breaking: preserve script path when S3 source_dir is provided (#941)

Author: laurenyu

src/sagemaker/chainer/estimator.py

@@ -214,7 +214,7 @@ def create_model(
         return ChainerModel(
             self.model_data,
             role or self.role,
-            entry_point or self.entry_point,
+            entry_point or self._model_entry_point(),
             source_dir=(source_dir or self._model_source_dir()),
             enable_cloudwatch_metrics=self.enable_cloudwatch_metrics,
             container_log_level=self.container_log_level,

src/sagemaker/estimator.py

@@ -1734,17 +1734,28 @@ def _stage_user_code_in_s3(self):
         )
 
     def _model_source_dir(self):
-        """Get the appropriate value to pass as source_dir to model constructor
-        on deploying
+        """Get the appropriate value to pass as ``source_dir`` to a model constructor.
 
         Returns:
-            str: Either a local or an S3 path pointing to the source_dir to be
-            used for code by the model to be deployed
+            str: Either a local or an S3 path pointing to the ``source_dir`` to be
+                used for code by the model to be deployed
         """
         return (
             self.source_dir if self.sagemaker_session.local_mode else self.uploaded_code.s3_prefix
         )
 
+    def _model_entry_point(self):
+        """Get the appropriate value to pass as ``entry_point`` to a model constructor.
+
+        Returns:
+            str: The path to the entry point script. This can be either an absolute path or
+                a path relative to ``self._model_source_dir()``.
+        """
+        if self.sagemaker_session.local_mode or (self._model_source_dir() is None):
+            return self.entry_point
+
+        return self.uploaded_code.script_name
+
     def hyperparameters(self):
         """Return the hyperparameters as a dictionary to use for training.
 

src/sagemaker/fw_utils.py

@@ -447,7 +447,7 @@ def tar_and_upload_dir(
             script name.
     """
     if directory and directory.lower().startswith("s3://"):
-        return UploadedCode(s3_prefix=directory, script_name=os.path.basename(script))
+        return UploadedCode(s3_prefix=directory, script_name=script)
 
     script_name = script if directory else os.path.basename(script)
     dependencies = dependencies or []

src/sagemaker/mxnet/estimator.py

@@ -218,10 +218,10 @@ def create_model(
 
         kwargs["name"] = self._get_or_create_name(kwargs.get("name"))
 
-        return MXNetModel(
+        model = MXNetModel(
             self.model_data,
             role or self.role,
-            entry_point or self.entry_point,
+            entry_point,
             framework_version=self.framework_version,
             py_version=self.py_version,
             source_dir=(source_dir or self._model_source_dir()),
@@ -235,6 +235,13 @@ def create_model(
             **kwargs
         )
 
+        if entry_point is None:
+            model.entry_point = (
+                self.entry_point if model._is_mms_version() else self._model_entry_point()
+            )
+
+        return model
+
     @classmethod
     def _prepare_init_params_from_job_description(cls, job_details, model_channel_name=None):
         """Convert the job description to init params that can be handled by the

src/sagemaker/pytorch/estimator.py

@@ -175,7 +175,7 @@ def create_model(
         return PyTorchModel(
             self.model_data,
             role or self.role,
-            entry_point or self.entry_point,
+            entry_point or self._model_entry_point(),
             framework_version=self.framework_version,
             py_version=self.py_version,
             source_dir=(source_dir or self._model_source_dir()),

src/sagemaker/rl/estimator.py

@@ -232,7 +232,7 @@ def create_model(
         if not entry_point and (source_dir or dependencies):
             raise AttributeError("Please provide an `entry_point`.")
 
-        entry_point = entry_point or self.entry_point
+        entry_point = entry_point or self._model_entry_point()
         source_dir = source_dir or self._model_source_dir()
         dependencies = dependencies or self.dependencies
 

src/sagemaker/sklearn/estimator.py

@@ -196,7 +196,7 @@ def create_model(
         return SKLearnModel(
             self.model_data,
             role,
-            entry_point or self.entry_point,
+            entry_point or self._model_entry_point(),
             source_dir=(source_dir or self._model_source_dir()),
             enable_cloudwatch_metrics=self.enable_cloudwatch_metrics,
             container_log_level=self.container_log_level,

src/sagemaker/xgboost/estimator.py

@@ -172,7 +172,7 @@ def create_model(
         return XGBoostModel(
             self.model_data,
             role,
-            entry_point or self.entry_point,
+            entry_point or self._model_entry_point(),
             framework_version=self.framework_version,
             source_dir=(source_dir or self._model_source_dir()),
             enable_cloudwatch_metrics=self.enable_cloudwatch_metrics,

tests/data/mxnet_mnist/sourcedir.tar.gz

@@ -32,11 +32,16 @@ def mxnet_training_job(
     sagemaker_session, mxnet_full_version, mxnet_full_py_version, cpu_instance_type
 ):
     with timeout(minutes=TRAINING_DEFAULT_TIMEOUT_MINUTES):
-        script_path = os.path.join(DATA_DIR, "mxnet_mnist", "mnist.py")
+        s3_prefix = "integ-test-data/mxnet_mnist"
         data_path = os.path.join(DATA_DIR, "mxnet_mnist")
 
+        s3_source = sagemaker_session.upload_data(
+            path=os.path.join(data_path, "sourcedir.tar.gz"), key_prefix="{}/src".format(s3_prefix)
+        )
+
         mx = MXNet(
-            entry_point=script_path,
+            entry_point="mxnet_mnist/mnist.py",
+            source_dir=s3_source,
             role="SageMakerRole",
             framework_version=mxnet_full_version,
             py_version=mxnet_full_py_version,
@@ -46,10 +51,10 @@ def mxnet_training_job(
         )
 
         train_input = mx.sagemaker_session.upload_data(
-            path=os.path.join(data_path, "train"), key_prefix="integ-test-data/mxnet_mnist/train"
+            path=os.path.join(data_path, "train"), key_prefix="{}/train".format(s3_prefix)
         )
         test_input = mx.sagemaker_session.upload_data(
-            path=os.path.join(data_path, "test"), key_prefix="integ-test-data/mxnet_mnist/test"
+            path=os.path.join(data_path, "test"), key_prefix="{}/test".format(s3_prefix)
         )
 
         mx.fit({"train": train_input, "test": test_input})
@@ -62,7 +67,13 @@ def test_attach_deploy(mxnet_training_job, sagemaker_session, cpu_instance_type)
 
     with timeout_and_delete_endpoint_by_name(endpoint_name, sagemaker_session):
         estimator = MXNet.attach(mxnet_training_job, sagemaker_session=sagemaker_session)
-        predictor = estimator.deploy(1, cpu_instance_type, endpoint_name=endpoint_name)
+        predictor = estimator.deploy(
+            1,
+            cpu_instance_type,
+            entry_point="mnist.py",
+            source_dir=os.path.join(DATA_DIR, "mxnet_mnist"),
+            endpoint_name=endpoint_name,
+        )
         data = numpy.zeros(shape=(1, 1, 28, 28))
         result = predictor.predict(data)
         assert result is not None