utils

func AddNodeLabel(cl *kubernetes.Clientset, nodeName string, key string, value string) error

func CheckDeploymentWithStandardKMMNFD(cl *kubernetes.Clientset, create bool) error

func CheckGpuLabel(rl v1.ResourceList, label string) bool

func CheckHelmDeployment(cl *kubernetes.Clientset, ns string, create bool) error

func CheckHelmOCDeployment(cl *kubernetes.Clientset, create bool) error

func CheckOCDeploymentWithStandardKMMNFD(cl *kubernetes.Clientset, create bool) error

func CreateConfigMap(ctx context.Context, cl *kubernetes.Clientset, ns string, cmName string, data map[string]string) error

func CreateDaemonset(cl *kubernetes.Clientset, ns string, name string, image string, matchLabels map[string]string, res *v1.ResourceRequirements) error

func CreateDaemonsetVerify(ctx context.Context, cl *kubernetes.Clientset, ns string,
	name string, image string, matchLabels map[string]string,
	res *v1.ResourceRequirements) error

func CreateMinioService(ctx context.Context, cl *kubernetes.Clientset, ns, hostName string) error

func CreateOpaqueSecret(ctx context.Context, cl *kubernetes.Clientset, name, ns string, keys map[string]string) error

func CreatePod(ctx context.Context, cl *kubernetes.Clientset, ns string,
	name string, image string, workerNodeName string) error

func CreateTLSSecret(ctx context.Context, cl *kubernetes.Clientset, name, ns string, crt, key []byte) error

func CreateTempFile(fileName string, data []byte) (*os.File, error)

func CurlMetrics(
	endpointIPs []string,
	token string,
	port int,
	secure bool,
	caCertPath string,
	clientCertPath string,
	clientKeyPath string,
) error

func DelDaemonset(cl *kubernetes.Clientset, ns string, name string) error

func DelRocmPods(ctx context.Context, cl *kubernetes.Clientset) error

func DelRocmPodsByNodeNames(ctx context.Context, cl *kubernetes.Clientset,
	workerNodeNames []string) error

func DeleteConfigMap(ctx context.Context, cl *kubernetes.Clientset, ns string, cmName string) error

func DeleteMinioService(ctx context.Context, cl *kubernetes.Clientset, ns string)

func DeleteNodeAppDaemonSet(cl *kubernetes.Clientset) error

func DeleteNodeLabel(cl *kubernetes.Clientset, nodeName string, key string) error

func DeleteOpaqueSecret(ctx context.Context, cl *kubernetes.Clientset, name, ns string)

func DeletePod(ctx context.Context, cl *kubernetes.Clientset, ns string,
	name string) error

func DeleteRebootPod(ctx context.Context, cl *kubernetes.Clientset, nodeName string, force bool)

func DeleteTLSSecret(ctx context.Context, cl *kubernetes.Clientset, name, ns string) error

func DeleteTempFile(file *os.File) error

func DeployNodeAppDaemonSet(cl *kubernetes.Clientset) error

func DeployResourcesFromFile(pathOrURL string, cl *kubernetes.Clientset, apiCl *apiextClient.Clientset, create bool) error

func DeployRocmPods(ctx context.Context, cl *kubernetes.Clientset,
	res *v1.ResourceRequirements) error

func DeployRocmPodsByNodeNames(ctx context.Context, cl *kubernetes.Clientset,
	workerNodeNames []string) error

func DeployRocmPytorchPods(ctx context.Context, cl *kubernetes.Clientset,
	res *v1.ResourceRequirements) error

func DevicePluginName(cfgName string) string

func ExecPodCmd(command string, ns string, name string, container string) (string, error)

func GenerateServiceAccountToken(clientset *kubernetes.Clientset, serviceAccountName, namespace string) (string, error)

func GetAMDGPUCount(ctx context.Context, cl *kubernetes.Clientset, resourceType string) (map[string]int, error)

func GetAMDGpuWorker(cl *kubernetes.Clientset, isOpenshift bool) []v1.Node

func GetClusterIP(clientset *kubernetes.Clientset, serviceName, namespace string) (string, error)

func GetClusterType(cfg *rest.Config) string

func GetGpuDriverVersion(name string) (string, error)

func GetJobLogs(clientset *kubernetes.Clientset, job *batchv1.Job) ([]string, error)

func GetNodeIP(ctx context.Context, cl *kubernetes.Clientset,
	nodeName string) (string, error)

func GetNodeIPs(clientset *kubernetes.Clientset) ([]string, error)

func GetNodeIPsForDaemonSet(clientset *kubernetes.Clientset, daemonSetName, namespace string) ([]string, error)

func GetNonAMDGpuWorker(cl *kubernetes.Clientset) []v1.Node

func GetPodNamesFromJob(clientset *kubernetes.Clientset, job *batchv1.Job) ([]string, error)

func GetRebootPod(nodeName string) *v1.Pod

func GetRocmInfo(name string) (string, error)

func GetServiceEndpoints(clientset *kubernetes.Clientset, serviceName, namespace string) ([]string, error)

func GetWorkerNodes(cl *kubernetes.Clientset) []*v1.Node

func HandleNodesReboot(ctx context.Context, cl *kubernetes.Clientset, nodes []v1.Node) error

func IsJSONParsable(s string) bool

func ListGpuDrivers(name string) (string, error)

func ListRocmPods(ctx context.Context, cl *kubernetes.Clientset) ([]string, error)

func ListRocmPodsByNodeNames(ctx context.Context,
	workerNodeNames []string) []string

func NFDWorkerName(isOpenshift bool) string

func NodeLabellerName(cfgName string) string

func NodeTaint(cl *kubernetes.Clientset, nodeName string) error

func PatchKMMDeploymentWithCIENVFlag(cl *kubernetes.Clientset) error

func PatchOperatorControllerDeploymentWithCIENVFlag(cl *kubernetes.Clientset) error

func RemoveMinioServiceAccount(ns, pod, container, accessKey string)

func Retry(f func() error, timeout time.Duration, period time.Duration) error

func RunCommand(command string)

func RunCommandOnNode(ctx context.Context, cl *kubernetes.Clientset, nodeName, command string) (string, error)

func SetGPUHealthOnNode(cl *kubernetes.Clientset, ns, gpuid, health, nodeName string) error

func SetupAccessKeysOnMinioServer(ns, pod, container, accessKey, secretKey string)

func SplitYAML(data []byte) [][]byte

func VerifyROCMPODResourceCount(ctx context.Context, cl *kubernetes.Clientset,
	gpuReqCount int, resourceType string) error