kubeflow.fairing.kubernetes package

Submodules

kubeflow.fairing.kubernetes.manager module

class kubeflow.fairing.kubernetes.manager.KubeManager

Bases: object

Handles communication with Kubernetes’ client.

create_deployment(namespace, deployment)

Create an V1Deployment in the specified namespace.

Parameters:
  • namespace – The custom resource
  • deployment – Deployment body to create
Returns:

object: Created V1Deployments.
create_isvc(namespace, isvc)

Create the provided InferenceService in the specified namespace.

Parameters:
  • namespace – The custom resource
  • InferenceService – The InferenceService body
Returns:

object: Created InferenceService.
create_job(namespace, job)

Creates a V1Job in the specified namespace.

Parameters:
  • namespace – The resource
  • job – Job defination as kubernetes
Returns:

object: Created Job.
create_secret(namespace, secret)

Create secret in the specified namespace.

Parameters:
  • namespace – The custom resource
  • secret – The secret body
Returns:

object: Created secret.
create_tf_job(namespace, job)

Create the provided TFJob in the specified namespace. The TFJob version is defined in TF_JOB_VERSION in fairing.constants. The version TFJob need to be installed before creating the TFJob.

Parameters:
  • namespace – The custom resource
  • job – The JSON schema of the Resource to create
Returns:

object: Created TFJob.
delete_deployment(name, namespace)

Delete an existing model deployment and relinquish all resources associated.

Parameters:
  • name – The deployment name
  • namespace – The custom resource
Returns:

obje deployment.
delete_isvc(name, namespace)

Delete the provided InferenceService in the specified namespace.

Parameters:
  • name – The custom object
  • namespace – The custom resource
Returns:

object: The deleted InferenceService.
delete_job(name, namespace)

Delete the specified job and related pods.

Parameters:
  • name – The job name
  • namespace – The resource
Returns:

object: the deleted job.
delete_tf_job(name, namespace)

Delete the provided TFJob in the specified namespace.

Parameters:
  • name – The custom object
  • namespace – The custom resource
Returns:

object: The deleted TFJob.
get_service_external_endpoint(name, namespace, selectors=None)

Get the service external endpoint as http://ip_or_hostname:5000/predict.

Parameters:
  • name – The sevice name
  • namespace – The custom resource
  • selectors – A selector to restrict the list of returned objects by their labels.
  • Defaults – to everything
Returns:

str: the service external endpoint.
log(name, namespace, selectors=None, container='', follow=True)

Get log of the specified pod.

Parameters:
  • name – The pod name
  • namespace – The custom resource
  • selectors – A selector to restrict the list of returned objects by their labels.
  • Defaults – to everything
  • container – The container for which to stream logs.
  • if – there is one container in the pod
  • follow – True or False (Default value = True)
Returns:

str: logs of the specified pod.
secret_exists(name, namespace)

Check if the secret exists in the specified namespace.

Parameters:
  • name – The secret name
  • namespace – The custom resource.
Returns:

bool: True if the secret exists, otherwise return False.

kubeflow.fairing.kubernetes.utils module

kubeflow.fairing.kubernetes.utils.get_resource_mutator(cpu=None, memory=None)

The mutator for getting the resource setting for pod spec.

The useful example: https://github.com/kubeflow/fairing/blob/master/examples/train_job_api/main.ipynb

Parameters:
  • cpu – Limits and requests for CPU resources (Default value = None)
  • memory – Limits and requests for memory (Default value = None)
Returns:

object: The mutator function for setting cpu and memory in pod spec.
kubeflow.fairing.kubernetes.utils.mounting_pvc(pvc_name, pvc_mount_path='/mnt')

The function for pod_spec_mutators to mount persistent volume claim.

Parameters:
  • pvc_name – The name of persistent volume claim
  • pvc_mount_path – Path for the persistent volume claim mounts to.
Returns:

object: function for mount the pvc to pods.

Module contents