[solved] Deployment problem - The shell object ... can't be saved to workspace RRS feed

  • Question

  • Using 9.4.7 with Python and the DeployClient to deploy a set of "models", one of which is a custom object named tenant_env.

    If I use the Python environment from a local ML Server install then the deployment works.

    Using an alternative environment gives me the following exception:

    HttpException: (400) Reason: Bad Request HTTP response headers: {'Date': 'Thu, 05 Mar 2020 18:52:57 GMT', 'Content-Type': 'application/json', 'Server': 'Kestrel', 'Content-Length': '239', 'X-Content-Type-Options': 'nosniff', 'x-ms-request-id': 'b0c235ec-64fa-4f41-a355-27fe20158252'} HTTP response body: {"Link":"","Message":"The shell object tenant_env can't be saved to workspace: '_create_function() takes from 2 to 6 positional arguments but 7 were given'","ExceptionType":"ArgumentException"}

    Nothing appears in the web or compute note logs (is there somewhere else to look?)

    I have been able to pickle the object in one Py environment and unpickle in the other, on my local machine.

    Can anyone either explain this or point me in a direction to investigate. I am at a loss, and web searches have provided no insight!

    Cheers, and TIA, Adam

    • Edited by A Cooper Friday, March 6, 2020 12:39 PM
    Thursday, March 5, 2020 7:09 PM

All replies

  • This was finally tracked down to the dill package version incompatibility.

    using dill, as on ML Server, resolved the issue.

    Friday, March 6, 2020 12:39 PM