def _Run(self, args, version):
index_ref = args.CONCEPTS.index.Parse()
region = index_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version, region=region):
operation = client.IndexesClient().PatchBeta(index_ref, args)
if args.metadata_file is not None:
# Update index content.
op_ref = indexes_util.ParseIndexOperation(operation.name)
log.status.Print(
constants.OPERATION_CREATION_DISPLAY_MESSAGE.format(
name=operation.name,
verb='update index',
id=op_ref.Name(),
sub_commands='--index={}'.format(index_ref.Name())))
# We will not wait for the operation since it can take up to hours.
return operation
# Update index meta data.
response_msg = operations_util.WaitForOpMaybe(
operations_client=operations.OperationsClient(),
op=operation,
op_ref=indexes_util.ParseIndexOperation(operation.name),
log_method='update')
if response_msg is not None:
response = encoding.MessageToPyValue(response_msg)
if 'name' in response:
log.UpdatedResource(response['name'],
kind='AI Platform index')
return response_msg
def Run(self, args):
region_ref = args.CONCEPTS.region.Parse()
region = region_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(
version=constants.BETA_VERSION, region=region):
return client.HpTuningJobsClient().List(
region=region_ref.RelativeName())
def _Run(self, args, version):
validation.ValidateDisplayName(args.display_name)
index_endpoint_ref = args.CONCEPTS.index_endpoint.Parse()
project_id = index_endpoint_ref.AsDict()['projectsId']
region = index_endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version, region=region):
index_endpoint_client = client.IndexEndpointsClient(
version=version)
if version == constants.GA_VERSION:
operation = index_endpoint_client.DeployIndex(
index_endpoint_ref, args)
else:
operation = index_endpoint_client.DeployIndexBeta(
index_endpoint_ref, args)
op_ref = index_endpoints_util.ParseIndexEndpointOperation(
operation.name)
# TODO(b/208506223): Support `--async` flag.
index_endpoint_id = op_ref.AsDict()['indexEndpointsId']
log.status.Print(
constants.OPERATION_CREATION_DISPLAY_MESSAGE.format(
name=operation.name,
verb='deploy index',
id=op_ref.Name(),
sub_commands='--index-endpoint={} [--project={}]'.format(
index_endpoint_id, project_id)))
return operation
def _Run(args, version):
"""Run Vertex AI online prediction."""
endpoint_ref = args.CONCEPTS.endpoint.Parse()
args.region = endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version,
region=args.region):
if args.request.startswith('@'):
request = console_io.ReadFromFileOrStdin(args.request[1:],
binary=True)
else:
request = args.request.encode('utf-8')
endpoints_client = client.EndpointsClient(version=version)
_, response = endpoints_client.RawPredict(endpoint_ref,
args.http_headers, request)
# Workaround since gcloud only supports protobufs as JSON objects. Since
# raw predict can return anything, write raw bytes to stdout.
if not args.IsSpecified('format'):
sys.stdout.buffer.write(response)
return None
# If user asked for formatting, assume it's a JSON object.
try:
return json.loads(response.decode('utf-8'))
except ValueError:
raise core_exceptions.Error(
'No JSON object could be decoded from the '
'HTTP response body:\n' + six.text_type(response))
def _Run(args, version):
"""Run method for update command."""
validation.ValidateDisplayName(args.display_name)
model_monitoring_job_ref = args.CONCEPTS.monitoring_job.Parse()
region = model_monitoring_job_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version, region=region):
try:
result = client.ModelMonitoringJobsClient(version=version).Patch(
model_monitoring_job_ref, args)
except errors.NoFieldsSpecifiedError:
available_update_args = [
'display_name',
'emails',
'prediction_sampling_rate',
'feature_thresholds',
'monitoring_config_from_file',
'monitoring_frequency',
'analysis_instance_schema',
'log_ttl',
'update_labels',
'clear_labels',
'remove_labels',
'anomaly_cloud_logging',
]
if not any(args.IsSpecified(arg) for arg in available_update_args):
raise
log.status.Print('No update to perform.')
return None
else:
log.UpdatedResource(
result.name, kind='Vertex AI model deployment monitoring job')
return result
def _Run(args, version):
"""Update an existing Vertex AI Tensorboard."""
tensorboard_ref = args.CONCEPTS.tensorboard.Parse()
args.region = tensorboard_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version, region=args.region):
tensorboards_client = client.TensorboardsClient(version=version)
try:
op = tensorboards_client.Patch(tensorboard_ref, args)
except errors.NoFieldsSpecifiedError:
available_update_args = [
'display_name',
'description',
'update_labels',
'clear_labels',
'remove_labels',
]
if not any(args.IsSpecified(arg) for arg in available_update_args):
raise
log.status.Print('No update to perform.')
return None
else:
log.UpdatedResource(op.name, kind='Vertex AI Tensorboard')
return op
def _Run(self, args):
model_ref = args.CONCEPTS.model.Parse()
region = model_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(
version=constants.BETA_VERSION, region=region):
response = client.ModelsClient().Get(model_ref)
return response
def _Run(self, args):
tensorboard_ref = args.CONCEPTS.tensorboard.Parse()
region = tensorboard_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(
version=constants.ALPHA_VERSION, region=region):
response = client.TensorboardsClient().Get(tensorboard_ref)
return response
def _Run(self, args, version):
validation.ValidateDisplayName(args.display_name)
index_endpoint_ref = args.CONCEPTS.index_endpoint.Parse()
region = index_endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version, region=region):
index_endpoint_client = client.IndexEndpointsClient(
version=version)
try:
if version == constants.GA_VERSION:
result = index_endpoint_client.Patch(
index_endpoint_ref, args)
else:
result = index_endpoint_client.PatchBeta(
index_endpoint_ref, args)
except errors.NoFieldsSpecifiedError:
available_update_args = [
'display_name', 'description', 'update_labels',
'clear_labels', 'remove_labels'
]
if not any(
args.IsSpecified(arg)
for arg in available_update_args):
raise
log.status.Print('No update to perform.')
return None
else:
log.UpdatedResource(result.name,
kind='Vertex AI index endpoint')
return result
def Run(self, args):
hptuning_job_ref = args.CONCEPTS.hptuning_job.Parse()
region = hptuning_job_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(
version=constants.BETA_VERSION, region=region):
response = client.HpTuningJobsClient().Get(
hptuning_job_ref.RelativeName())
return response
def _Run(args, version):
tensorboard_ref = args.CONCEPTS.tensorboard.Parse()
region = tensorboard_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version=version,
region=region):
response = client.TensorboardsClient(
version=version).Get(tensorboard_ref)
return response
def _Run(self, args):
region_ref = args.CONCEPTS.region.Parse()
region = region_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(
version=constants.ALPHA_VERSION, region=region):
return client.TensorboardsClient().List(limit=args.limit,
page_size=args.page_size,
region_ref=region_ref,
sort_by=args.sort_by)
def _Run(self, args):
model_ref = args.CONCEPTS.model.Parse()
region = model_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(
version=constants.BETA_VERSION, region=region):
operation = client.ModelsClient().Delete(model_ref)
return operations_util.WaitForOpMaybe(
operations_client=operations.OperationsClient(),
op=operation,
op_ref=models_util.ParseModelOperation(operation.name))
def Run(self, args):
custom_job_ref = args.CONCEPTS.custom_job.Parse()
region = custom_job_ref.AsDict()['locationsId']
validation.ValidateRegion(region)
with endpoint_util.AiplatformEndpointOverrides(
version=self._api_version, region=region):
response = client.CustomJobsClient(version=self._api_version).Get(
custom_job_ref.RelativeName())
return response
def _Run(args, version):
tensorboard_run_ref = args.CONCEPTS.tensorboard_run_id.Parse()
region = tensorboard_run_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(
version=version, region=region):
return client.TensorboardTimeSeriesClient(version=version).List(
tensorboard_run_ref=tensorboard_run_ref,
limit=args.limit,
page_size=args.page_size,
sort_by=args.sort_by)
def _Run(args, version):
tensorboard_time_series_ref = args.CONCEPTS.tensorboard_time_series.Parse()
region = tensorboard_time_series_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version=version,
region=region):
response = client.TensorboardTimeSeriesClient(version=version).Read(
tensorboard_time_series_ref=tensorboard_time_series_ref,
max_data_points=args.max_data_points,
data_filter=args.filter)
return response
def _Run(args, version):
"""Deploy a model to an existing Vertex AI endpoint."""
validation.ValidateDisplayName(args.display_name)
if version != constants.GA_VERSION:
validation.ValidateAutoscalingMetricSpecs(
args.autoscaling_metric_specs)
endpoint_ref = args.CONCEPTS.endpoint.Parse()
args.region = endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version,
region=args.region):
endpoints_client = client.EndpointsClient(version=version)
operation_client = operations.OperationsClient()
if version == constants.GA_VERSION:
op = endpoints_client.DeployModel(
endpoint_ref,
args.model,
args.region,
args.display_name,
machine_type=args.machine_type,
accelerator_dict=args.accelerator,
min_replica_count=args.min_replica_count,
max_replica_count=args.max_replica_count,
enable_access_logging=args.enable_access_logging,
disable_container_logging=args.disable_container_logging,
service_account=args.service_account,
traffic_split=args.traffic_split,
deployed_model_id=args.deployed_model_id)
else:
op = endpoints_client.DeployModelBeta(
endpoint_ref,
args.model,
args.region,
args.display_name,
machine_type=args.machine_type,
accelerator_dict=args.accelerator,
min_replica_count=args.min_replica_count,
max_replica_count=args.max_replica_count,
autoscaling_metric_specs=args.autoscaling_metric_specs,
enable_access_logging=args.enable_access_logging,
enable_container_logging=args.enable_container_logging,
service_account=args.service_account,
traffic_split=args.traffic_split,
deployed_model_id=args.deployed_model_id)
response_msg = operations_util.WaitForOpMaybe(
operation_client, op, endpoints_util.ParseOperation(op.name))
if response_msg is not None:
response = encoding.MessageToPyValue(response_msg)
if 'deployedModel' in response and 'id' in response[
'deployedModel']:
log.status.Print(('Deployed a model to the endpoint {}. '
'Id of the deployed model: {}.').format(
endpoint_ref.AsDict()['endpointsId'],
response['deployedModel']['id']))
return response_msg
def _Run(args, version):
"""Undeploy a model fro man existing AI Platform endpoint."""
endpoint_ref = args.CONCEPTS.endpoint.Parse()
args.region = endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version,
region=args.region):
endpoints_client = client.EndpointsClient(version=version)
operation_client = operations.OperationsClient()
op = endpoints_client.UndeployModelBeta(endpoint_ref, args)
return operations_util.WaitForOpMaybe(
operation_client, op, endpoints_util.ParseOperation(op.name))
def Run(self, args):
hptuning_job_ref = args.CONCEPTS.hptuning_job.Parse()
name = hptuning_job_ref.Name()
region = hptuning_job_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(
version=constants.BETA_VERSION, region=region):
response = client.HpTuningJobsClient().Cancel(
hptuning_job_ref.RelativeName())
log.status.Print(
constants.HPTUNING_JOB_CANCEL_DISPLAY_MESSAGE.format(id=name))
return response
def _Run(self, args, version):
index_endpoint_ref = args.CONCEPTS.index_endpoint.Parse()
region = index_endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version, region=region):
operation = client.IndexEndpointsClient().UndeployIndexBeta(
index_endpoint_ref, args)
return operations_util.WaitForOpMaybe(
operations_client=operations.OperationsClient(),
op=operation,
op_ref=index_endpoints_util.ParseIndexEndpointOperation(
operation.name))
def _Run(self, args):
tensorboard_ref = args.CONCEPTS.tensorboard.Parse()
region = tensorboard_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(
version=constants.ALPHA_VERSION, region=region):
operation = client.TensorboardsClient().Delete(tensorboard_ref)
return operations_util.WaitForOpMaybe(
operations_client=operations.OperationsClient(),
op=operation,
op_ref=tensorboards_util.ParseTensorboardOperation(
operation.name))
def _Run(args, version):
tensorboard_run_ref = args.CONCEPTS.tensorboard_run.Parse()
region = tensorboard_run_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version=version,
region=region):
operation = client.TensorboardRunsClient(
version=version).Delete(tensorboard_run_ref)
return operations_util.WaitForOpMaybe(
operations_client=operations.OperationsClient(),
op=operation,
op_ref=tensorboards_util.ParseTensorboardOperation(operation.name))
def _Run(args, version):
"""Create a new Vertex AI Tensorboard run."""
tensorboard_exp_ref = args.CONCEPTS.tensorboard_experiment.Parse()
region = tensorboard_exp_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version, region=region):
tensorboard_runs_client = client.TensorboardRunsClient(version=version)
response = tensorboard_runs_client.Create(tensorboard_exp_ref, args)
response_msg = encoding.MessageToPyValue(response)
if 'name' in response_msg:
log.status.Print(('Created Vertex AI Tensorboard run: {}.').format(
response_msg['name']))
return response
def Run(self, args):
hptuning_job_ref = args.CONCEPTS.hptuning_job.Parse()
region = hptuning_job_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(
version=constants.BETA_VERSION, region=region):
relative_name = hptuning_job_ref.RelativeName()
return log_util.StreamLogs(
hptuning_job_ref.AsDict()['hyperparameterTuningJobsId'],
continue_function=client.HpTuningJobsClient().CheckJobComplete(
relative_name),
polling_interval=args.polling_interval,
task_name=args.task_name,
allow_multiline=args.allow_multiline_logs)
