Maintenance windows with SLO

[JAR1]

Hi all.

How do Maintenance windows work for my SLOs?

For example, if the availability in a service "A" is down while a Maintenance windows, how is it calculated in the SLO?

Thanks 

JAR

[Gerhard-K · Accepted solution]

Hello Jar1,

SLOs itself do not consider set maintenance windows, except if they are related to synthetic monitoring. How the specific SLO reacts to the changing situation, based on an active maintenance window, depends on the metrics used and defined timeframes of the SLO evaluation and the maintenance window. For example, if the SLO evaluation window overlaps with a maintenance window, depending on the evaluated parameter, the SLO might not show any data. 

Best,

Gerhard

[Juwei_Zhang1]

"depends" the word confuse me. In general SLO is to measure service level objective (success hit/total hit per dyna definition). it includes all service or we /user selected service but must be most or user-facing service.

maintenance window usually means shut off service for a while for deployment and recycling service. During that time period, either no load the services or service failure but expected. those failed request during MW should not be counted in SLO. this is what we understand and expect

Suppose SLO select -1w past seven day and maintW is included. Please clarify the request, any request happend during the MaintW is included or excluded in SLO calculation?

[Gerhard-K]

Hi @Juwei_Zhang1 ,

You're right; if, for example, the services, which mainly contribute to the overall status of an SLO, are down, it is not per se a problem and hence, a violation of the targeted objective. 

However, if a maintenance window needs to be included or excluded depends on the selected SLA. If you have a contractual SLA with a customer, you might need to consider any downtime in your evaluation. But, I agree, typically for internal SLAs maintenance windows don't need to be considered in the same way.

SLOs don't consider maintenance windows unless the underlying SLI refers to synthetic monitoring. This means, in case of a (planned downtime) no requests are used for evaluation from the selected service.