The are many means through which clients can reach the apiserver:

• direct connection
• kubernetes service
• internal load balancer
• others

Currently, it is impossible to detect any issues happening in-between clients and the apiserver and be alerted. For that reason, we need to introduce alerts for each type of connection that would inform users about unavailabilities and potential network issues.

For that purpose, SLO-based alerts are a great fit since they are heavily customizable and we can define different availability targets for each type of connection. However, it requires having a lot of insights about the system, which we currently lack. For that purpose, https://issues.redhat.com/browse/API-1373 was created.

Essentially the goal is to create alerts that would be similar to the existing KubeAPIErrorBudgetBurn but would specifically target client requests and the different hosts they are sent to.