Function validateK8sCompute

pkg/operator/resources/validations.go:119–152

(api *userconfig.API, maxMemMap map[string]kresource.Quantity)

var _neuronDevicePluginMemReserve = kresource.MustParse("100Mi")

func validateK8sCompute(api *userconfig.API, maxMemMap map[string]kresource.Quantity) error {
	clusterNodeGroupNames := strset.New(config.ClusterConfig.GetNodeGroupNames()...)
	for _, ngName := range api.NodeGroups {
		if !clusterNodeGroupNames.Has(ngName) {
			return errors.Wrap(ErrorInvalidNodeGroupSelector(ngName, config.ClusterConfig.GetNodeGroupNames()), userconfig.NodeGroupsKey)
		}
	}

	compute := userconfig.GetPodComputeRequest(api)

	for _, ng := range config.ClusterConfig.NodeGroups {
		if api.NodeGroups != nil && !slices.HasString(api.NodeGroups, ng.Name) {
			continue
		}

		nodeCPU, nodeMem, nodeGPU, nodeInf := getNodeCapacity(ng.InstanceType, maxMemMap)

		if compute.CPU != nil && nodeCPU.Cmp(compute.CPU.Quantity) < 0 {
			continue
		} else if compute.Mem != nil && nodeMem.Cmp(compute.Mem.Quantity) < 0 {
			continue
		} else if compute.GPU > nodeGPU {
			continue
		} else if compute.Inf > nodeInf {
			continue
		}

		// we found a node group that has capacity
		return nil
	}

	// no nodegroups have capacity
	return ErrorNoAvailableNodeComputeLimit(api, compute, maxMemMap)
}

func getNodeCapacity(instanceType string, maxMemMap map[string]kresource.Quantity) (kresource.Quantity, kresource.Quantity, int64, int64) {
	instanceMetadata := aws.InstanceMetadatas[config.ClusterConfig.Region][instanceType]