When used in a dataset factory, azureml's credentials are not setup correctly and creates a crash instead. For instance, the following crashes with missing credentials :

"{name}_csv":
    type: utils.azure_ml_dataset.AzureMLDataset
    azureml_dataset: raw
    root_dir: data/01_raw/
    dataset:
        type: pandas.CSVDataset
        filepath: "{name}.csv"

The issue is that when the catalog is created, the dataset factories are not actual datasets, but dataset_pattern, which are resolved afterward. Then the hook for after_catalog_created is called, but still no actual dataset for the dataset_pattern.

I've discussed with some people on kedro's slack channels, and they seem to agree that the best way to pass the credentials would be with a after_context_created hook.

I'll open a pull request soon to fix that, with the following pattern instead :

"{name}_csv":
    type: utils.azure_ml_dataset.AzureMLDataset
    azureml_dataset: raw
    root_dir: data/01_raw/
    credentials: azureml # added credentials, and implicited the credentials to azureml
    dataset:
        type: pandas.CSVDataset
        filepath: "{name}.csv"