marvinbuss / aml-docker Goto Github PK

Hello 👋
I've been using the 1.25 version of this image with great success for some time now, but recently I've come across an error that's bugging me and I can't really pinpoint its root cause.
Basically, what I'm trying to do is to have a custom GitHub Action that would push a CSV file from the git repository into an Azure ML workspace and register it as a new dataset. I've tested several versions of the code, including:

• Registering the dataset from pandas dataframe

df = pd.read_csv('data/train.csv')
datastore = workspace.get_default_datastore()
dataset = Dataset.Tabular.register_pandas_dataframe(
    df, (datastore, "data/train.csv"), "Dataset")

• Manually uploading the file and then loading it from a datastore

datastore = workspace.get_default_datastore()
datastore.upload(src_dir="data", target_path="data", overwrite=True)
dataset = Dataset.Tabular.from_delimited_files(datastore.path("data/train.csv"), validate=False)

In both cases, I get the following exception:

Error Message: ScriptExecutionException was caused by WriteStreamsException.
  WriteStreamsException was caused by UnexpectedException.
    Unexpected exception while writing files with writer 'parquet'.
      DatastoreResolutionException was caused by UnexpectedException.
        Unexpected failure making request to fetching info for Datastore '<my-datastore-name>' in subscription: '<my-subscription-id>', resource group: '<my-resource-group>', workspace: '<my-workspace>'. Using base service url: https://westeurope.experiments.azureml.net. HResult: 0x80131501.
          The SSL connection could not be established, see inner exception.

This occurs in register_pandas_dataframe() line in the first example and in the from_delimited_files() line in the second one. Interestingly, the data is successfully uploaded to the datastore in the second example, what fails is the second step of getting a reference to it from the datastore. This seems strange since the error message indicates a connectivity problem, but the data was uploaded correctly.

Initially, I thought this is a GitHub-related issue, but I run the same container on my local machine and the results were the same. Running the same code locally, in a Jupyter notebook, works fine, the data is pushed and the new dataset is available and correctly referenced in the code. This leads me to believe that the issue is related to the container itself. I've spent a fair amount of time trying to debug it but I haven't reached anything conclusive.
Could anyone help me with this one? Thanks a lot.