example: What is average percent of good events to total events in every minute for last two weeks.
Is an SLI composed of a relation of two metrics. Most commonly called good (numerator) and total events (denominator). When picking objective for your service you specify how big part of total events should be good events.
SLI where metrics returned form a query are compared to an arbitrarily chosen threshold. A comparator can be lt(<), le(<=), gt(>), ge(>=). Based on the result of a comparison is determined if the error budget is burnt or not.
It is SLO that is composed of few different objectives that can be seen as SLOs as well. Each objective that is part of composite SLO can have its own threshold, query and data source. Composite SLO burns its error budget when any of SLOs beneath burns their error budget. User should be able to specify weight that will decide if composite SLO is burning faster or slower depending on which inner SLO is burning its error budget.
Books that are great to learn about SLO concept: