fix: Default AsyncPredictor upload prefix to endpoint name (#6337)

AsyncPredictor accepts name=None by default, but predict() and
predict_async() called name_from_base(self.name) when uploading input
data without an input_path, which raised
"TypeError: 'NoneType' object is not subscriptable".

Fall back to the wrapped predictor's endpoint name when no name is
given, and document the name argument. An explicitly provided name is
still used as the S3 key prefix.

Fixes #3210
Fixes #4774

Author: lucasjia-aws

src/sagemaker/predictor_async.py

@@ -40,6 +40,9 @@ def __init__(
             predictor (sagemaker.predictor.Predictor): General ``Predictor``
                 object has useful methods and variables. ``AsyncPredictor``
                 stands on top of it with capability for async inference.
+            name (str): Optional. Name used as the prefix of the Amazon S3 key when
+                input data is uploaded for async inference. If not specified, the
+                endpoint name is used. (Default: None)
         """
         warn_v2_deprecation(
             feature="AsyncPredictor",
@@ -200,10 +203,13 @@ def _upload_data_to_s3(
             my_uuid = str(uuid.uuid4())
             timestamp = sagemaker_timestamp()
             bucket = self.sagemaker_session.default_bucket()
+            # ``name`` is optional; fall back to the endpoint name so the default
+            # ``AsyncPredictor(predictor)`` construction can upload input data.
+            base_name = self.name or self.endpoint_name
             key = s3.s3_path_join(
                 self.sagemaker_session.default_bucket_prefix,
                 "async-endpoint-inputs",
-                name_from_base(self.name, short=True),
+                name_from_base(base_name, short=True),
                 "{}-{}".format(timestamp, my_uuid),
             )
 

tests/unit/test_predictor_async.py

@@ -174,6 +174,31 @@ def test_async_predict_call_with_data():
     assert result.output_path == ASYNC_OUTPUT_LOCATION
 
 
+def test_async_predict_call_with_data_and_no_name():
+    # Regression test for https://github.com/aws/sagemaker-python-sdk/issues/3210
+    sagemaker_session = empty_sagemaker_session()
+    predictor_async = AsyncPredictor(Predictor(ENDPOINT, sagemaker_session))
+    assert predictor_async.name is None
+
+    result = predictor_async.predict_async(data=DUMMY_DATA)
+
+    _, put_kwargs = sagemaker_session.s3_client.put_object.call_args
+    assert put_kwargs["Bucket"] == BUCKET_NAME
+    assert put_kwargs["Key"].startswith("async-endpoint-inputs/{}-".format(ENDPOINT))
+    assert predictor_async._input_path == "s3://{}/{}".format(BUCKET_NAME, put_kwargs["Key"])
+    assert result.output_path == ASYNC_OUTPUT_LOCATION
+
+
+def test_async_predict_call_with_data_and_name_uses_name():
+    sagemaker_session = empty_sagemaker_session()
+    predictor_async = AsyncPredictor(Predictor(ENDPOINT, sagemaker_session), name=ASYNC_PREDICTOR)
+
+    predictor_async.predict_async(data=DUMMY_DATA)
+
+    _, put_kwargs = sagemaker_session.s3_client.put_object.call_args
+    assert put_kwargs["Key"].startswith("async-endpoint-inputs/{}-".format(ASYNC_PREDICTOR))
+
+
 def test_async_predict_call_with_data_and_input_path():
     sagemaker_session = empty_sagemaker_session()
     predictor_async = AsyncPredictor(Predictor(ENDPOINT, sagemaker_session))