Ensure that self.manifest is defined before call profiler (#2488)

* Ensure that self.manifest is defined before call profiler

docs/performance_guide.md

@@ -77,6 +77,8 @@ You can find more information on TorchServe benchmarking [here](https://github.c
 
 TorchServe has native support for the PyTorch profiler which will help you find performance bottlenecks in your code.
 
+If you created a custom `handle` or `initialize` method overwriting the BaseHandler, you must define the `self.manifest` attribute to be able to run `_infer_with_profiler`.  
+
 ```
 export ENABLE_TORCH_PROFILER=TRUE
 ```

ts/torch_handler/base_handler.py

@@ -333,6 +333,9 @@ def handle(self, data, context):
         is_profiler_enabled = os.environ.get("ENABLE_TORCH_PROFILER", None)
         if is_profiler_enabled:
             if PROFILER_AVAILABLE:
+                if self.manifest is None:
+                    # profiler will use to get the model name
+                    self.manifest = context.manifest
                 output, _ = self._infer_with_profiler(data=data)
             else:
                 raise RuntimeError(

ts/torch_handler/unit_tests/test_base_handler.py

@@ -5,9 +5,10 @@
 Ensures it can load and execute an example model
 """
 
+import os
 import pytest
 
-from ts.torch_handler.base_handler import BaseHandler
+from ts.torch_handler.base_handler import BaseHandler, PROFILER_AVAILABLE
 
 
 @pytest.fixture()
@@ -30,3 +31,13 @@ def test_batch_handle(handler, base_model_context):
     processed = handler.handle(list_data, base_model_context)
 
     assert processed == [1, 0]
+
+
+def test_inference_with_profiler_works_with_custom_initialize_method(handler, base_model_context):
+    handler.manifest = None
+    PROFILER_AVAILABLE = True
+    os.environ["ENABLE_TORCH_PROFILER"] = "1"
+
+    list_data = [[1.0, 2.0], [4.0, 3.0]]
+    processed = handler.handle(list_data, base_model_context)
+    assert processed == [1, 0]