v1alpha1

type Annotations struct {
	Usergroupid string `json:"usergroupid"`
	Userid      string `json:"userid"`
	Priority    string `json:"priority"`
	Scheduler   string `json:"scheduler"`
	Walltime    int    `json:"walltime"`
}

type Framework struct {
	Name FrameworkName `json:"name"`
	Type JobType       `json:"type"`
}

type FrameworkName string

type JobType string

type MasterSpec struct {
	EtcdEndpoint string                      `json:"etcd-endpoint"`
	Resources    corev1.ResourceRequirements `json:"resources"`
	ReplicaSpec  *v1beta1.ReplicaSet         `json:"replicaSpec"`
	Envs         map[string]string           `json:"envs"`

	//for preStop
	GracePeriodSeconds *int64              `json:"grace_period_seconds"`
	PreStopCmd         []string            `json:"pre_stop_cmd"`
	Tolerations        []corev1.Toleration `json:"tolerations"`
	NodeSelector       map[string]string   `json:"node_selector"`
	LivenessProbe      *corev1.Probe       `json:"liveness_probe"`
	ReadinessProbe     *corev1.Probe       `json:"readiness_probe"`
}

type PserverSpec struct {
	Entrypoint       string                        `json:"entrypoint"`
	MinInstance      int                           `json:"min-instance"`
	MaxInstance      int                           `json:"max-instance"`
	Resources        corev1.ResourceRequirements   `json:"resources"`
	ReplicaSpec      *v1beta1.ReplicaSet           `json:"replicaSpec"`
	ImagePullSecrets []corev1.LocalObjectReference `json:"imagePullSecrets"`
	Envs             map[string]string             `json:"envs"`
	//for preStop
	GracePeriodSeconds *int64              `json:"grace_period_seconds"`
	PreStopCmd         []string            `json:"pre_stop_cmd"`
	Tolerations        []corev1.Toleration `json:"tolerations"`
	NodeSelector       map[string]string   `json:"node_selector"`
	//IndexSucceed marks if the operator has added labels to pservers successfully in the initial phase
	IndexSucceed   bool          `json:"index_succeed"`
	LivenessProbe  *corev1.Probe `json:"liveness_probe"`
	ReadinessProbe *corev1.Probe `json:"readiness_probe"`
}

type ResourceState string

type ScaleResults string

type TrainerJobScaleRecord struct {
	// ScaleTimestamp is the time to scale a TrainingJob
	ScaleTimestamp metav1.Time `json:"scaleTimestamp"`
	// Additional is the additional the job to scale
	Additional int32 `json:"additional"`
	// Status is the result of the scale。
	Status ScaleResults `json:"status"`
	// reason is the reason for the scale failed.
	// +optional
	Reason string `json:"reason,omitempty"`
}

type TrainerJobScaleRecords struct {
	ScaleRecords []*TrainerJobScaleRecord
}

type TrainerSpec struct {
	EtcdEndpoint     string                        `json:"etcd-endpoint"`
	Entrypoint       string                        `json:"entrypoint"`
	Workspace        string                        `json:"workspace"`
	MinInstance      int                           `json:"min-instance"`
	MaxInstance      int                           `json:"max-instance"`
	Resources        corev1.ResourceRequirements   `json:"resources"`
	ReplicaSpec      *batchv1.Job                  `json:"replicaSpec"`
	ImagePullSecrets []corev1.LocalObjectReference `json:"imagePullSecrets"`
	Envs             map[string]string             `json:"envs"`
	//for preStop
	GracePeriodSeconds *int64              `json:"grace_period_seconds"`
	PreStopCmd         []string            `json:"pre_stop_cmd"`
	Tolerations        []corev1.Toleration `json:"tolerations"`
	NodeSelector       map[string]string   `json:"node_selector"`
	//IndexSucceed marks if the operator has added labels to trainers successfully in the initial phase
	IndexSucceed   bool          `json:"index_succeed"`
	LivenessProbe  *corev1.Probe `json:"liveness_probe"`
	ReadinessProbe *corev1.Probe `json:"readiness_probe"`
}

type TrainingJob struct {
	metav1.TypeMeta   `json:",inline"`
	metav1.ObjectMeta `json:"metadata,omitempty"`
	Spec              TrainingJobSpec   `json:"spec"`
	Status            TrainingJobStatus `json:"status"`
}

type TrainingJobList struct {
	metav1.TypeMeta `json:",inline"`
	metav1.ListMeta `json:"metadata"`
	// Items means the list of paddle job/TrainingJob
	Items []TrainingJob `json:"items"`
}

type TrainingJobPhase string

type TrainingJobSpec struct {
	// General job attributes.
	Image             string               `json:"image"`
	HostNetwork       bool                 `json:"host_network"`
	Port              int                  `json:"port"`
	PortsNum          int                  `json:"ports_num"`
	PortsNumForSparse int                  `json:"ports_num_for_sparse"`
	TrainerPort       int                  `json:"trainer_port"`
	TrainerPortsNum   int                  `json:"trainer_ports_num"`
	FaultTolerant     bool                 `json:"fault_tolerant"`
	LocalJob          bool                 `json:"local_job"` // LocalJob indicates if the job is local job or cluster job
	Passes            int                  `json:"passes"`
	Volumes           []corev1.Volume      `json:"volumes"`
	VolumeMounts      []corev1.VolumeMount `json:"VolumeMounts"`

	// TODO how to use these two params in matrix
	Mountpath string `json:"mountpath"`
	Nfsmount  string `json:"nfsmount"`

	Annotations Annotations `json:"annotations"`

	//TrainingJob components.
	Master  MasterSpec  `json:"master"`
	Pserver PserverSpec `json:"pserver"`
	Trainer TrainerSpec `json:"trainer"`

	IsNccl    bool       `json:"is_nccl"`
	FrameWork *Framework `json:"frame_work"`
	//Scheduling components.
	SchedulerName string `json:"schedulerName,omitempty"`
	PodGroupName  string `json:"podGroupName,omitempty"`

	// Matrix field indicates whether the backend container is matrix
	Matrix bool `json:"matrix"`
}

type TrainingJobStatus struct {
	// Phase is phase of TrainingJob
	Phase TrainingJobPhase `json:"phase"`
	// Reason is the reason of job phase failed
	Reason string `json:"reason"`
	// ScaleStatus is autoscale status of trainer jobs
	// TODO(ZhengQi): this will used in autoscale mode in future.
	ScaleRecords TrainerJobScaleRecords `json:"scale_records"`
	// ReplicaStatuses is detail status of resources
	// TODO(ZhengQi): should we only considered trainer job now?
	ReplicaStatuses []*TrainingResourceStatus `json:"replica_statuses"`
	// StartTime marks when the trainingjob is Running
	StartTime metav1.Time `json:"startTime"`
	// Released marks resource have been released
	Released bool `json:"released"`
}

type TrainingResourceStatus struct {
	// TrainingResourceType the type of TrainingJob resource, include MASTER PSERVER and TRAINER
	TrainingResourceType `json:"training_resource_type"`
	// State is the state of a type of resource
	State ResourceState `json:"state"`
	// ResourceStates is the number of resource in different state
	ResourceStates map[ResourceState]int `json:"resource_states"`
}

type TrainingResourceType string