My co-workers are using pre-build docker images for our developing environment in Azure Machine Learning Service.
In a separate script, they have registered these environments with the command myenv.register(workspace=ws). In another script, I should use their environment for testing our model.
In order to get one of their environments, I use the command registered_env = Environment.get(ws, 'the-specific-environment-name')
Unfortunately, this does not work when I use registered_env for the experiment. I get the error "Authentication failed for container registry name_of_their_container_registry.azurecr.io". The experiment run works perfectly when I copy their environment definition code into my script instead of using the command registered_env = Environment.get(ws, 'the-specific-environment-name').
However, I cannot copy everytime their environment definition code into my script.
How can I get the environment into my script which has been defined in another script?
This StackOverFlow post is quite related to my problem:
https://stackoverflow.com/questions/71131403/registering-and-getting-an-environment-in-azure-machine-learning-studio-that-der
To illustrate what my problem is, here are some code samples.
This code sample is working:
registry = ContainerRegistry()
registry.address = <DockerRegistryAddress>
registry.username = <UserName>
registry.password = <Password>
exemplarily_env_docker_image = Environment.from_docker_image('exemplarily-env_Docker-image-AzureRegistry', <DockerImageAddress>, container_registry=registry, conda_specification=None, pip_requirements=None)
exemplarily_env_docker_image.python.user_managed_dependencies = True
# Registering and getting of an environment that derives from a Docker Image is not working because the credentials are not saved
exemplarily_env_docker_image.register(workspace=ws)
model = Model(ws, 'exemplarily_model')
inference_config = InferenceConfig(environment=exemplarily_env_docker_image,
source_directory='./source_dir',
entry_script='./score.py')
deployment_config = LocalWebservice.deploy_configuration(port=6789)
service = Model.deploy(
ws,
"myservice",
[model],
inference_config,
deployment_config,
overwrite=True,
)
service.wait_for_deployment(show_output=True)
print(service.get_logs())
Now, I do a small change and the code sample is not working anymore:
registry = ContainerRegistry()
registry.address = <DockerRegistryAddress>
registry.username = <UserName>
registry.password = <Password>
exemplarily_env_docker_image = Environment.from_docker_image('exemplarily-env_Docker-image-AzureRegistry', <DockerImageAddress>, container_registry=registry, conda_specification=None, pip_requirements=None)
exemplarily_env_docker_image.python.user_managed_dependencies = True
# Registering and getting of an environment that derives from a Docker Image is not working because the credentials are not saved
exemplarily_env_docker_image.register(workspace=ws)
model = Model(ws, 'exemplarily_model')
reg_env = Environment.get(ws, "exemplarily-env_Docker-image-AzureRegistry")
inference_config = InferenceConfig(environment=reg_env,
source_directory='./source_dir',
entry_script='./score.py')
deployment_config = LocalWebservice.deploy_configuration(port=6789)
service = Model.deploy(
ws,
"myservice",
[model],
inference_config,
deployment_config,
overwrite=True,
)
service.wait_for_deployment(show_output=True)
print(service.get_logs())
What is working:
registry = ContainerRegistry()
registry.address = <DockerRegistryAddress>
registry.username = <UserName>
registry.password = <Password>
exemplarily_env_docker_image = Environment.from_docker_image('exemplarily-env_Docker-image-AzureRegistry', <DockerImageAddress>, container_registry=registry, conda_specification=None, pip_requirements=None)
exemplarily_env_docker_image.python.user_managed_dependencies = True
# Registering and getting of an environment that derives from a Docker Image is not working because the credentials are not saved
exemplarily_env_docker_image.save_to_directory(path="./env", overwrite=True)
model = Model(ws, 'exemplarily_model')
reg_env = Environment.load_from_directory(path="./env")
inference_config = InferenceConfig(environment=reg_env,
source_directory='./source_dir',
entry_script='./score.py')
deployment_config = LocalWebservice.deploy_configuration(port=6789)
service = Model.deploy(
ws,
"myservice",
[model],
inference_config,
deployment_config,
overwrite=True,
)
service.wait_for_deployment(show_output=True)
print(service.get_logs())
Why is the middle code sample not working? Is this a bug?
