kubeflow.fairing.deployers.job package

Submodules

kubeflow.fairing.deployers.job.job module

class kubeflow.fairing.deployers.job.job.Job(namespace=None, runs=1, output=None, cleanup=True, labels=None, job_name='fairing-job-', stream_log=True, deployer_type='job', pod_spec_mutators=None, annotations=None)

Bases: kubeflow.fairing.deployers.deployer.DeployerInterface

Handle all the k8s’ template building for a training

create_resource()

create job

deploy(pod_spec)

deploy the training job using k8s client lib

Parameters:pod_spec – pod spec of deployed training job

do_cleanup()

clean up the pods after job finished

generate_deployment_spec(pod_template_spec)

Generate a V1Job initialized with correct completion and

parallelism (for HP search) and with the provided V1PodTemplateSpec

Parameters:pod_template_spec – V1PodTemplateSpec

generate_pod_template_spec(pod_spec)

Generate a V1PodTemplateSpec initiazlied with correct metadata

and with the provided pod_spec

Parameters:pod_spec – pod spec

get_logs()

get logs from the deployed job

set_anotations(annotations)

set_labels(labels, deployer_type)

set labels for the pods of a deployed job

Parameters:

• labels – dictionary of labels {label_name:label_value}
• deployer_type – deployer type name

Module contents