v1

package
v0.0.0-...-e91ad15 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Sep 27, 2024 License: Apache-2.0 Imports: 5 Imported by: 0

Documentation

Overview

Package v1 is the v1 version of the API. +groupName=kubeflow.org

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at 

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

Index

Constants

View Source 
const (
	// ReplicaIndexLabel represents the label key for the replica-index, e.g. the value is 0, 1, 2.. etc
	ReplicaIndexLabel = "replica-index"

	// ReplicaTypeLabel represents the label key for the replica-type, e.g. the value is ps , worker etc.
	ReplicaTypeLabel = "replica-type"

	// ReplicaNameLabel represents the label key for the replica-name, the value is replica name.
	ReplicaNameLabel = "replica-name"

	// GroupNameLabel represents the label key for group name, e.g. the value is kubeflow.org
	GroupNameLabel = "group-name"

	// JobNameLabel represents the label key for the job name, the value is job name
	JobNameLabel = "job-name"

	// JobRoleLabel represents the label key for the job role, e.g. the value is master
	JobRoleLabel = "job-role"
)
View Source 
const (
	KubeDLPrefix = "kubedl.io"

	// LabelPlatform indicates the platform running instances from.
	LabelPlatform = KubeDLPrefix + "/platform"
	// LabelGangScheduler indicates a specific gang scheduler name.
	LabelGangScheduler = KubeDLPrefix + "/gang-scheduler"
	// LabelDisableGangScheduling indicates job disables gang scheduling.
	LabelDisableGangScheduling = KubeDLPrefix + "/disable-gang-scheduling"
	// LabelGangSchedulingJobName indicates name of gang scheduled job.
	LabelGangSchedulingJobName = KubeDLPrefix + "/gang-job-name"
	// LabelCronName indicates the name of cron who created this job.
	LabelCronName = KubeDLPrefix + "/cron-name"

	// LabelTargetNode is the target node allocated by gpu coordinator.
	LabelTargetNode = "alloc-group.scheduling.sigs.k8s.io/target_node"
	// LabelAllocToken is the token of the alloc-consumer pod, which is the UID of the corresponding alloc-reserver pod.
	LabelAllocToken = "alloc-group.scheduling.sigs.k8s.io/token"

	// AnnotationGitSyncConfig annotate git sync configurations.
	AnnotationGitSyncConfig = KubeDLPrefix + "/git-sync-config"
	// AnnotationTenancyInfo annotate tenancy information.
	AnnotationTenancyInfo = KubeDLPrefix + "/tenancy"
	// AnnotationKubeflowTenancyInfo annotate tenancy information.
	AnnotationKubeflowTenancyInfo = "kubeflow.org/tenancy"
	// AnnotationGPUVisibleDevices is the visible gpu devices in view of process.
	AnnotationGPUVisibleDevices = "gpus." + KubeDLPrefix + "/visible-devices"
	// AnnotationSkipDAGScheduling skips dag scheduling scheme for special workloads.
	AnnotationSkipDAGScheduling = KubeDLPrefix + "/skip-dag-scheduling"

	// AnnotationTensorBoardConfig annotate tensorboard configurations.
	AnnotationTensorBoardConfig = KubeDLPrefix + "/tensorboard-config"
	// ReplicaTypeTensorBoard is the type for TensorBoard.
	ReplicaTypeTensorBoard ReplicaType = "TensorBoard"
	// ReplicaTypeAllReduceWorker is the type for all-reduce workers
	ReplicaTypeAllReduceWorker ReplicaType = "Worker"
	// JobReplicaTypeAIMaster means the AIMaster role for all job
	JobReplicaTypeAIMaster ReplicaType = "AIMaster"

	// AllocGroupJobName marks which job the allocgroup belongs to.
	// The value is the job name.
	AllocGroupJobName = KubeDLPrefix + "/allocgroupjobname"

	// AnnotationInternalNetworkMode annotate pod network mode
	// Only used internal.
	AnnotationInternalNetworkMode = KubeDLPrefix + "/network-mode"

	// AnnotationEnableHostNetwork annotate job enable hostnetwork mode.
	AnnotationEnableHostNetwork = KubeDLPrefix + "/enable-hostnetwork"
)

Constant label/annotation keys for job configuration.

View Source 
const (
	DefaultKubeDLNamespace                 = "kubedl"
	ResourceNvidiaGPU      v1.ResourceName = "nvidia.com/gpu"
)

Variables

This section is empty.

Functions

func RegisterDefaults 

func RegisterDefaults(scheme *runtime.Scheme) error

RegisterDefaults adds defaulters functions to the given scheme. Public to allow building arbitrary schemes. All generated defaulters are covering - they call all nested defaulters.

Types

type CleanPodPolicy 

type CleanPodPolicy string

CleanPodPolicy describes how to deal with pods when the job is finished. 

const (
	CleanPodPolicyUndefined CleanPodPolicy = ""
	CleanPodPolicyAll       CleanPodPolicy = "All"
	CleanPodPolicyRunning   CleanPodPolicy = "Running"
	CleanPodPolicyNone      CleanPodPolicy = "None"
)

type ControllerInterface 

type ControllerInterface interface {
	// Returns the Controller name
	ControllerName() string

	// Returns the GroupVersionKind of the API
	GetAPIGroupVersionKind() schema.GroupVersionKind

	// Returns the GroupVersion of the API
	GetAPIGroupVersion() schema.GroupVersion

	// Returns the Group Name(value) in the labels of the job
	GetGroupNameLabelValue() string

	// Returns the Job from Informer Cache
	GetJobFromInformerCache(namespace, name string) (v1.Object, error)

	// Returns the Job from API server
	GetJobFromAPIClient(namespace, name string) (v1.Object, error)

	// GetPodsForJob returns the pods managed by the job. This can be achieved by selecting pods using label key "job-name"
	// i.e. all pods created by the job will come with label "job-name" = <this_job_name>
	GetPodsForJob(job interface{}) ([]*corev1.Pod, error)

	// GetServicesForJob returns the services managed by the job. This can be achieved by selecting services using label key "job-name"
	// i.e. all services created by the job will come with label "job-name" = <this_job_name>
	GetServicesForJob(job interface{}) ([]*corev1.Service, error)

	// DeleteJob deletes the job
	DeleteJob(job interface{}) error

	// UpdateJobStatus updates the job status and job conditions
	UpdateJobStatus(job interface{}, replicas map[ReplicaType]*ReplicaSpec, jobStatus *JobStatus, restart bool) error

	// UpdateJobStatusInApiServer updates the job status in API server
	UpdateJobStatusInApiServer(job interface{}, jobStatus *JobStatus) error

	// CreateService creates the service
	CreateService(job interface{}, service *corev1.Service) error

	// DeleteService deletes the service
	DeleteService(job interface{}, name string, namespace string) error

	// CreatePod creates the pod
	CreatePod(job interface{}, pod *corev1.Pod) error

	// DeletePod deletes the pod
	DeletePod(job interface{}, pod *corev1.Pod) error

	// SetClusterSpec sets the cluster spec for the pod
	SetClusterSpec(ctx context.Context, job interface{}, podTemplate *corev1.PodTemplateSpec, rtype, index string) error

	// Returns the default container name in pod
	GetDefaultContainerName() string

	// Get the default container port name
	GetDefaultContainerPortName() string

	// Get the default container port number
	GetDefaultContainerPortNumber() int32

	// Get replicas reconcile orders so that replica type with higher priority can be created earlier.
	GetReconcileOrders() []ReplicaType

	// Returns if this replica type with index specified is a master role.
	// MasterRole pod will have "job-role=master" set in its label
	IsMasterRole(replicas map[ReplicaType]*ReplicaSpec, rtype ReplicaType, index int) bool
}

ControllerInterface defines the Interface to be implemented by custom operators. e.g. tf-operator needs to implement this interface

type DAGCondition 

type DAGCondition struct {
	// Upstream defines which replica type is the source tigger.
	Upstream ReplicaType `json:"upstream"`
	// OnPhase defines at which phase the upstream replica will trigger this condition.
	OnPhase v1.PodPhase `json:"onPhase"`
}

type GPUTopologyPolicy 

type GPUTopologyPolicy struct {
	// Platform indicates the platform a job is running on, e.g., ASI or ACK.
	Platform *string `json:"platform,omitempty"`
	// IsTopologyAware indicates whether a job is aware of the GPU topology. To topology-aware jobs, GPU coordinator
	// tries its best to allocate GPUs with fast communication links. Workers of a topology-aware job are aware of the
	// GPU devices of peer-workers on the same node, so that they could talk via p2p links like NVLink.
	IsTopologyAware bool `json:"isTopologyAware,omitempty"`
}

GPUTopologyPolicy specifies gpu topology-related policies that impact the scheduling decisions of the GPU coordinator.

type JobCondition 

type JobCondition struct {
	// Type of job condition.
	Type JobConditionType `json:"type"`
	// Status of the condition, one of True, False, Unknown.
	Status v1.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"`
}

+k8s:deepcopy-gen=true JobCondition describes the state of the job at a certain point.

func (*JobCondition) DeepCopy 

func (in *JobCondition) DeepCopy() *JobCondition

DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new JobCondition.

func (*JobCondition) DeepCopyInto 

func (in *JobCondition) DeepCopyInto(out *JobCondition)

DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.

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"`
}

+k8s:deepcopy-gen=true JobStatus represents the current observed state of the training Job.

func (*JobStatus) DeepCopy 

func (in *JobStatus) DeepCopy() *JobStatus

DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new JobStatus.

func (*JobStatus) DeepCopyInto 

func (in *JobStatus) DeepCopyInto(out *JobStatus)

DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.

type NetworkMode 

type NetworkMode string

NetworkMode specify network mode for job. 

const (
	// HostNetworkMode means the pods of the job will use host network.
	HostNetworkMode NetworkMode = "host"
)

type ProceedCondition 

type ProceedCondition struct {
	Type  ReplicaType      `json:"type,omitempty"`
	Phase JobConditionType `json:"phase,omitempty"`
}

type ReplicaSpec 

type ReplicaSpec struct {
	// Replicas is the desired number of replicas of the given template.
	// If unspecified, defaults to 1.
	Replicas *int32 `json:"replicas,omitempty"`

	// Template is the object that describes the pod that
	// will be created for this replica. RestartPolicy in PodTemplateSpec
	// will be overide by RestartPolicy in ReplicaSpec
	Template v1.PodTemplateSpec `json:"template,omitempty"`

	// Restart policy for all replicas within the job.
	// One of Always, OnFailure, Never and ExitCode.
	// Default to Never.
	RestartPolicy RestartPolicy `json:"restartPolicy,omitempty"`

	// DependOn represents a list of upstream vertex conditions to be dependent on for this RepicaType to start.
	// For example, in TensorFlow workers depend on ps to start first. If not set, KubeDL will populates the
	// default DependOn based on each framework's requirements. This feature is enabled by default, and can be
	// disabled with DAGScheduling feature gate.
	DependOn []DAGCondition `json:"-"`
}

+k8s:deepcopy-gen=true ReplicaSpec is a description of the replica

func (*ReplicaSpec) DeepCopy 

func (in *ReplicaSpec) DeepCopy() *ReplicaSpec

DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new ReplicaSpec.

func (*ReplicaSpec) DeepCopyInto 

func (in *ReplicaSpec) DeepCopyInto(out *ReplicaSpec)

DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.

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"`

	// The number of pods which reached phase Failed and reason is Evicted,
	// it is included in the number of Failed.
	Evicted int32 `json:"evicted,omitempty"`
}

ReplicaStatus represents the current observed state of the replica.

func (*ReplicaStatus) DeepCopy 

func (in *ReplicaStatus) DeepCopy() *ReplicaStatus

DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new ReplicaStatus.

func (*ReplicaStatus) DeepCopyInto 

func (in *ReplicaStatus) DeepCopyInto(out *ReplicaStatus)

DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.

type ReplicaType 

type ReplicaType string

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

type RestartPolicy 

type RestartPolicy string

RestartPolicy describes how the replicas should be restarted. Only one of the following restart policies may be specified. If none of the following policies is specified, the default one is RestartPolicyAlways. 

const (
	RestartPolicyAlways    RestartPolicy = "Always"
	RestartPolicyOnFailure RestartPolicy = "OnFailure"
	RestartPolicyNever     RestartPolicy = "Never"

	// RestartPolicyExitCode policy means that user should add exit code by themselves,
	// The job operator will check these exit codes to
	// determine the behavior when an error occurs:
	// - 1-127: permanent error, do not restart.
	// - 128-255: retryable error, will restart the pod.
	RestartPolicyExitCode RestartPolicy = "ExitCode"
)

type RunPolicy 

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

	// GPUTopologyPolicy defines the policy declared for desired GPU topology pattern.
	// +optional
	GPUTopologyPolicy *GPUTopologyPolicy `json:"gpuTopologyPolicy,omitempty"`

	// ProceedCondition defines a simple dag which holds other replica before one replica running
	// +optional
	ProceedCondition *ProceedCondition `json:"proceedCondition,omitempty"`
}

+k8s:deepcopy-gen=true 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.

func (*RunPolicy) DeepCopy 

func (in *RunPolicy) DeepCopy() *RunPolicy

DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new RunPolicy.

func (*RunPolicy) DeepCopyInto 

func (in *RunPolicy) DeepCopyInto(out *RunPolicy)

DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.

type SchedulingPolicy 

type SchedulingPolicy struct {
	MinAvailable *int32 `json:"minAvailable,omitempty"`
}

SchedulingPolicy encapsulates various scheduling policies of the distributed training job, for example `minAvailable` for gang-scheduling.

func (*SchedulingPolicy) DeepCopy 

func (in *SchedulingPolicy) DeepCopy() *SchedulingPolicy

DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new SchedulingPolicy.

func (*SchedulingPolicy) DeepCopyInto 

func (in *SchedulingPolicy) DeepCopyInto(out *SchedulingPolicy)

DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.

type SuccessPolicy 

type SuccessPolicy string

SuccessPolicy is the policy to mark the job as succeeded. 

const (
	// SuccessPolicyDefault indicates the job is succeeded by default.
	SuccessPolicyDefault SuccessPolicy = ""
	// SuccessPolicyAllWorkers indicates the job is succeeded if all workers are succeeded.
	SuccessPolicyAllWorkers SuccessPolicy = "AllWorkers"
)

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.