Exploring Quality of Service (QoS) Contracts

Understanding the scheduling process

When we send a request to the Kubernetes API to create a Pod (directly or through one of the controllers), it initiates the scheduling process. To be more precise, what happens next (where it will decide to run a Pod) depends hugely on the resources we defined for the containers that form the Pod. In a nutshell, Kubernetes will decide to deploy a Pod, whenever it is possible, inside one of the nodes that have enough available memory.

Memory requests and limits

When memory requests are defined, Pods will get the memory they requested. If the memory usage of one of the containers exceeds the requested amount, or if some other Pod needs that memory, the Pod hosting it might be terminated. Please note that we wrote that a Pod might be terminated. Whether that will happen depends on the requests from other Pods and the available memory in the cluster. On the other hand, containers that exceed their memory limits are always terminated ...