Use Azure ML Environment from Share ML Registry with Python SDK

Question

We have created an environment in azure ML shared registry. The environment was built into the registry. Here is what it looks like:

A DevOps pipeline creates this environment:

az ml environment create --name "$azureEnvPrefix" --build-context $dockerContext --dockerfile-path Dockerfile --registry-name $REGISTRY_NAME --tag "dev-$timestamp producer-CI"
    

The environment seems fine. But, when we use the same environment in ml pipeline using:

try:
    cred = ManagedIdentityCredential()
    ml_client_registry = MLClient(credential=cred,
                        registry_name=e.registry_name,
                        registry_location=e.registry_region)
    registyr_envs = [env for env in ml_client_registry.environments.list(e.monitoring)]
except Exception as e:
    print("Unable to get environment")
    raise e
environment_monitoring = ml_client_registry.environments.get(name=monitoring, version=registyr_envs[0].version)

It works but using this environment with runconfig for a PythonScriptStep results in error:

run_config_monitoring = RunConfiguration()
run_config_monitoring.environment = environment_monitoring

param_actuals_feature_name = PipelineParameter(
    name='actuals_feature_name', default_value='payment_amt_acc')

StepMonitoring = PythonScriptStep(
    script_name=e.script_path,
    arguments=["--actuals_feature_name", param_actuals_feature_name],
    source_directory=e.sources_directory,
    name='Monitoring',
    compute_target=aml_compute_cpu,
    allow_reuse=False,
    runconfig=run_config_monitoring
)

step_sequence = StepSequence(steps=[StepMonitoring])
pipeline = Pipeline(workspace=ws, steps=step_sequence)

It results in error:

Traceback (most recent call last): File "/anaconda/envs/localpy38/lib/python3.8/runpy.py", line 194, in _run_module_as_main return _run_code(code, main_globals, None, File "/anaconda/envs/localpy38/lib/python3.8/runpy.py", line 87, in _run_code exec(code, run_globals) File "/mnt/batch/tasks/shared/LS_root/mounts/clusters/ourrehmansweden/code/Users/ourrehman/Sweden_cashflow_forecasting_aml/ml_service/pipelines/build_pipeline_active_monitoring.py", line 157, in main(args) File "/mnt/batch/tasks/shared/LS_root/mounts/clusters/ourrehmansweden/code/Users/ourrehman/Sweden_cashflow_forecasting_aml/ml_service/pipelines/build_pipeline_active_monitoring.py", line 70, in main environment_active_monitoring = ml_client_registry.environments.get(name=e.aml_env_name_active_monitoring, version=registyr_envs[0].version) File "/anaconda/envs/localpy38/lib/python3.8/site-packages/azureml/core/_experiment_method.py", line 104, in wrapper return init_func(self, *args, **kwargs) File "/anaconda/envs/localpy38/lib/python3.8/site-packages/azureml/pipeline/core/pipeline.py", line 180, in init self._graph = self._graph_builder.build(self._name, steps, finalize=False) File "/anaconda/envs/localpy38/lib/python3.8/site-packages/azureml/pipeline/core/builder.py", line 1497, in build graph = self.construct(name, steps) File "/anaconda/envs/localpy38/lib/python3.8/site-packages/azureml/pipeline/core/builder.py", line 1519, in construct self.process_collection(steps) File "/anaconda/envs/localpy38/lib/python3.8/site-packages/azureml/pipeline/core/builder.py", line 1555, in process_collection builder.process_collection(collection) File "/anaconda/envs/localpy38/lib/python3.8/site-packages/azureml/pipeline/core/builder.py", line 1809, in process_collection nodes = self._base_builder.process_collection(item) File "/anaconda/envs/localpy38/lib/python3.8/site-packages/azureml/pipeline/core/builder.py", line 1549, in process_collection return self.process_step(collection) File "/anaconda/envs/localpy38/lib/python3.8/site-packages/azureml/pipeline/core/builder.py", line 1593, in process_step node = step.create_node(self._graph, self._default_datastore, self._context) File "/anaconda/envs/localpy38/lib/python3.8/site-packages/azureml/pipeline/steps/python_script_step.py", line 243, in create_node return super(PythonScriptStep, self).create_node( File "/anaconda/envs/localpy38/lib/python3.8/site-packages/azureml/pipeline/core/_python_script_step_base.py", line 156, in create_node module_builder=module_builder, runconfig=repr(self._runconfig)) File "/anaconda/envs/localpy38/lib/python3.8/site-packages/azureml/core/runconfig.py", line 1352, in repr run_config_dict = _serialize_to_dict(self) File "/anaconda/envs/localpy38/lib/python3.8/site-packages/azureml/core/_serialization_utils.py", line 44, in _serialize_to_dict result[name] = field_info.field_type._serialize_to_dict( File "/anaconda/envs/localpy38/lib/python3.8/site-packages/azureml/core/environment.py", line 1685, in _serialize_to_dict if environment.python.conda_dependencies is not None: AttributeError: 'Environment' object has no attribute 'python'

I see that the code which complains is azureml/core/environment.py", line 1685:

if environment is not None and not isinstance(environment, EnvironmentReference):
    if environment.python.conda_dependencies is not None:
        inline = environment.python.conda_dependencies._conda_dependencies
        environment_dict["python"]["condaDependencies"] = inline

Is there something that I am missing, or we can't use environments like this? Would be thankful if anyone could help!