Only create the endpoint if it does not already exist, fixes #182 in k8s

llmops/common/deployment/kubernetes_endpoint.py

@@ -1,5 +1,5 @@
 """
-This module creates Kubernetes managed endpoint as flow deployment process.
+This module creates Kubernetes managed endpoint as flow deployment process if it does not already exist.
 
 Args:
 --base_path: Base path of the use case. Where flows, data,
@@ -27,7 +27,7 @@
 from llmops.common.logger import llmops_logger
 from llmops.common.experiment_cloud_config import ExperimentCloudConfig
 
-logger = llmops_logger("provision_endpoint")
+logger = llmops_logger("kubernetes_endpoint")
 
 
 def create_kubernetes_endpoint(
@@ -51,6 +51,8 @@ def create_kubernetes_endpoint(
         credential=DefaultAzureCredential(),
     )
 
+    existing_endpoints = ml_client.online_endpoints.list(local=False)
+
     config_file = open(real_config)
     endpoint_config = json.load(config_file)
 
@@ -61,21 +63,35 @@ def create_kubernetes_endpoint(
                 endpoint_desc = elem["ENDPOINT_DESC"]
                 compute_name = elem["COMPUTE_NAME"]
 
-                endpoint = KubernetesOnlineEndpoint(
-                    name=endpoint_name,
-                    description=endpoint_desc,
-                    compute=compute_name,
-                    auth_mode="key",
-                    tags={"build_id": build_id} if build_id else {},
-                    properties={
-                        "enforce_access_to_default_secret_stores": True,
-                    },
-                )
-
-                logger.info(f"Creating endpoint {endpoint.name}")
-                ml_client.online_endpoints.begin_create_or_update(
-                    endpoint=endpoint
-                ).result()
+                # See if endpoint with name endpoint_name already exists
+                endpoint = next(
+                    (
+                        e for e in existing_endpoints
+                        if e.name == endpoint_name
+                        ),
+                    None)
+
+                if endpoint is None:
+                    endpoint = KubernetesOnlineEndpoint(
+                        name=endpoint_name,
+                        description=endpoint_desc,
+                        compute=compute_name,
+                        auth_mode="key",
+                        tags={"build_id": build_id} if build_id else {},
+                        properties={
+                            "enforce_access_to_default_secret_stores": True,
+                        },
+                    )
+
+                    logger.info(f"Creating endpoint {endpoint.name}")
+                    ml_client.online_endpoints.begin_create_or_update(
+                        endpoint=endpoint
+                    ).result()
+                else:
+                    logger.info(
+                        f"Skipping create as endpoint"
+                        f"{endpoint.name} already exists"
+                        )
 
                 logger.info(f"Obtaining endpoint {endpoint.name} identity")
                 principal_id = ml_client.online_endpoints.get(

tests/test_create_kubernetes_endpoint.py

@@ -19,8 +19,7 @@ def _set_required_env_vars():
     monkeypatch.setenv("RESOURCE_GROUP_NAME", "TEST_RESOURCE_GROUP_NAME")
     monkeypatch.setenv("WORKSPACE_NAME", "TEST_WORKSPACE_NAME")
 
-
-def test_create_kubernetes_endpoint():
+def test_create_kubernetes_endpoint_when_not_exists():
     """Test create_kubernetes_endpoint."""
     env_name = "dev"
     endpoint_name = "k8s-test-endpoint"
@@ -32,6 +31,8 @@ def test_create_kubernetes_endpoint():
         # Mock the MLClient
         ml_client_instance = Mock()
         mock_ml_client.return_value = ml_client_instance
+        ml_client_instance.online_endpoints.list.return_value = []
+
         # Create the endpoint
         create_kubernetes_endpoint(env_name, str(RESOURCE_PATH))
 
@@ -61,3 +62,33 @@ def test_create_kubernetes_endpoint():
         assert created_endpoint.description == endpoint_description
         assert created_endpoint.compute == compute_name
         assert created_endpoint.auth_mode == "key"
+
+def test_create_kubernetes_endpoint_when_exists():
+    """Test create_kubernetes_endpoint."""
+    env_name = "dev"
+    endpoint_name = "k8s-test-endpoint"
+    with patch(
+        "llmops.common.deployment.kubernetes_endpoint.MLClient"
+    ) as mock_ml_client:
+        # Mock the MLClient
+        ml_client_instance = Mock()
+        mock_ml_client.return_value = ml_client_instance
+
+        mock_endpoint = Mock()
+        mock_endpoint.name = endpoint_name
+        ml_client_instance.online_endpoints.list.return_value = [
+            mock_endpoint
+        ]
+
+        # Create the endpoint
+        create_kubernetes_endpoint(env_name, str(RESOURCE_PATH))
+
+        # Assert online_endpoints.begin_create_or_update is called once
+        create_endpoint_calls = (
+            ml_client_instance.online_endpoints.begin_create_or_update
+        )
+
+        # Endpoint should not be created if it already exists as it would
+        # set the traffic to zero for existing deployments and there are 
+        # no properties we need to update on the endpoint
+        assert create_endpoint_calls.call_count == 0