Problem when launching a child run on AzureML

Question

Got a error when trying to launch child runs from a pipeline step.

In pipeline step I have this:

import argparse
from azureml.core import Run, ScriptRunConfig
from azureml.core.model import Model
from azureml.core.environment import Environment

run = Run.get_context()
compute_target = run.experiment.workspace.compute_targets[run.get_details()['target']]

parser = argparse.ArgumentParser("args")
parser.add_argument("--train_path", type=str, help="train_path")
parser.add_argument("--test_path", type=str, help="test_path")
args = parser.parse_args()

child_config_xgboost = ScriptRunConfig(
    source_directory=".",
    script='xgboost_model.py',
    arguments=['--train_path', args.train_path, "--test_path", args.test_path],
    compute_target=compute_target,
    environment=run.get_environment()
)

child_xgboost=run.submit_child(child_config_xgboost)
child_xgboost.wait_for_completion()

I receive this error:

Traceback (most recent call last):
  File "train.py", line 7, in <module>
    compute_target = run.experiment.workspace.compute_targets[run.get_details()['target']]
  File "/azureml-envs/azureml_912ebce86aa851a4f789bc2c01e320a3/lib/python3.6/site-packages/azureml/core/workspace.py", line 1009, in compute_targets
    compute_target.name: compute_target for compute_target in ComputeTarget.list(self)}
  File "/azureml-envs/azureml_912ebce86aa851a4f789bc2c01e320a3/lib/python3.6/site-packages/azureml/core/compute/compute.py", line 535, in list
    env_obj = child.deserialize(workspace, env)
  File "/azureml-envs/azureml_912ebce86aa851a4f789bc2c01e320a3/lib/python3.6/site-packages/azureml/core/compute/computeinstance.py", line 638, in deserialize
    target._initialize(workspace, object_dict)
  File "/azureml-envs/azureml_912ebce86aa851a4f789bc2c01e320a3/lib/python3.6/site-packages/azureml/core/compute/computeinstance.py", line 130, in _initialize
    status.created_by_user_org)
  File "/azureml-envs/azureml_912ebce86aa851a4f789bc2c01e320a3/lib/python3.6/site-packages/azureml/core/compute/computeinstance.py", line 947, in _get_user_display_name
    headers = {"Authorization": "Bearer " + workspace._auth._get_graph_token()}
  File "/azureml-envs/azureml_912ebce86aa851a4f789bc2c01e320a3/lib/python3.6/site-packages/azureml/core/authentication.py", line 1455, in _get_graph_token
    raise AuthenticationException("AzureMLTokenAuthentication._get_graph_token "
UserScriptException: UserScriptException:
 Message: AzureMLTokenAuthentication._get_graph_token not yet supported.
 InnerException AuthenticationException:
 Message: AzureMLTokenAuthentication._get_graph_token not yet supported.
 InnerException None
 ErrorResponse 
{
    "error": {
        "code": "UserError",
        "inner_error": {
            "code": "Authentication"
        },
        "message": "AzureMLTokenAuthentication._get_graph_token not yet supported."
    }
}
 ErrorResponse 
{
    "error": {
        "code": "UserError",
        "message": "AzureMLTokenAuthentication._get_graph_token not yet supported."
    }
}

How can I launch a child run using a ScriptRunConfig with the same environment and compute target as the parent run?

Answer 1

@MARCO PINHEIRO Thanks, Find more details here for secret and for submit child..

from azureml.core import Experiment, Run  
  
run = Run.get_context()  
secret_value = run.get_secret(name="mysecret")  

Here is the example using ScriptRunConfig(), you can access the key vault when you use ScriptRunConfig().

Also, Here are the details to create an environment and configure and run training runs.