mpioperatortracker

package module
v0.0.0-...-29fc9a6 Latest Latest
Warning

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

 Go to latest Published: Jul 18, 2022 License: BSD-2-Clause Imports: 23 Imported by: 1

README

mpioperatortracker

Experimental MPIOperator support for DRMAA2OS. This project evaluates the usefulness of DRMAA2 for managing MPIOperator jobs running on Kubernetes.

How MPIOperator Works

MPIOperator provides a framework / Kubernetes operator for running MPI jobs based on IntelMPI or OpenMPI on Kubernetes. For that it implements a custom resource definition (CRD) for the kind MPIJob. Submitting jobs require writing these specific yaml files.

What mpioperatortracker is

It is a DRMAA2 implementation of the jobtracker interface which allows to hook MPIOpertor jobs into the DRMAA2 Go framework so that they can be submitted, supervised, and managed from the well defined DRMAA2 interfaces. The DRMAA2 JobTemplate is used for submitting MPIOperator jobs. The DRMAA2 JobInfo struct is used for getting the status of a job.

How to use it

Converting a DRMAA2 Job Template to an MPIOperator Job

JobInfo Fields

Job Control Mapping

Job State Mapping

Examples

Documentation

Index

Constants

View Source 
const ExtensionMPIImplementation = "mpiImplementation"
View Source 
const ExtensionRunAsUser = "runAsUser"
View Source 
const ExtensionSSHMountPath = "sshMountPath"
View Source 
const ExtensionSlotsPerWorker = "slotsPerWorker"
View Source 
const ExtensionWorkerCommand = "workerCommand"
View Source 
const ExtensionWorkerImage = "workerImage"

Variables

This section is empty.

Functions

func ConvertJobTemplateToMPIJob 

func ConvertJobTemplateToMPIJob(jt drmaa2interface.JobTemplate) (kubeflow.MPIJobSpec, error)

func CreateJob 

func CreateJob(ctx context.Context, mpiClient clientset.Interface, mpiJob *kubeflow.MPIJob, waitForJob bool) (*kubeflow.MPIJob, error)

func DeleteJob 

func DeleteJob(ctx context.Context, mpiClient clientset.Interface, namespace, jobName string) error

func DescribeJob 

func DescribeJob(ctx context.Context, mpiClient clientset.Interface, namespace, jobName string) (*kubeflow.MPIJob, error)

func GetClient 

func GetClient(restConfig *rest.Config) (clientset.Interface, error)

func GetJobInfo 

func GetJobInfo(ctx context.Context, mpiClient clientset.Interface, namespace, jobName string) (drmaa2interface.JobInfo, error)

func GetJobState 

func GetJobState(ctx context.Context, mpiClient clientset.Interface, namespace, jobName string) (drmaa2interface.JobState, string, error)

func GetLauncherResourceLimitExtension 

func GetLauncherResourceLimitExtension(jt drmaa2interface.JobTemplate) v1.ResourceList

func GetLauncherResourceRequestExtension 

func GetLauncherResourceRequestExtension(jt drmaa2interface.JobTemplate) v1.ResourceList

func GetMPIImplementationExtension 

func GetMPIImplementationExtension(jt drmaa2interface.JobTemplate) kubeflow.MPIImplementation

func GetRunAsUserExtension 

func GetRunAsUserExtension(jt drmaa2interface.JobTemplate) int64

func GetSSHMountPathExtension 

func GetSSHMountPathExtension(jt drmaa2interface.JobTemplate) string

func GetSlotsPerWorkerExtension 

func GetSlotsPerWorkerExtension(jt drmaa2interface.JobTemplate) int32

func GetWorkerCommandExtension 

func GetWorkerCommandExtension(jt drmaa2interface.JobTemplate) []string

func GetWorkerImageExtension 

func GetWorkerImageExtension(jt drmaa2interface.JobTemplate) string

func GetWorkerResourceLimitExtension 

func GetWorkerResourceLimitExtension(jt drmaa2interface.JobTemplate) v1.ResourceList

func GetWorkerResourceRequestExtension 

func GetWorkerResourceRequestExtension(jt drmaa2interface.JobTemplate) v1.ResourceList

func InstallMPIOperator 

func InstallMPIOperator(kubeconfigPath string) error

func JobInfoFromMPIJob 

func JobInfoFromMPIJob(mpiJob *kubeflow.MPIJob) (jobInfo drmaa2interface.JobInfo)

func JobStateFromCondition 

func JobStateFromCondition(lastCondition common.JobCondition) (drmaa2interface.JobState, string, error)

func ListJobs 

func ListJobs(ctx context.Context, mpiClient clientset.Interface, namespace string) ([]kubeflow.MPIJob, error)

func NewMPIJob 

func NewMPIJob(spec kubeflow.MPIJobSpec) (job kubeflow.MPIJob)

func NewRestConfig 

func NewRestConfig(kubeConfigPath string) (*rest.Config, error)

func SetLauncherResourceLimitExtension 

func SetLauncherResourceLimitExtension(jt drmaa2interface.JobTemplate, limits v1.ResourceList) drmaa2interface.JobTemplate

func SetLauncherResourceRequestsExtension 

func SetLauncherResourceRequestsExtension(jt drmaa2interface.JobTemplate, requests v1.ResourceList) drmaa2interface.JobTemplate

func SetMPIImplementationExtension 

func SetMPIImplementationExtension(jt drmaa2interface.JobTemplate, mpiImplementation string) drmaa2interface.JobTemplate

func SetRunAsUserExtension 

func SetRunAsUserExtension(jt drmaa2interface.JobTemplate, runAsUser int64) drmaa2interface.JobTemplate

func SetSSHMountPathExtension 

func SetSSHMountPathExtension(jt drmaa2interface.JobTemplate, sshMountPath string) drmaa2interface.JobTemplate

func SetSlotsPerWorkerExtension 

func SetSlotsPerWorkerExtension(jt drmaa2interface.JobTemplate, slotsPerWorker int32) drmaa2interface.JobTemplate

func SetVolumeMounts 

func SetVolumeMounts(jt drmaa2interface.JobTemplate, vm []VolumeMountSpec) drmaa2interface.JobTemplate

func SetWorkerCommandExtension 

func SetWorkerCommandExtension(jt drmaa2interface.JobTemplate, workerCommand ...string) drmaa2interface.JobTemplate

func SetWorkerImageExtension 

func SetWorkerImageExtension(jt drmaa2interface.JobTemplate, workerImage string) drmaa2interface.JobTemplate

func SetWorkerResourceLimitExtension 

func SetWorkerResourceLimitExtension(jt drmaa2interface.JobTemplate, limits v1.ResourceList) drmaa2interface.JobTemplate

func SetWorkerResourceRequestsExtension 

func SetWorkerResourceRequestsExtension(jt drmaa2interface.JobTemplate, requests v1.ResourceList) drmaa2interface.JobTemplate

Types

type MPIOperatorTracker 

type MPIOperatorTracker struct {
	// contains filtered or unexported fields
}

func NewMPIOperatorTracker 

func NewMPIOperatorTracker(kubeconfigPath string, testInstallMPIOperator bool) (*MPIOperatorTracker, error)

func (*MPIOperatorTracker) AddArrayJob 

func (t *MPIOperatorTracker) AddArrayJob(jt drmaa2interface.JobTemplate, begin int, end int, step int, maxParallel int) (string, error)

AddArrayJob makes a mass submission of jobs defined by the same job template. Many HPC workload manager support job arrays for submitting 10s of thousands of similar jobs by one call. The additional parameters define how many jobs are submitted by defining a TASK_ID range. Begin is the first task ID (like 1), end is the last task ID (like 10), step is a positive integeger which defines the increments from one task ID to the next task ID (like 1). maxParallel is an arguments representating an optional functionality which instructs the backend to limit maxParallel tasks of this job arary to run in parallel. Note, that jobs use the TASK_ID environment variable to identifiy which task they are and determine that way what to do (like which data set is accessed).

func (*MPIOperatorTracker) AddJob 

func (t *MPIOperatorTracker) AddJob(jobTemplate drmaa2interface.JobTemplate) (string, error)

AddJob typically submits or starts a new job at the backend. The function returns the unique job ID or an error if job submission (or starting of the job in case there is no queueing system) has failed.

func (*MPIOperatorTracker) DeleteJob 

func (t *MPIOperatorTracker) DeleteJob(jobID string) error

DeleteJob removes a job from a potential internal database. It does not stop a job. A job must be in an endstate (terminated, failed) in order to call DeleteJob. In case of an error or the job is not in an end state error must be returned. If the backend does not support cleaning up resources for a finished job nil should be returned.

func (*MPIOperatorTracker) JobControl 

func (t *MPIOperatorTracker) JobControl(jobID string, action string) error

JobControl sends a request to the backend to either "terminate", "suspend", "resume", "hold", or "release" a job. The strings are fixed and are defined by the JobControl constants. This could change in the future to be limited only to constants representing the actions. When the request is not accepted by the system the function must return an error.

func (*MPIOperatorTracker) JobInfo 

func (t *MPIOperatorTracker) JobInfo(jobID string) (drmaa2interface.JobInfo, error)

JobInfo returns the job status of a job in form of a JobInfo struct or an error.

func (*MPIOperatorTracker) JobState 

func (t *MPIOperatorTracker) JobState(jobID string) (drmaa2interface.JobState, string, error)

JobState returns the DRMAA2 state and substate (free form string) of the job.

func (*MPIOperatorTracker) ListArrayJobs 

func (t *MPIOperatorTracker) ListArrayJobs(arrayjobID string) ([]string, error)

ListArrayJobs returns all job IDs an job array ID (or array job ID) represents or an error.

func (MPIOperatorTracker) ListJobCategories 

func (MPIOperatorTracker) ListJobCategories() ([]string, error)

ListJobCategories returns a list of job categories which can be used in the JobCategory field of the job template. The list is informational. An example is returning a list of supported container images. AddJob() and AddArrayJob() processes a JobTemplate and hence also the JobCategory field.

func (*MPIOperatorTracker) ListJobs 

func (t *MPIOperatorTracker) ListJobs() ([]string, error)

ListJobs returns all visible job IDs or an error.

func (*MPIOperatorTracker) Wait 

func (t *MPIOperatorTracker) Wait(jobID string, timeout time.Duration, states ...drmaa2interface.JobState) error

Wait blocks until the job is either in one of the given states, the max. waiting time (specified by timeout) is reached or an other internal error occured (like job was not found). In case of a timeout also an error must be returned.

type VolumeMountSpec 

type VolumeMountSpec struct {
	MountPath  string // path to mount volume inside the container
	ReadOnly   bool
	VolumeType string // like pvc, or cm/configmap
	VolumeName string // like pvc-name or configmap-name
}

func GetVolumeMounts 

func GetVolumeMounts(jt drmaa2interface.JobTemplate) []VolumeMountSpec

Source Files

View all Source files

Directories

Path Synopsis
examples
pi Module

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.