Importing models where the artifacts location is in s3

[richardyc]

Hi team, I met an exception when importing an exported model, where the target mlflow deployment uses s3 to store artifacts. The model got uploaded to s3 successfully, but failed at "import_version" function path check:

    def _import_version(self, model_name, src_vr, dst_run_id, dst_source, sleep_time):
        """
        :param model_name: Model name.
        :param src_vr: Source model version.
        :param dst_run: Destination run.
        :param dst_source: Destination version 'source' field.
        :param sleep_time: Seconds to wait for model version crreation.
        """
        dst_source = dst_source.replace("file://","") # OSS MLflow
        if not dst_source.startswith("dbfs:") and not os.path.exists(dst_source):
            raise MlflowExportImportException(f"'source' argument for MLflowClient.create_model_version does not exist: {dst_source}")

Raising an exception 'source' argument for MLflowClient.create_model_version does not exist: s3:...... I commented out the check and the model version got imported successfully. Shall we add s3 paths to supported dst_source? Thanks all

[TS9001]

I have the completely same issue... It would be fantastic to support s3, lots of people (and especially companies) are using that.

[richardyc]

Just pushed a quick fix, can I get a review? Thank you!