Improvements

Author: pskiran1

qa/L0_backend_python/model_readiness/test.sh

@@ -112,28 +112,46 @@ done
 #
 echo -e "\n***\n*** Testing User-Defined is_model_ready() Function\n***"
 
-# Define all test models
-USER_READY_MODELS=(
-    "is_model_ready_fn_returns_true"
-    "is_model_ready_fn_returns_false"
-    "is_model_ready_fn_raises_error"
-    "is_model_ready_fn_returns_non_boolean"
-    "is_model_ready_fn_timeout"
-    "is_model_ready_fn_coroutine_returns_true"
-    "is_model_ready_fn_returns_true_decoupled"
-)
-
-# Create model directories and copy files
-for MODEL in "${USER_READY_MODELS[@]}"; do
-    mkdir -p ./models/$MODEL/1/
-    cp ./test_models/$MODEL/model.py ./models/$MODEL/1/model.py
-    if [ "$MODEL" == "is_model_ready_fn_returns_true_decoupled" ]; then
-        cp ./test_models/$MODEL/config.pbtxt ./models/$MODEL/config.pbtxt
+# Helper function to set up test models with different readiness behaviors based on config parameters
+setup_readiness_test_model() {
+    local model_name=$1
+    local return_value=$2
+    local delay_secs=$3
+
+    mkdir -p ./models/$model_name/1/
+    if [ "$model_name" == "is_model_ready_fn_coroutine_returns_true" ]; then
+        cp ./test_models/readiness_coroutine_model.py ./models/$model_name/1/model.py
     else
-        cp ./models/identity_fp32/config.pbtxt ./models/$MODEL/config.pbtxt
-        sed -i "s/^name:.*/name: \"$MODEL\"/" ./models/$MODEL/config.pbtxt
+        cp ./test_models/readiness_model.py ./models/$model_name/1/model.py
     fi
-done
+    cp ./models/identity_fp32/config.pbtxt ./models/$model_name/config.pbtxt
+    sed -i "s/^name:.*/name: \"$model_name\"/" ./models/$model_name/config.pbtxt
+    cat >> ./models/$model_name/config.pbtxt << EOF
+parameters: {
+  key: "READINESS_FN_RETURN_VALUE"
+  value: { string_value: "$return_value" }
+}
+parameters: {
+  key: "READINESS_FN_DELAY_SECS"
+  value: { string_value: "$delay_secs" }
+}
+EOF
+}
+
+# Create readiness test models using shared model.py + config parameters
+setup_readiness_test_model "is_model_ready_fn_returns_true" "true" "0.1"
+setup_readiness_test_model "is_model_ready_fn_returns_false" "false" "0.1"
+setup_readiness_test_model "is_model_ready_fn_raises_error" "exception" "0.1"
+setup_readiness_test_model "is_model_ready_fn_returns_non_boolean" "non_boolean" "0.1"
+setup_readiness_test_model "is_model_ready_fn_timeout" "true" "8"
+setup_readiness_test_model "is_model_ready_fn_coroutine_returns_true" "coroutine" "0.1"
+
+# Decoupled model has a unique execute() and its own config
+mkdir -p ./models/is_model_ready_fn_returns_true_decoupled/1/
+cp ./test_models/is_model_ready_fn_returns_true_decoupled/model.py \
+    ./models/is_model_ready_fn_returns_true_decoupled/1/model.py
+cp ./test_models/is_model_ready_fn_returns_true_decoupled/config.pbtxt \
+    ./models/is_model_ready_fn_returns_true_decoupled/config.pbtxt
 
 # Start server with all models
 SERVER_ARGS="--model-repository=$(pwd)/models --backend-directory=${BACKEND_DIR} --log-verbose=1"

qa/L0_backend_python/model_readiness/test_model_readiness.py

@@ -53,9 +53,7 @@ def callback(user_data, result, error):
 
 
 def prepare_infer_args(input_value):
-    """
-    Create InferInput/InferRequestedOutput lists
-    """
+    """Create InferInput and InferRequestedOutput lists for decoupled inference."""
     input_data = np.array([[input_value]], dtype=np.int32)
     infer_input = [grpcclient.InferInput("IN", input_data.shape, "INT32")]
     infer_input[0].set_data_from_numpy(input_data)
@@ -65,7 +63,7 @@ def prepare_infer_args(input_value):
 
 def collect_responses(user_data, expected_responses_count):
     """
-    Collect responses from user_data until the final response flag is seen.
+    Collect up to `expected_responses_count` responses from user_data.
     """
     errors = []
     responses = []
@@ -88,7 +86,7 @@ def collect_responses(user_data, expected_responses_count):
 
 
 def call_inference_identity_model(model_name, protocol, client):
-    """Helper to test inference functionality"""
+    """Send an inference request and verify the output matches the input."""
     shape = (1, 8)
     input_data = np.ones(shape, dtype=np.float32)
 
@@ -167,9 +165,7 @@ def setUp(self):
         self.client_grpc = grpcclient.InferenceServerClient(url=URL_GRPC)
 
     def _run_inference_decoupled(self, index, model_name, expected_responses_count):
-        """
-        Helper function for streaming inference.
-        """
+        """Send a decoupled streaming inference request and verify responses."""
         user_data = UserData()
         with grpcclient.InferenceServerClient(URL_GRPC) as triton_client:
             try:
@@ -284,8 +280,7 @@ def test_is_model_ready_returns_true(self):
                 f"iteration {i} - GRPC client - Model {model_name} should be READY",
             )
 
-            # Inference should work normally
-            # readiness check functionality should not affect inference
+            # Verify inference is unaffected by readiness checks.
             call_inference_identity_model(model_name, "http", self.client_http)
             call_inference_identity_model(model_name, "grpc", self.client_grpc)
 
@@ -304,8 +299,7 @@ def test_is_model_ready_returns_false(self):
                 f"iteration {i} - GRPC client - Model {model_name} should be NOT READY",
             )
 
-            # Inference should work normally
-            # readiness check functionality should not affect inference
+            # Verify inference is unaffected by readiness checks.
             call_inference_identity_model(model_name, "http", self.client_http)
             call_inference_identity_model(model_name, "grpc", self.client_grpc)
 
@@ -324,12 +318,11 @@ def test_is_model_ready_raises_exception(self):
                 f"iteration {i} - GRPC client - Model {model_name} should be NOT READY (exception)",
             )
 
-            # Inference should work normally
-            # readiness check functionality should not affect inference
+            # Verify inference is unaffected by readiness checks.
             call_inference_identity_model(model_name, "http", self.client_http)
             call_inference_identity_model(model_name, "grpc", self.client_grpc)
 
-        # Test good model afterwards to ensure server is healthy
+        # Verify a healthy model is still ready to confirm server stability.
         model_name = "is_model_ready_fn_returns_true"
         for i in range(num_requests):
             self.assertTrue(
@@ -341,8 +334,7 @@ def test_is_model_ready_raises_exception(self):
                 f"iteration {i} - GRPC client - Model {model_name} should be READY",
             )
 
-            # Inference should work normally
-            # readiness check functionality should not affect inference
+            # Verify inference is unaffected by readiness checks.
             call_inference_identity_model(model_name, "http", self.client_http)
             call_inference_identity_model(model_name, "grpc", self.client_grpc)
 
@@ -361,12 +353,11 @@ def test_is_model_ready_returns_non_boolean(self):
                 f"iteration {i} - GRPC client - Model {model_name} should be NOT READY (wrong return type)",
             )
 
-            # Inference should work normally
-            # readiness check functionality should not affect inference
+            # Verify inference is unaffected by readiness checks.
             call_inference_identity_model(model_name, "http", self.client_http)
             call_inference_identity_model(model_name, "grpc", self.client_grpc)
 
-        # Test good model afterwards to ensure server is healthy
+        # Verify a healthy model is still ready to confirm server stability.
         model_name = "is_model_ready_fn_returns_true"
         for i in range(num_requests):
             self.assertTrue(
@@ -378,8 +369,7 @@ def test_is_model_ready_returns_non_boolean(self):
                 f"iteration {i} - GRPC client - Model {model_name} should be READY",
             )
 
-            # Inference should work normally
-            # readiness check functionality should not affect inference
+            # Verify inference is unaffected by readiness checks.
             call_inference_identity_model(model_name, "http", self.client_http)
             call_inference_identity_model(model_name, "grpc", self.client_grpc)
 
@@ -450,7 +440,7 @@ def do_inference(protocol, index):
             except Exception as e:
                 infer_errors[protocol].append((index, str(e)))
 
-        # Launch concurrent ready checks
+        # Launch concurrent readiness and inference requests.
         http_threads = []
         for i in range(num_requests):
             t1 = threading.Thread(target=check_model_readiness, args=("http", i))
@@ -466,8 +456,6 @@ def do_inference(protocol, index):
         for t in http_threads:
             self.assertFalse(t.is_alive(), f"HTTP threads are not completed")
 
-        time.sleep(5)
-
         grpc_threads = []
         for i in range(num_requests):
             t1 = threading.Thread(target=check_model_readiness, args=("grpc", i))
@@ -483,8 +471,6 @@ def do_inference(protocol, index):
         for t in grpc_threads:
             self.assertFalse(t.is_alive(), f"gRPC threads are not completed")
 
-        time.sleep(5)
-
         # Verify no errors in readiness checks
         self.assertEqual(
             len(ready_errors["http"]), 0, f"HTTP errors: {ready_errors['http']}"