CapacityBuffer

The CapacityBuffer custom resource lets you provision spare capacity within a cluster. You can create CapacityBuffers yourself, or GKE can create and manage them for you. To have GKE manage CapacityBuffers, configure capacity buffers in custom ComputeClasses by using the spec.buffers field in a ComputeClass. A CapacityBuffer causes the cluster autoscaler to scale up nodes in advance by simulating placeholder Pods in the cluster autoscaler's memory. This process helps to ensure that spare capacity is available to handle sudden workload spikes or to speed up scaling events. These placeholder Pods are for internal cluster autoscaler use only and aren't visible in your cluster.

CapacityBuffer specification

spec:
  provisioningStrategy: string
  podTemplateRef:
    name: string
  scalableRef:
    apiGroup: string
    kind: string
    name: string
  replicas: integer
  percentage: integer
  limits:
    string: string

Fields
spec.provisioningStrategy optional	string Defines how GKE provisions the buffer. Supported values are "buffer.x-k8s.io/active-capacity" (default) and "buffer.gke.io/standby-capacity" (standby buffer).
spec.podTemplateRef optional	object A reference to a PodTemplate resource in the same namespace that defines the resource requirements for a single unit of buffer capacity. This template provides the specification for placeholder Pods which the cluster autoscaler simulates to provision buffer capacity. You must specify only one of podTemplateRef or scalableRef.
spec.scalableRef optional	object A reference to an object of a kind that has a scale subresource. This allows the CapacityBuffer to manage the buffer by scaling based on an existing scalable resource, such as a Deployment, ReplicaSet, StatefulSet, ReplicationController, or Job. You must specify only one of podTemplateRef or scalableRef.
spec.replicas optional	integer Defines the selected number of buffer units to provision. If you don't set replicas or percentage, GKE creates as many units as fit within defined resource limits. If both are set, GKE uses the maximum of the two values provided they both fit within the limits.
spec.percentage optional	integer Defines the selected buffer capacity as a percentage of the current replicas for the workload specified in the scalableRef field. This setting applies only if you set the scalableRef field. This value can be greater than 100%. GKE calculates the absolute number of replicas from the percentage and rounds up to the nearest whole number, with a minimum of 0.
spec.limits optional	map (string: quantity) Limits the number of units GKE creates for this buffer based on total resource requests (for example, CPU and memory). If you set replicas or percentage, these limits serve as an upper bound, ensuring that buffer capacity does not exceed them. If you don't set replicas or percentage, GKE creates as many units as fit within defined resource limits.

CapacityBuffer status

The status section indicates the observed state of the CapacityBuffer resource.

Fields
status.podTemplateRef	object The observed reference to the PodTemplate that was used to provision the buffer. If this field is not set, and the conditions indicate an error, it provides details about the error state.
status.replicas	integer The number of buffer units that GKE should provision based on fields in the specification such as limits, percentage, and replicas.
status.podTemplateGeneration	integer The observed generation of the PodTemplate, which determines if the status is up-to-date with the selected spec.podTemplateRef.
status.conditions	array (object) List of status conditions for the buffer. The Ready condition indicates if GKE successfully provisioned the buffer. The ReadyForProvisioning condition type indicates if the buffer is compatible with the underlying infrastructure's billing model.
status.provisioningStrategy	string The provisioning strategy used by the buffer.