v1alpha1

type AutoScaleStrategy struct {
	MinReplicas *int32                 `json:"minReplicas,omitempty"`
	MaxReplicas *int32                 `json:"maxReplicas,omitempty"`
	AutoScaler  corev1.ObjectReference `json:"autoScaler"`
}

type BatchingStrategy struct {
	// BatchSize represents batch size of requests to be forwarded to backend
	// service.
	BatchSize int32 `json:"batchSize"`
	// TimeoutSeconds represents max waiting duration to aggregate batch requests.
	TimeoutSeconds *int32 `json:"timeoutSeconds,omitempty"`
}

type FrameworkType string

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

	Spec   InferenceSpec   `json:"spec,omitempty"`
	Status InferenceStatus `json:"status,omitempty"`
}

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

type InferenceSpec struct {
	// Framework specify the backend inference framework.
	Framework FrameworkType `json:"framework"`

	// Predictors defines a list of predictor specifications and serving-related strategies.
	Predictors []PredictorSpec `json:"predictors"`
}

type InferenceStatus struct {
	// InferenceEndpoints exposes available serving service endpoint.
	InferenceEndpoint string `json:"inferenceEndpoint,omitempty"`
	// PredictorStatuses exposes current observed status for each predictor.
	PredictorStatuses []PredictorStatus `json:"predictorStatuses,omitempty"`
}

type PredictorSpec struct {
	// Name indicates the predictor name.
	Name string `json:"name"`

	// ModelVersion specifies the name of target model-with-version to be loaded
	// in serving service, ModelVersion object has to be created before serving
	// service deployed.
	ModelVersion string `json:"modelVersion,omitempty"`

	// ModelPath is the loaded model filepath in container.
	ModelPath *string `json:"modelPath,omitempty"`

	// Replicas specify the expected predictor replicas.
	Replicas *int32 `json:"replicas,omitempty"`

	// TrafficWeight defines the traffic split weights across multiple predictors
	TrafficWeight *int32 `json:"trafficWeight,omitempty"`

	// Template describes a template of predictor pod with its properties.
	Template corev1.PodTemplateSpec `json:"template"`

	// AutoScale describes auto-scaling strategy for predictor.
	AutoScale *AutoScaleStrategy `json:"autoScale,omitempty"`

	// Batching specify batching strategy to control how to forward a batch of
	// aggregated inference requests to backend service.
	Batching *BatchingStrategy `json:"batching,omitempty"`
}

type PredictorStatus struct {
	// Name is the name of current predictor.
	Name string `json:"name"`
	// Replicas is the expected replicas of current predictor.
	Replicas int32 `json:"replicas"`
	// ReadyReplicas is the ready replicas of current predictor.
	ReadyReplicas int32 `json:"readyReplicas"`
	// TrafficPercent is the traffic distribution routed to this service.
	TrafficPercent *int32 `json:"trafficPercent,omitempty"`
	// Endpoint is the available endpoints of current predictor.
	Endpoint string `json:"endpoint"`
}