kfapi

Documentation

Index

• type JobCondition
• type JobConditionType
• type JobStatus
• type MPIJobSpec
• type MXJobSpec
• type PaddleJobSpec
• type PyTorchJobSpec
• type ReplicaStatus
• type ReplicaType
• type RunPolicy
• type TFJobSpec
• type XGBoostJobSpec

Types

type JobCondition

type JobCondition struct {
	// Type of job condition.
	Type JobConditionType `json:"type"`
	// Status of the condition, one of True, False, Unknown.
	Status corev1.ConditionStatus `json:"status"`
	// The reason for the condition's last transition.
	Reason string `json:"reason,omitempty"`
	// A human readable message indicating details about the transition.
	Message string `json:"message,omitempty"`
	// The last time this condition was updated.
	LastUpdateTime metav1.Time `json:"lastUpdateTime,omitempty"`
	// Last time the condition transitioned from one status to another.
	LastTransitionTime metav1.Time `json:"lastTransitionTime,omitempty"`
}

JobCondition describes the state of the job at a certain point.

type JobConditionType

type JobConditionType string

JobConditionType defines all kinds of types of JobStatus.

const (
	// JobCreated means the job has been accepted by the system,
	// but one or more of the pods/services has not been started.
	// This includes time before pods being scheduled and launched.
	JobCreated JobConditionType = "Created"

	// JobRunning means all sub-resources (e.g. services/pods) of this job
	// have been successfully scheduled and launched.
	// The training is running without error.
	JobRunning JobConditionType = "Running"

	// JobRestarting means one or more sub-resources (e.g. services/pods) of this job
	// reached phase failed but maybe restarted according to it's restart policy
	// which specified by user in v1.PodTemplateSpec.
	// The training is freezing/pending.
	JobRestarting JobConditionType = "Restarting"

	// JobSucceeded means all sub-resources (e.g. services/pods) of this job
	// reached phase have terminated in success.
	// The training is complete without error.
	JobSucceeded JobConditionType = "Succeeded"

	// JobFailed means one or more sub-resources (e.g. services/pods) of this job
	// reached phase failed with no restarting.
	// The training has failed its execution.
	JobFailed JobConditionType = "Failed"
)

type JobStatus

type JobStatus struct {
	// Conditions is an array of current observed job conditions.
	Conditions []JobCondition `json:"conditions"`

	// ReplicaStatuses is map of ReplicaType and ReplicaStatus,
	// specifies the status of each replica.
	ReplicaStatuses map[ReplicaType]*ReplicaStatus `json:"replicaStatuses"`

	// Represents time when the job was acknowledged by the job controller.
	// It is not guaranteed to be set in happens-before order across separate operations.
	// It is represented in RFC3339 form and is in UTC.
	StartTime *metav1.Time `json:"startTime,omitempty"`

	// Represents time when the job was completed. It is not guaranteed to
	// be set in happens-before order across separate operations.
	// It is represented in RFC3339 form and is in UTC.
	CompletionTime *metav1.Time `json:"completionTime,omitempty"`

	// Represents last time when the job was reconciled. It is not guaranteed to
	// be set in happens-before order across separate operations.
	// It is represented in RFC3339 form and is in UTC.
	LastReconcileTime *metav1.Time `json:"lastReconcileTime,omitempty"`
}

JobStatus represents the current observed state of the training Job.

type MPIJobSpec

type MPIJobSpec struct {
	SlotsPerWorker  *int32                                                    `json:"slotsPerWorker,omitempty"`
	RunPolicy       RunPolicy                                                 `json:"runPolicy,omitempty"`
	MainContainer   string                                                    `json:"mainContainer,omitempty"`
	MPIReplicaSpecs map[operationv1.MPIReplicaType]*operationv1.KFReplicaSpec `json:"mpiReplicaSpecs"`
}

MPIJobSpec resource definiton.

type MXJobSpec

type MXJobSpec struct {
	RunPolicy      RunPolicy                                                `json:"runPolicy,omitempty"`
	JobMode        operationv1.MXJobModeType                                `json:"jobMode,omitempty"`
	MXReplicaSpecs map[operationv1.MXReplicaType]*operationv1.KFReplicaSpec `json:"mxReplicaSpecs"`
}

MXJobSpec is a desired state description of the MXNetJob.

type PaddleJobSpec

type PaddleJobSpec struct {
	RunPolicy          RunPolicy                                                    `json:"runPolicy,omitempty"`
	ElasticPolicy      *operationv1.PaddleElasticPolicy                             `json:"elasticPolicy,omitempty"`
	PaddleReplicaSpecs map[operationv1.PaddleReplicaType]*operationv1.KFReplicaSpec `json:"paddleReplicaSpecs"`
}

PaddleJobSpec is a desired state description of the TFJob.

type PyTorchJobSpec

type PyTorchJobSpec struct {
	RunPolicy           RunPolicy                                                     `json:"runPolicy,omitempty"`
	ElasticPolicy       *operationv1.PytorchElasticPolicy                             `json:"elasticPolicy,omitempty"`
	NprocPerNode        *string                                                       `json:"nprocPerNode,omitempty"`
	PyTorchReplicaSpecs map[operationv1.PyTorchReplicaType]*operationv1.KFReplicaSpec `json:"pytorchReplicaSpecs"`
}

PyTorchJobSpec is a desired state description of the PyTorchJob.

type ReplicaStatus

type ReplicaStatus struct {
	// The number of actively running pods.
	Active int32 `json:"active,omitempty"`

	// The number of pods which reached phase Succeeded.
	Succeeded int32 `json:"succeeded,omitempty"`

	// The number of pods which reached phase Failed.
	Failed int32 `json:"failed,omitempty"`
}

ReplicaStatus represents the current observed state of the replica.

type ReplicaType

type ReplicaType string

ReplicaType represents the type of the replica. Each operator needs to define its own set of ReplicaTypes.

type RunPolicy

type RunPolicy struct {
	// CleanPodPolicy defines the policy to kill pods after the job completes.
	// Default to Running.
	CleanPodPolicy *operationv1.CleanPodPolicy `json:"cleanPodPolicy,omitempty"`

	// TTLSecondsAfterFinished is the TTL to clean up jobs.
	// It may take extra ReconcilePeriod seconds for the cleanup, since
	// reconcile gets called periodically.
	// Default to infinite.
	TTLSecondsAfterFinished *int32 `json:"ttlSecondsAfterFinished,omitempty"`

	// Specifies the duration in seconds relative to the startTime that the job may be active
	// before the system tries to terminate it; value must be positive integer.
	// +optional
	ActiveDeadlineSeconds *int64 `json:"activeDeadlineSeconds,omitempty"`

	// Optional number of retries before marking this job failed.
	// +optional
	BackoffLimit *int32 `json:"backoffLimit,omitempty"`

	// SchedulingPolicy defines the policy related to scheduling, e.g. gang-scheduling
	// +optional
	SchedulingPolicy *operationv1.SchedulingPolicy `json:"schedulingPolicy,omitempty"`
}

type TFJobSpec

type TFJobSpec struct {
	RunPolicy           RunPolicy                                                `json:"runPolicy,omitempty"`
	EnableDynamicWorker bool                                                     `json:"enableDynamicWorker,omitempty"`
	SuccessPolicy       *operationv1.TFSuccessPolicy                             `json:"successPolicy,omitempty"`
	TFReplicaSpecs      map[operationv1.TFReplicaType]*operationv1.KFReplicaSpec `json:"tfReplicaSpecs"`
}

TFJobSpec is a desired state description of the TFJob.

type XGBoostJobSpec

type XGBoostJobSpec struct {
	RunPolicy       RunPolicy                                                 `json:"runPolicy,omitempty"`
	XGBReplicaSpecs map[operationv1.XGBReplicaType]*operationv1.KFReplicaSpec `json:"xgbReplicaSpecs"`
}

XGBoostJobSpec is a desired state description of the XGBoostJob.