From 0dcae3fe29a6541c9eccaf70f19345865cb53591 Mon Sep 17 00:00:00 2001
From: kthui <18255193+kthui@users.noreply.github.com>
Date: Thu, 30 May 2024 00:14:04 -0700
Subject: [PATCH] Update improper tests for added checking in response sender

---
 .../response_sender/response_sender_test.py   | 30 ++++++++++++++-----
 1 file changed, 22 insertions(+), 8 deletions(-)

diff --git a/qa/L0_backend_python/response_sender/response_sender_test.py b/qa/L0_backend_python/response_sender/response_sender_test.py
index 8634d435bc..296e762313 100644
--- a/qa/L0_backend_python/response_sender/response_sender_test.py
+++ b/qa/L0_backend_python/response_sender/response_sender_test.py
@@ -540,8 +540,11 @@ def test_decoupled_one_response_on_and_post_return(self):
 
     # Non-decoupled model send 2 response before return.
     def test_non_decoupled_two_response_pre_return(self):
-        # Note: The 2 responses sent will make their way to the frontend, but only the
-        #       response at index 0 will be sent back to the client.
+        # Note: The 1st response will make its way to the client, but sending the 2nd
+        #       response will raise an exception which stops the model. However, since
+        #       the exception happens before the model returns, it will be caught by the
+        #       stub process which pass it to the backend and sent an error response
+        #       with final flag.
         self._assert_non_decoupled_infer_with_expected_response_success(
             number_of_response_before_return=2,
             send_complete_final_flag_before_return=True,
@@ -554,9 +557,15 @@ def test_non_decoupled_two_response_pre_return(self):
         )
 
     # Non-decoupled model send 2 response after return.
+    @unittest.skip("Model unload will hang, see the TODO comment.")
     def test_non_decoupled_two_response_post_return(self):
-        # Note: The 2 responses sent will make their way to the frontend, but only the
-        #       response at index 0 will be sent back to the client.
+        # Note: The 1st response will make its way to the client, but sending the 2nd
+        #       response will raise an exception which stops the model. Since the
+        #       exception happens after the model returns, it cannot be caught by the
+        #       stub (i.e. in a daemon thread), so nothing will happen.
+        # TODO: Since the stub does not know if the model failed after returning, the
+        #       complete final flag is not sent and will hang when unloading the model.
+        #       How to detect such event and close the response factory?
         self._assert_non_decoupled_infer_with_expected_response_success(
             number_of_response_before_return=0,
             send_complete_final_flag_before_return=False,
@@ -570,8 +579,10 @@ def test_non_decoupled_two_response_post_return(self):
 
     # Non-decoupled model send 1 response and return 1 response.
     def test_non_decoupled_one_response_pre_and_on_return(self):
-        # Note: The 2 responses sent will make their way to the frontend, but only the
-        #       response at index 0 will be sent back to the client.
+        # Note: The sent response will make its way to the client and complete final.
+        #       The returned response will see the response sender is closed and raise
+        #       an exception. The backend should see the request is closed and do
+        #       nothing upon receiving the error from stub.
         self._assert_non_decoupled_infer_with_expected_response_success(
             number_of_response_before_return=1,
             send_complete_final_flag_before_return=True,
@@ -585,8 +596,11 @@ def test_non_decoupled_one_response_pre_and_on_return(self):
 
     # Non-decoupled model return 1 response and send 1 response.
     def test_non_decoupled_one_response_on_and_pre_return(self):
-        # Note: The 2 responses sent will make their way to the frontend, but only the
-        #       response at index 0 will be sent back to the client.
+        # Note: The returned response will send the response to the client and complete
+        #       final. The sent response will see the response sender is closed and
+        #       raise an exception. Since the exception happens after the model returns,
+        #       it cannot be caught by the stub (i.e. in a daemon thread), so nothing
+        #       will happen.
         self._assert_non_decoupled_infer_with_expected_response_success(
             number_of_response_before_return=0,
             send_complete_final_flag_before_return=False,