nvcaconfig

type AgentConfig struct {
	AgentTimeConfig `mapstructure:",squash" yaml:",inline"`

	LogLevel string `yaml:",omitempty"`

	FeatureFlags []string `yaml:",omitempty"`

	SharedStorage             SharedStorageConfig             `yaml:",omitempty"`
	InternalPersistentStorage InternalPersistentStorageConfig `yaml:",omitempty"`

	ICMSURL           string            `yaml:",omitempty"`
	KubeconfigPath    string            `yaml:",omitempty"`
	SvcAddress        string            `yaml:",omitempty"`
	AdminAddr         string            `yaml:",omitempty"`
	DebugAddr         string            `yaml:",omitempty"`
	SystemNamespace   string            `yaml:",omitempty"`
	RequestsNamespace string            `yaml:",omitempty"`
	NamespaceLabels   map[string]string `yaml:",omitempty"`
	ComputeBackend    string            `yaml:",omitempty"`
	StaticGPUCapacity uint64            `yaml:",omitempty"`
	// MinHealthcheckRefreshWait forces the NVCA internal healthchecker
	// to wait at least this long between refresh calls,
	// in case the healthchecker is too chatty in specific instances.
	MinHealthcheckRefreshWait time.Duration `yaml:",omitempty"`

	// MaintenanceMode indicates the operational mode of NVCA
	MaintenanceMode MaintenanceMode `yaml:",omitempty"`

	// HelmRepository restriction
	HelmRepositoryPrefix string `yaml:",omitempty"`

	// ReVal service config
	HelmReValServiceURL string `yaml:",omitempty"`

	// ROS URL
	RolloverServiceURL string `yaml:",omitempty"`

	// CSIVolumeMountOptions for PVC provisioning
	CSIVolumeMountOptions []string `yaml:",omitempty"`

	// Function Deployment Stages service config
	FunctionDeploymentStagesServiceURL string `yaml:",omitempty"`

	// NVCA Operater version
	OperatorVersion string `yaml:",omitempty"`
	// Kubernetes version override
	KubernetesVersionOverride string `yaml:",omitempty"`

	// ImageCredentialHelperImage is the image tag for "nvcf-image-credential-helper",
	// for third party registry cred updates.
	//
	// See https://github.com/NVIDIA/nvcf/nvcf-backend-cluster/image-credential-helper
	ImageCredentialHelperImage string `yaml:",omitempty"`

	// Skip self-destruct even if ICMS sends SELF_DESTRUCT
	SkipSelfDestruct bool `yaml:",omitempty"`
	// Force self-destruct mode for testing
	ForceSelfDestruct bool `yaml:",omitempty"`

	// Secret Mirror
	SecretMirrorSourceNamespace string `yaml:",omitempty"`
	SecretMirrorLabelSelector   string `yaml:",omitempty"`
	// Tolerations are applied to the NVCA agent pod.
	Tolerations []corev1.Toleration `yaml:",omitempty"`

	// UtilsResources contains resources required by the utils container.
	// UtilsResources is subtracted from instance type resources
	// when the agent calculates them for registration.
	// Quantities must be strings.
	UtilsResources ResourceList `yaml:",omitempty"`

	// AdditionalResourceOverhead contains the collective resource overhead from
	// cluster reserved capacity like DaemonSet containers.
	// AdditionalResourceOverhead is subtracted from instance type resources
	// when the agent calculates them for registration.
	// Quantities must be strings.
	AdditionalResourceOverhead ResourceList `yaml:",omitempty"`

	// BYOOResources contains resources required by the BYOO (Bring Your Own Observability)
	// OTel collector container. BYOOResources is subtracted from instance type resources
	// when the agent calculates them for registration.
	// Quantities must be strings.
	BYOOResources ResourceRequirements `yaml:",omitempty"`

	// BYOOFluentBitResources contains resources required by the BYOO (Bring Your Own Observability)
	// FluentBit container. BYOOFluentBitResources is subtracted from instance type resources
	// when the agent calculates them for registration.
	// Quantities must be strings.
	BYOOFluentBitResources ResourceRequirements `yaml:",omitempty"`
}

type AgentTimeConfig struct {
	CredRenewInterval              time.Duration `yaml:",omitempty"`
	HeartbeatInterval              time.Duration `yaml:",omitempty"`
	SyncQueueInterval              time.Duration `yaml:",omitempty"`
	SyncRequestStatusInterval      time.Duration `yaml:",omitempty"`
	SyncAcknowledgeRequestInterval time.Duration `yaml:",omitempty"`
	PeriodicInstanceStatusInterval time.Duration `yaml:",omitempty"`
	RolloverServiceUpdateInterval  time.Duration `yaml:",omitempty"`
	ICMSRequestAckInterval         time.Duration `yaml:",omitempty"`
	ICMSRequestAckRetryTimeout     time.Duration `yaml:",omitempty"`
}

type AllowedExtraKubernetesTypeConfig struct {
	// Group of resource.
	// https://kubernetes.io/docs/reference/using-api/#api-groups
	Group string `json:"group"`
	// Version of resource.
	// https://kubernetes.io/docs/reference/using-api/#api-versioning
	Version string `json:"version"`
	// Kind of resource.
	// https://github.com/kubernetes/community/blob/master/contributors/devel/sig-architecture/api-conventions.md#types-kinds
	Kind string `json:"kind"`
	// Resource name, used in API path.
	// https://kubernetes.io/docs/reference/using-api/api-concepts/#standard-api-terminology
	Resource string `json:"resource"`
}

type AuthzConfig struct {
	// ClientID must come from a secret and set in-memory.
	// It will not be stored when the config is written.
	ClientID string `mapstructure:"-" yaml:"-"`
	// ClientSecretKey must come from a secret and set in-memory.
	// It will not be stored when the config is written.
	ClientSecretKey            string `mapstructure:"-" yaml:"-"`
	ClientSecretsEnvFile       string `yaml:",omitempty"`
	TokenFetchFailureThreshold uint64 `yaml:",omitempty"`
	TokenURL                   string `yaml:",omitempty"`
	TokenScope                 string `yaml:",omitempty"`
	PublicKeysetEndpoint       string `yaml:",omitempty"`
	NGCServiceAPIKeyFile       string `yaml:",omitempty"`
	// NGCServiceAPIKey must come from a secret and set in-memory.
	// It will not be stored when the config is written.
	NGCServiceAPIKey string `mapstructure:"-" yaml:"-"`
	// SelfManagedVaultSecretsJSONPath is the path to the secrets.json file created by Vault
	// for self-managed environments.
	SelfManagedVaultSecretsJSONPath string `yaml:",omitempty"`
}

type Config struct {
	Environment Environment       `yaml:",omitempty"`
	Cluster     NVCFClusterConfig `yaml:",omitempty"`
	Agent       AgentConfig       `yaml:",omitempty"`
	Webhook     WebhookConfig     `yaml:",omitempty"`
	Workload    WorkloadConfig    `yaml:",omitempty"`
	Authz       AuthzConfig       `yaml:",omitempty"`
	Tracing     TracingConfig     `yaml:",omitempty"`
}

type Environment string

type InternalPersistentStorageConfig struct {
	StorageClassName string `yaml:",omitempty"`
	// The desired hard limit for storage in the IPS PVC.
	// More info: https://kubernetes.io/docs/concepts/policy/resource-quotas/
	HardResourceQuota ResourceList `yaml:",omitempty"`
}

type MaintenanceMode string

type NVCFClusterConfig struct {
	Name          string   `yaml:",omitempty"`
	ID            string   `yaml:",omitempty"`
	GroupName     string   `yaml:",omitempty"`
	GroupID       string   `yaml:",omitempty"`
	NCAID         string   `yaml:",omitempty"`
	Region        string   `yaml:",omitempty"`
	Attributes    []string `yaml:",omitempty"`
	CloudProvider string   `yaml:",omitempty"`

	// ValidationPolicy defines Helm ReVal validation policy data passed to ReVal when rendering a Helm chart.
	ValidationPolicy *ValidationPolicyConfig `yaml:",omitempty"`
}

type OTELExporter string

type ResourceList corev1.ResourceList

type ResourceRequirements struct {
	Limits   ResourceList           `yaml:",omitempty"`
	Requests ResourceList           `yaml:",omitempty"`
	Claims   []corev1.ResourceClaim `yaml:",omitempty"`
}

type SharedStorageConfig struct {
	Server   SharedStorageServerConfig   `yaml:",omitempty"`
	TaskData SharedStorageTaskDataConfig `yaml:",omitempty"`
}

type SharedStorageServerConfig struct {
	// Image for the shared storage pod's SMB container.
	Image string `yaml:",omitempty"`
	// ContainerResources are the requests/limits set for the shared storage pod's SMB container
	ContainerResources ResourceRequirements `yaml:",omitempty"`
}

type SharedStorageTaskDataConfig struct {
	// StorageClassName the storage class to use for the task data
	// if none specified ephemeral storage will be used
	StorageClassName *string `yaml:",omitempty"`
	// PVMountOptions represents the mount options for the PV
	PVMountOptions []string `yaml:",omitempty"`
	// StorageCapacity of the provisioned volume.
	StorageCapacity resource.Quantity `yaml:",omitempty"`
}

type TracingConfig struct {
	Exporter             OTELExporter `yaml:",omitempty"`
	LightstepServiceName string       `yaml:",omitempty"`
	// LightstepAccessToken must come from a secret and set in-memory.
	// It will not be stored when the config is written.
	LightstepAccessToken string `mapstructure:"-" yaml:"-"`
}

type ValidationPolicyConfig struct {
	// Name of the policy. This maps to values enumerated by ReVal.
	// Currently "Default" and "Unrestricted" are supported.
	Name string `json:"name"`
	// AllowedExtraKubernetesTypes are a list of Kubernetes group-version-kind + resource sets
	// that the cluster has resources defined for, and allows deploying objects of that type.
	AllowedExtraKubernetesTypes []AllowedExtraKubernetesTypeConfig `json:"allowedExtraKubernetesTypes,omitempty"`
}

type WebhookConfig struct {
	SvcAddress      string            `yaml:",omitempty"`
	TLSKeyFile      string            `yaml:",omitempty"`
	TLSCertFile     string            `yaml:",omitempty"`
	TLSSecretName   string            `yaml:",omitempty"`
	DCGMAnnotations map[string]string `yaml:",omitempty"`
}

type WorkloadConfig struct {
	WorkloadTimeConfig `mapstructure:",squash" yaml:",inline"`

	// Tolerations are applied to translated workload pod specs and related infra pods.
	Tolerations []corev1.Toleration `yaml:",omitempty"`

	// FunctionEnvOverrides is a map of environment variable overrides
	// applied to function workloads before translation.
	// Example keys: INIT_CONTAINER, UTILS_CONTAINER, OTEL_CONTAINER, ESS_AGENT_CONTAINER
	FunctionEnvOverrides map[string]string `yaml:",omitempty"`

	// TaskEnvOverrides is a map of environment variable overrides
	// applied to task workloads before translation.
	// Example keys: INIT_CONTAINER, UTILS_CONTAINER, ESS_AGENT_CONTAINER
	TaskEnvOverrides map[string]string `yaml:",omitempty"`
}

type WorkloadTimeConfig struct {
	// MaxRunningTimeout is the max duration an operation (ex. Helm install, cache Job run) can run for before being considered failed.
	MaxRunningTimeout time.Duration `yaml:",omitempty"`
	// ModelCacheIdlePeriod is the max duration a model cache PV can exist while not in use.
	ModelCacheIdlePeriod time.Duration `yaml:",omitempty"`
	// ModelCacheIdleCleanupPeriod is the period of the model cache cleanup runner.
	ModelCacheIdleCleanupPeriod time.Duration `yaml:",omitempty"`
	// ModelCacheROPVCBindTimeGracePeriod is the max duration an RO PVC can be unbound during cache init.
	ModelCacheROPVCBindTimeGracePeriod time.Duration `yaml:",omitempty"`
	// ModelCacheVolumeDetachmentTimeout is the max duration to wait for PV detachment on cache cleanup.
	ModelCacheVolumeDetachmentTimeout time.Duration `yaml:",omitempty"`
	// defaultWorkerDegradationTimeout is the duration after which a Pod is considered degraded if its status indicates so.
	WorkerDegradationTimeout time.Duration `yaml:",omitempty"`
	// WorkerStartupTimeout is the duration after which a non-".status.phase=(Ready|Succeeded)" Pod is marked failed.
	WorkerStartupTimeout time.Duration `yaml:",omitempty"`
	// PodLaunchThresholdSecondsOnFailedRestarts is the duration after which a Pod with failed and restarting containers is marked failed.
	PodLaunchThresholdSecondsOnFailedRestarts time.Duration `yaml:",omitempty"`
	// PodLaunchThresholdMinutesOnInitFailure is the duration after which a Pod with failed init containers is marked failed.
	PodLaunchThresholdMinutesOnInitFailure time.Duration `yaml:",omitempty"`
	// PodLaunchThresholdMinutesOnInitFailure is the duration after which an un-scheduled Pod is marked failed.
	PodScheduledThreshold time.Duration `yaml:",omitempty"`
	// PodLaunchThresholdMinutesOnInitFailure is the duration after which an init cache Job with unsuccessful Pods is marked failed.
	InitCacheJobFailureThreshold time.Duration `yaml:",omitempty"`
	// PodLaunchThresholdMinutesOnInitFailure is the duration after which a Pod with container pull issues is marked failed.
	MaxImagePullErrorThreshold time.Duration `yaml:",omitempty"`
	// NamespaceStuckTimeout is the duration after which a terminating Namespace is considered stuck.
	NamespaceStuckTimeout time.Duration `yaml:",omitempty"`
	// FailingObjectsBackoffTimeout is the duration to retry transient events (FailedMount, FailedAttachVolume) before marking as failed.
	FailingObjectsBackoffTimeout time.Duration `yaml:",omitempty"`
	// FailingObjectsBackoffRequeueInterval is the interval to requeue reconciliation when objects are failing within the backoff period.
	FailingObjectsBackoffRequeueInterval time.Duration `yaml:",omitempty"`
}