Unable to use private docker registry with latest Azure ML release

Question

Since the latest Azure ML release, we have been unable to submit any job using a private docker registry. Same jobs were working before the new release.
We configure the job as follows (all of this is automated and the code has not changed):

base_image_name = 'REDACTED.azurecr.io/lb/learning_box_azure_compute:0.1.15_1601582281'

# Set the container registry information  

myenv = Environment(name="lb")  

myenv.docker.enabled = True  
myenv.docker.base_image = base_image_name  
myenv.docker.base_image_registry.address = 'REDACTED.azurecr.io/lb/'  
myenv.docker.base_image_registry.username, myenv.docker.base_image_registry.password = get_docker_secrets()  
myenv.python.user_managed_dependencies = True  

myenv.python.interpreter_path = "/opt/miniconda/bin/python"  

Instead of successful job submission, we are instead getting:
{
"error": {
"message": "Activity Failed:\n{\n \"error\": {\n \"code\": \"UserError\",\n \"message\": \"Unable to get image details : Specified base docker image REDACTED.azurecr.io/lb/learning_box_azure_compute:0.1.15_16\",\n \"details\": []\n },\n \"correlation\": {\n \"operation\": null,\n \"request\": \"c41448d429f9c80b\"\n },\n \"environment\": \"eastus\",\n \"location\": \"eastus\",\n \"time\": \"2020-11-09T21:40:39.699533Z\",\n \"componentName\": \"execution-worker\"\n}"
}
}
The image has not changed (we tried a few different ones from prior successful jobs) and the use of the SDK has not changed.
Has anybody else encountered a similar problem since the Nov 5 upgrade (https://learn.microsoft.com/en-us/azure/machine-learning/azure-machine-learning-release-notes)?
This is a major block as we cannot proceed with any project that depend on Azure ML at this time.

Answer 1

@Fabien Campagne Thanks for the details, with fully qualified base image name you do not need to specify container registry address. container registry address itself should be just a host name.