The issue: Azure/azure-sdk-for-python#31166
There is conda env with python dependencies you must have to reproduce the issue.
List of dependencies can be found in conda.lock.yml file.
To create a virtual env out of that file:
conda env update --prune --file conda.lock.yml
conda activate neptune-recspoc
You must have a valid Microsoft azure subscription and Azure machine learning workspace. There is a python jupyter notebook which should provide step-by-step guidance for the issue reproduction.
There is also might be required to provide a valid AML environment in the yaml configurations. Any should work.
Location: msft-sdk-v2-test-batch-invoke-bug/aml/pipelines/collaborative_filtering_yaml/sdk-deploy-and-test-light.ipynb
Usage: Substitute your subscription id, resource group id and workspace name in the notebook.
Execute cell by cell until you reach the last one with .invoke() function call.
While calling the last cell - take a look into your workspace batch endpoint jobs,
if it submits multiple jobs and in the end, you have an error out of API calls to the AML REST endpoint - then you reproduced the issue.
if it submits only one single job and there is no errors out of .invoke() call - the issue wasn't reproduced and it is expected behaviour.
More details can be found in the original github issue request: Azure/azure-sdk-for-python#31166