v1alpha1

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

	Spec   ElasticDLJobSpec `json:"spec,omitempty"`
	Status common.JobStatus `json:"status,omitempty"`
}

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

type ElasticDLJobSpec struct {
	// RunPolicy encapsulates various runtime policies of the distributed training
	// job, for example how to clean up resources and how long the job can stay
	// active.
	common.RunPolicy `json:",inline"`

	// A map of ElasticDLReplicaType (type) to ReplicaSpec (value). Specifies the ElasticDL cluster configuration.
	// For example,
	//   {
	//     "Master": ElasticDLReplicaSpec,
	//   }
	ElasticDLReplicaSpecs map[common.ReplicaType]*common.ReplicaSpec `json:"elasticdlReplicaSpecs"`
}

type LegacyV1Alpha1 struct {
	// Deprecated. Specifies the desired number of DeprecatedGPUs the MPIJob should run on.
	// Mutually exclusive with the `Replicas` field.
	// Note that this is deprecated in favor of `ProcessingUnits` field.
	// +optional
	DeprecatedGPUs *int32 `json:"gpus,omitempty"`

	// The maximum number of GPUs available per node.
	// Note that this will be ignored if the GPU resources are explicitly
	// specified in the MPIJob pod spec.
	// This is deprecated in favor of `ProcessingUnitsPerNode` field.
	GPUsPerNode *int32 `json:"gpusPerNode,omitempty"`

	// Specifies the desired number of processing units the MPIJob should run on.
	// Mutually exclusive with the `Replicas` field.
	// +optional
	ProcessingUnits *int32 `json:"processingUnits,omitempty"`

	// The maximum number of processing units available per node.
	// Note that this will be ignored if the processing resources are explicitly
	// specified in the MPIJob pod spec.
	// +optional
	ProcessingUnitsPerNode *int32 `json:"processingUnitsPerNode,omitempty"`

	// The processing resource type, e.g. 'nvidia.com/gpu' or 'cpu'.
	// Defaults to 'nvidia.com/gpu'
	// +optional
	ProcessingResourceType string `json:"processingResourceType,omitempty"`

	// Run the launcher on the master.
	// Defaults to false.
	// +optional
	LauncherOnMaster bool `json:"launcherOnMaster,omitempty"`

	// Specifies the desired number of replicas the MPIJob should run on.
	// The `PodSpec` should specify the number of processing units.
	// Mutually exclusive with the `GPUs` or `ProcessingUnits` fields.
	// +optional
	Replicas *int32 `json:"replicas,omitempty"`

	// Describes the pod that will be created when executing an MPIJob.
	Template corev1.PodTemplateSpec `json:"template,omitempty"`
}

type LegacyV1Alpha2 struct {
	// MPIDistribution specifies name of the MPI framwork which is used
	// Defaults to "OpenMPI"
	// Options includes "OpenMPI", "IntelMPI" and "MPICH"
	MPIDistribution *MPIDistributionType `json:"mpiDistribution,omitempty"`
}

type MPIDistributionType string

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

	Spec   MPIJobSpec      `json:"spec,omitempty"`
	Status apiv1.JobStatus `json:"status,omitempty"`
}

type MPIJobLegacySpec struct {
	// RunPolicy is inline embedded in MPIJobSpec in both v1alpha1.
	*apiv1.RunPolicy `json:",inline"`

	// LegacyV1Alpha1 is legacy fields in v1alpha1 api definition.
	*LegacyV1Alpha1 `json:",inline"`

	// LegacyV1Alpha2 is legacy fields in v1alpha2 api definition.
	*LegacyV1Alpha2 `json:",inline"`
}

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

type MPIJobSpec struct {
	// Specifies the number of slots per worker used in hostfile.
	// Defaults to 1.
	// +optional
	SlotsPerWorker *int32 `json:"slotsPerWorker,omitempty"`

	// `MPIReplicaSpecs` contains maps from `MPIReplicaType` to `ReplicaSpec` that
	// specify the MPI replicas to run.
	MPIReplicaSpecs map[apiv1.ReplicaType]*apiv1.ReplicaSpec `json:"mpiReplicaSpecs"`

	// MainContainer specifies name of the main container which
	// executes the MPI code.
	MainContainer string `json:"mainContainer,omitempty"`

	// `RunPolicy` encapsulates various runtime policies of the distributed training
	// job, for example how to clean up resources and how long the job can stay
	// active.
	RunPolicy apiv1.RunPolicy `json:"runPolicy,omitempty"`

	// LegacySpec reserves the deprecated fields for backward compatibility.
	*MPIJobLegacySpec `json:",inline"`
}

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

	Spec   MarsJobSpec   `json:"spec,omitempty"`
	Status MarsJobStatus `json:"status,omitempty"`
}

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

type MarsJobSpec struct {
	// RunPolicy encapsulates various runtime policies of the distributed training
	// job, for example how to clean up resources and how long the job can stay
	// active.
	commonv1.RunPolicy `json:",inline"`

	// WorkerMemoryTuningPolicy provides multiple memory tuning policies to mars worker
	// spec, such as cache size, cold data paths...
	WorkerMemoryTuningPolicy *MarsWorkerMemoryTuningPolicy `json:"workerMemoryTuningPolicy,omitempty"`

	// WebHost is the domain address of webservice that expose to external users.
	WebHost *string `json:"webHost,omitempty"`

	// MarsReplicaSpecs is a map of MarsReplicaType(key) to ReplicaSpec(value),
	// specifying replicas and template of each type.
	MarsReplicaSpecs map[commonv1.ReplicaType]*commonv1.ReplicaSpec `json:"marsReplicaSpecs"`
}

type MarsJobStatus struct {
	commonv1.JobStatus `json:",inline"`
	// WebServiceAddresses is a list of available webservices addresses for users, its length
	// equals with WebServices.Replicas.
	WebServiceAddresses []string `json:"webServiceAddresses,omitempty"`
}

type MarsWorkerMemoryTuningPolicy struct {
	// PlasmaStore specify the socket path of plasma store that handles shared memory
	// between all worker processes.
	PlasmaStore *string `json:"plasmaStore,omitempty"`

	// LockFreeFileIO indicates whether spill dirs are dedicated or not.
	LockFreeFileIO *bool `json:"lockFreeFileIO,omitempty"`

	// SpillDirs specify multiple directory paths, when size of in-memory objects is
	// about to reach the limitation, mars workers will swap cold data out to spill dirs
	// and persist in ephemeral-storage.
	SpillDirs []string `json:"spillDirs,omitempty"`

	// WorkerCachePercentage specify the percentage of total available memory size can
	// be used as cache, it will be overridden by workerCacheSize if it is been set.
	WorkerCachePercentage *int32 `json:"workerCachePercentage,omitempty"`

	// WorkerCacheSize specify the exact cache quantity can be used.
	WorkerCacheSize *resource.Quantity `json:"workerCacheSize,omitempty"`
}

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

	Spec   PyTorchJobSpec   `json:"spec,omitempty"`
	Status common.JobStatus `json:"status,omitempty"`
}

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

type PyTorchJobSpec struct {
	// RunPolicy encapsulates various runtime policies of the distributed training
	// job, for example how to clean up resources and how long the job can stay
	// active.
	common.RunPolicy `json:",inline"`

	// A map of PyTorchReplicaType (type) to ReplicaSpec (value). Specifies the PyTorch cluster configuration.
	// For example,
	//   {
	//     "Master": PyTorchReplicaSpec,
	//     "Worker": PyTorchReplicaSpec,
	//   }
	PyTorchReplicaSpecs map[common.ReplicaType]*common.ReplicaSpec `json:"pytorchReplicaSpecs"`

	// ModelVersion represents the model output by this job run.
	// +optional
	ModelVersion *v1alpha1.ModelVersionSpec `json:"modelVersion,omitempty"`

	// CacheBackend is used to configure the cache engine for job
	// +optional
	CacheBackend *cachev1alpha1.CacheBackendSpec `json:"cacheBackend"`
}

type PyTorchJobStatus struct {
}

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

	Spec   TFJobSpec          `json:"spec,omitempty"`
	Status commonv1.JobStatus `json:"status,omitempty"`
}

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

type TFJobSpec struct {
	// RunPolicy encapsulates various runtime policies of the distributed training
	// job, for example how to clean up resources and how long the job can stay
	// active.
	commonv1.RunPolicy `json:",inline"`

	// SuccessPolicy defines the policy to mark the TFJob as succeeded when the job does not contain chief or master
	// role.
	// Value "" means the default policy that the job is succeeded if all workers are succeeded or worker 0 completed,
	// Value "AllWorkers" means the job is succeeded if all workers are succeeded.
	// Default to ""
	// +optional
	SuccessPolicy *commonv1.SuccessPolicy `json:"successPolicy,omitempty"`

	// A map of TFReplicaType (type) to ReplicaSpec (value). Specifies the TF cluster configuration.
	// For example,
	//   {
	//     "PS": ReplicaSpec,
	//     "Worker": ReplicaSpec,
	//   }
	TFReplicaSpecs map[commonv1.ReplicaType]*commonv1.ReplicaSpec `json:"tfReplicaSpecs"`

	// ModelVersion represents the model output by this job run.
	// +optional
	ModelVersion *v1alpha1.ModelVersionSpec `json:"modelVersion,omitempty"`

	// CacheBackend is used to configure the cache engine for job
	// +optional
	CacheBackend *cachev1alpha1.CacheBackendSpec `json:"cacheBackend"`
}

type TFReplicaType commonv1.ReplicaType

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

	Spec   XDLJobSpec   `json:"spec,omitempty"`
	Status v1.JobStatus `json:"status,omitempty"`
}

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

type XDLJobSpec struct {
	// RunPolicy encapsulates various runtime policies of the distributed training
	// job, for example how to clean up resources and how long the job can stay
	// active.
	v1.RunPolicy `json:",inline"`

	// XDLReplicaSpecs is map of ReplicaType and ReplicaSpec
	// specifies the XDL replicas to run.
	// For example,
	//   {
	//     "PS": ReplicaSpec,
	//     "Worker": ReplicaSpec,
	//   }
	XDLReplicaSpecs map[v1.ReplicaType]*v1.ReplicaSpec `json:"xdlReplicaSpecs"`

	// MinFinishWorkerNum specifies the minimum number of successfully finished
	// workers such that the job is treated as successful. Not specifying this
	// value means all worker should be successfully finished.
	MinFinishWorkerNum *int32 `json:"minFinishWorkNum,omitempty"`

	// MinFinishWorkPercentage specifies the minimum percentage of all workers
	// that should be finished successfully such that the job is treated as successful.
	// MinFinishWorkPercentage takes precedence over  MinFinishWorkerNum if both are
	// specified.
	MinFinishWorkerPercentage *int32 `json:"minFinishWorkRate,omitempty"`
}

type XDLJobStatus struct {
}

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

	Spec   XGBoostJobSpec   `json:"spec,omitempty"`
	Status XGBoostJobStatus `json:"status,omitempty"`
}

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

type XGBoostJobSpec struct {
	// RunPolicy encapsulates various runtime policies of the distributed training
	// job, for example how to clean up resources and how long the job can stay
	// active.
	RunPolicy commonv1.RunPolicy `json:",inline"`

	// XGBoostReplicaSpecs is map of ReplicaType and ReplicaSpec
	// specifies the XGBoost replicas to run.
	// For example,
	//   {
	//     "PS": ReplicaSpec,
	//     "Worker": ReplicaSpec,
	//   }
	XGBReplicaSpecs map[commonv1.ReplicaType]*commonv1.ReplicaSpec `json:"xgbReplicaSpecs"`
}

type XGBoostJobStatus struct {
	commonv1.JobStatus `json:",inline"`
}