v1alpha1

type ClusterPolicy struct {
	metav1.TypeMeta   `json:",inline"`
	metav1.ObjectMeta `json:"metadata,omitempty"`

	Spec   ClusterPolicySpec   `json:"spec,omitempty"`
	Status ClusterPolicyStatus `json:"status,omitempty"`
}

type ClusterPolicyCustomDefaulter struct{}

type ClusterPolicyCustomValidator struct{}

type ClusterPolicyList struct {
	metav1.TypeMeta `json:",inline"`
	metav1.ListMeta `json:"metadata,omitempty"`
	Items           []ClusterPolicy `json:"items"`
}

type ClusterPolicySpec struct {
	// Which type of resource registration is to be used: Device plugin (dp) or Dynamic Resource Allocation (dra).
	// +kubebuilder:validation:Enum=dp;dra
	ResourceRegistration string `json:"resourceRegistration"`

	// To enable resource monitoring via XPU or not. Deploys GPU Plugin or DRA with monitoring enabled and
	// XPU Manager DaemonSet if true.
	ResourceMonitoring bool `json:"resourceMonitoring,omitempty"`

	// Use NFD rule to label nodes.
	UseNFDLabeling bool `json:"useNFDLabeling,omitempty"`

	// Deploy Kubernetes components to integrate with Prometheus.
	PrometheusIntegration bool `json:"prometheusIntegration,omitempty"`

	// Set up Kueue queues for node resources
	// +optional
	EnableKueue bool `json:"enableKueue,omitempty"`

	// Define Kueue queues
	Kueue *KueueQueueSpec `json:"kueue,omitempty"`

	// Enable health monitoring in DP/DRA
	// These values are applied to all the Intel GPU devices in the cluster.
	// Mechanism to monitor the values differ between DP and DRA. DP uses LevelZero API
	// directly, while DRA relies on the health status provided by XPU Manager.
	HealthinessSpec *HealthinessSpec `json:"health,omitempty"`

	// +optional
	DynamicResourceAllocationSpec DynamicResourceAllocationSpec `json:"dra"`
	// +optional
	DevicePluginSpec DevicePluginSpec `json:"dp"`
	// +optional
	XpuManagerSpec XpuManagerSpec `json:"xpu"`

	// Pull secret is shared with all the deployments.
	// +optional
	PullSecret *v1.LocalObjectReference `json:"pullSecret,omitempty"`

	NodeSelector map[string]string `json:"nodeSelector,omitempty"`
	Tolerations  []v1.Toleration   `json:"tolerations,omitempty"`

	// LogLevel to overwrite the default log level of the components.
	// +kubebuilder:validation:Range=0:4
	// +kubebuilder:validation:Default=2
	LogLevel int32 `json:"logLevel,omitempty"`
}

type ClusterPolicyStatus struct {
	DevicePluginStatus string   `json:"devicePluginStatus,omitempty"`
	DRAStatus          string   `json:"draStatus,omitempty"`
	XPUManagerStatus   string   `json:"xpuManagerStatus,omitempty"`
	Errors             []string `json:"errors,omitempty"`
}

type ClusterQueueSpec struct {
	// Name of the cluster queue
	Name string `json:"name"`
	// List of Kueue LocalQueues to create for this ClusterQueue
	LocalQueues []LocalQueueSpec `json:"localQueues"`
}

type DevicePluginSpec struct {
	// Container image for the GPU plugin
	PluginImage string `json:"plugin,omitempty"`
	// Container image for the Level Zero companion container
	LevelzeroImage string `json:"levelzero,omitempty"`

	// AllowIDs and DenyIDs are used to control which devices are registered as resources.
	// Allow or deny certain PCI Device IDs. Both cannot be used together. Format is '0xabcd'.
	AllowIDs []string `json:"allowIDs,omitempty"`
	DenyIDs  []string `json:"denyIDs,omitempty"`

	// ByPathMode controls DRI by-path entries are exposed by the plugin.
	// +kubebuilder:validation:Enum=single;all;none
	ByPathMode string `json:"byPathMode,omitempty"`

	// +kubebuilder:validation:Range=0:4
	// +kubebuilder:validation:Default=1
	LogLevel int32 `json:"logLevel,omitempty"`
}

type DynamicResourceAllocationSpec struct {
	Image string `json:"image,omitempty"`

	LogLevel int32 `json:"logLevel,omitempty"`

	// DeviceTaints controls whether DRA applies taints to the GPU devices if
	// the devices are indicated as unhealthy by the health monitoring.
	DeviceTaints bool `json:"deviceTaints,omitempty"`

	// Enable DRA Pod's health check.
	// +kubebuilder:default=true
	PodHealthCheck bool `json:"podHealthCheck,omitempty"`
}

type GPUFirmwareContent struct {
	// Container image containing the firmware files under /update directory.
	ContainerImage string `json:"containerImage"`

	Files []GPUFirmwareFile `json:"files"`
}

type GPUFirmwareFile struct {
	// +kubebuilder:validation:Enum=GFX;GFX_DATA;GFX_CODE_DATA;GFX_PSCBIN;AMC;FAN_TABLE;VR_CONFIG;OPROM_CODE;OPROM_DATA
	Type string `json:"type"`
	// Filename of the firmware file without any directories.
	FileName string `json:"filename"`
	// SHA256 checksum of the firmware file. Format: sha256:<64 hex characters>.
	// When set, content.containerImage must be digest-pinned (image@sha256:...).
	// +kubebuilder:validation:Pattern=`^sha256:[0-9a-f]{64}$`
	// +optional
	Checksum string `json:"checksum,omitempty"`
}

type GPUFirmwareUpdate struct {
	metav1.TypeMeta   `json:",inline"`
	metav1.ObjectMeta `json:"metadata,omitempty"`

	Spec   GPUFirmwareUpdateSpec   `json:"spec,omitempty"`
	Status GPUFirmwareUpdateStatus `json:"status,omitempty"`
}

type GPUFirmwareUpdateCustomValidator struct {
}

type GPUFirmwareUpdateList struct {
	metav1.TypeMeta `json:",inline"`
	metav1.ListMeta `json:"metadata,omitempty"`
	Items           []GPUFirmwareUpdate `json:"items"`
}

type GPUFirmwareUpdateSpec struct {
	// List of GPU firmwares to be updated.
	Content GPUFirmwareContent `json:"content"`

	// +kubebuilder:validation:Enum=canary;direct
	UpdateMethod string `json:"updateMethod"`

	// Updater container to run the firmware update.
	// Should container xpu-smi tool that is used to perform the updates.
	UpdaterImage string `json:"updaterImage"`

	// Pull secret for the updater image and the content image, if needed.
	ImagePullSecret string `json:"imagePullSecret,omitempty"`

	// Target PCI Device ID in case the node has multiple GPU types. Format is '0xabcd'.
	PCIDeviceID string `json:"pciDeviceID,omitempty"`

	// Taint key to be applied to nodes during firmware update.
	// Taint value is always NoSchedule.
	// +kubebuilder:default=gpufirmware-update
	UpdateTaint string `json:"updateTaint"`

	// Tolerations needed to be scheduled to the node.
	Tolerations []v1.Toleration `json:"tolerations,omitempty"`

	// Node selector to target specific nodes for firmware update.
	NodeSelector map[string]string `json:"nodeSelector,omitempty"`

	// AMCCredentialsSecret is the name of a Kubernetes Secret (in the operator namespace)
	// containing 'username' and 'password' keys used for the AMC firmware update method
	// (redfish interface). Required when any firmware file has type AMC.
	// +optional
	AMCCredentialsSecret string `json:"amcCredentialsSecret,omitempty"`

	// InsecureSkipTLSVerify disables TLS certificate verification when the operator
	// contacts the content image registry to check reachability or verify file checksums.
	// Use this when the registry uses a self-signed or private-CA certificate.
	// Note: this flag only affects the operator's pre-verification calls; the actual
	// firmware-update DaemonSet image pull is controlled by the node's container runtime.
	// +optional
	InsecureSkipTLSVerify bool `json:"insecureSkipTLSVerify,omitempty"`

	// HoldAfterCanary, when true, pauses the update in the canary_done state after a
	// successful canary update and waits for HoldAfterCanary to be set to false.
	// When false (default), the full rollout starts automatically.
	// +optional
	HoldAfterCanary bool `json:"holdAfterCanary,omitempty"`
}

type GPUFirmwareUpdateStatus struct {
	// Indicates the current state of the firmware update process.
	// Moves from "" (not started) -> "draining" -> "updating" -> "cleanup" -> "completed" or "error".
	State string `json:"state"`
	// Messages contains informational or error messages related to the update process.
	Messages []string `json:"messages"`

	// Substatus for nodes.
	NodeInfos GPUFirmwareUpdateSubsetStatus `json:"nodeInfos"`
}

type GPUFirmwareUpdateSubsetStatus struct {
	All       []string `json:"allNodes,omitempty"`
	Pending   []string `json:"pendingNodes,omitempty"`
	Draining  []string `json:"drainingNodes,omitempty"`
	Updating  []string `json:"updatingNodes,omitempty"`
	Completed []string `json:"completedNodes,omitempty"`
	Error     []string `json:"errorNodes,omitempty"`
	Jobs      []string `json:"jobs,omitempty"`
}

type HealthinessSpec struct {
	// Not supported by Device Plugin
	// +kubebuilder:validation:Minimum=1
	// +kubebuilder:validation:Maximum=3600
	// +kubebuilder:default:=5
	CheckIntervalSeconds int32 `json:"checkIntervalSeconds,omitempty"`

	// +kubebuilder:validation:Minimum=1
	// +kubebuilder:validation:Maximum=130
	// +kubebuilder:default:=100
	CoreTemperatureThreshold int32 `json:"coreTemperatureThreshold,omitempty"`

	// +kubebuilder:validation:Minimum=1
	// +kubebuilder:validation:Maximum=130
	// +kubebuilder:default:=100
	MemoryTemperatureThreshold int32 `json:"memoryTemperatureThreshold,omitempty"`
}

type KueueQueueSpec struct {
	// Cluster queues for dividing resources evenly
	EqualResources []ClusterQueueSpec `json:"equalResources"`
}

type LocalQueueSpec struct {
	// Name of the cluster queue
	Name string `json:"name"`
	// Namespace for the local queue
	Namespace string `json:"namespace"`
}

type XpuManagerSpec struct {
	Image string `json:"image,omitempty"`

	// +kubebuilder:validation:Range=0:3
	// +kubebuilder:validation:Default=2
	LogLevel int32 `json:"logLevel,omitempty"`

	// ConfigMapOverride allows overriding the default OpenTelemetry Collector configuration used by XPU Manager.
	// Configmap has to be in the same namespace as the operator and contain a key "config.yaml" with the configuration content.
	// The value should be a YAML string containing the configuration. If not set, a default configuration will be used.
	ConfigMapOverride string `json:"configMapOverride,omitempty"`

	// Set monitoring resource name for Device Plugin use. If not set, the default resource
	// name "gpu.intel.com/xe_monitoring" will be used.
	// +kubebuilder:validation:Enum=i915_monitoring;xe_monitoring;monitoring
	MonitoringResource string `json:"monitoringResource,omitempty"`
}