I see the following issues:

• If there are several files under the model dir, one of them is picked up in a non-deterministic way. A registered model is a logical container for one or more files that make up your model. For example, if you have a model that's stored in multiple files, you can register them as a single model in the workspace. (https://docs.microsoft.com/en-us/azure/machine-learning/how-to-deploy-and-where#registermodel)
• I could potentially have subdirectories with "." in their names, and the glob() would match their names
• I could potentially have a model file without "." in the name.

A cleaner approach might be to use Model.get_model_path(), which directly returns the file name, is compact and can be used in multi model deployment. The downside is that it requires the model name to be hard-coded (or we could define an environment variable in the AML Environment).

I think a better approach until we enable Environments, might be to keep the original code and remove the if '.pkl' in file: clause, and add an assertion that exactly one file was found.

@algattik what about this one (just updated)? It will work in all cases for single model, and I can use my favorite name now rather than to use undocumented pkl extension:)