Adding unit tests for apply

Author: akram

src/codeflare_sdk/common/utils/unit_test_support.py

@@ -27,10 +27,15 @@
 
 
 def createClusterConfig():
+    config = createClusterConfigWithNumWorkers()
+    return config
+
+
+def createClusterConfigWithNumWorkers(num_workers=2):
     config = ClusterConfiguration(
         name="unit-test-cluster",
         namespace="ns",
-        num_workers=2,
+        num_workers=num_workers,
         worker_cpu_requests=3,
         worker_cpu_limits=4,
         worker_memory_requests=5,
@@ -41,13 +46,21 @@ def createClusterConfig():
     return config
 
 
-def createClusterWithConfig(mocker):
+def createClusterWithConfigAndNumWorkers(mocker, num_workers=2, dynamic_client=None):
     mocker.patch("kubernetes.config.load_kube_config", return_value="ignore")
     mocker.patch(
         "kubernetes.client.CustomObjectsApi.get_cluster_custom_object",
         return_value={"spec": {"domain": "apps.cluster.awsroute.org"}},
     )
-    cluster = Cluster(createClusterConfig())
+    cluster = Cluster(createClusterConfigWithNumWorkers(num_workers))
+    mocker.patch.object(cluster, "get_dynamic_client", return_value=dynamic_client)
+    mocker.patch.object(cluster, "down", return_value=None)
+    mocker.patch.object(cluster, "config_check", return_value=None)
+    return cluster
+
+
+def createClusterWithConfig(mock_config):
+    cluster = createClusterWithConfigAndNumWorkers(mock_config)
     return cluster
 
 
@@ -383,6 +396,48 @@ def mocked_ingress(port, cluster_name="unit-test-cluster", annotations: dict = N
     return mock_ingress
 
 
+# Global dictionary to maintain state in the mock
+cluster_state = {}
+
+
+# The mock side_effect function for server_side_apply
+def mock_server_side_apply(resource, body=None, name=None, namespace=None, **kwargs):
+    # Simulate the behavior of server_side_apply:
+    # Update a mock state that represents the cluster's current configuration.
+    # Stores the state in a global dictionary for simplicity.
+
+    global cluster_state
+
+    if not resource or not body or not name or not namespace:
+        raise ValueError("Missing required parameters for server_side_apply")
+
+    # Extract worker count from the body if it exists
+    try:
+        worker_count = (
+            body["spec"]["workerGroupSpecs"][0]["replicas"]
+            if "spec" in body and "workerGroupSpecs" in body["spec"]
+            else None
+        )
+    except KeyError:
+        worker_count = None
+
+    # Apply changes to the cluster_state mock
+    cluster_state[name] = {
+        "namespace": namespace,
+        "worker_count": worker_count,
+        "body": body,
+    }
+
+    # Return a response that mimics the behavior of a successful apply
+    return {
+        "status": "success",
+        "applied": True,
+        "name": name,
+        "namespace": namespace,
+        "worker_count": worker_count,
+    }
+
+
 @patch.dict("os.environ", {"NB_PREFIX": "test-prefix"})
 def create_cluster_all_config_params(mocker, cluster_name, is_appwrapper) -> Cluster:
     mocker.patch(

src/codeflare_sdk/ray/cluster/cluster.py

@@ -88,6 +88,14 @@ def __init__(self, config: ClusterConfiguration):
         if is_notebook():
             cluster_up_down_buttons(self)
 
+    def get_dynamic_client(self):
+        """Return a dynamic client, optionally mocked in tests."""
+        return DynamicClient(get_api_client())
+
+    def config_check(self):
+        """Return a dynamic client, optionally mocked in tests."""
+        return config_check()
+
     @property
     def _client_headers(self):
         k8_client = get_api_client()
@@ -144,7 +152,7 @@ def up(self):
         the Kueue localqueue.
         """
         # TODO: Add deprecation message in favor of apply()
-        # print( "WARNING: The up() is planned for deprecation in favor of apply().")        
+        # print( "WARNING: The up() is planned for deprecation in favor of apply().")
 
         # check if RayCluster CustomResourceDefinition exists if not throw RuntimeError
         self._throw_for_no_raycluster()
@@ -182,30 +190,27 @@ def up(self):
         except Exception as e:  # pragma: no cover
             return _kube_api_error_handling(e)
 
-
     def apply(self, force=False):
         """
         Applies the Cluster yaml using server-side apply.
         If 'force' is set to True, conflicts will be forced.
         """
+        # check if RayCluster CustomResourceDefinition exists if not throw RuntimeError
+        self._throw_for_no_raycluster()
+        namespace = self.config.namespace
         # Ensure Kubernetes configuration is loaded
-        config_check()
-
-        # Create a dynamic client for interacting with custom resources
-        dynamic_client = DynamicClient(get_api_client())
-
         try:
+            self.config_check()
             # Get the RayCluster custom resource definition
-            api = dynamic_client.resources.get(
-                api_version="ray.io/v1",
-                kind="RayCluster"
+            api = self.get_dynamic_client().resources.get(
+                api_version="ray.io/v1", kind="RayCluster"
             )
         except Exception as e:
             raise RuntimeError("Failed to get RayCluster resource: " + str(e))
 
         # Read the YAML file and parse it into a dictionary
         try:
-            with open(self.resource_yaml, 'r') as f:
+            with open(self.resource_yaml, "r") as f:
                 resource_body = yaml.safe_load(f)
         except FileNotFoundError:
             raise RuntimeError(f"Resource YAML file '{self.resource_yaml}' not found.")
@@ -216,15 +221,14 @@ def apply(self, force=False):
         resource_name = resource_body.get("metadata", {}).get("name")
         if not resource_name:
             raise ValueError("The resource must have a 'metadata.name' field.")
-
         try:
             # Use server-side apply
             resp = api.server_side_apply(
                 body=resource_body,
                 name=resource_name,
                 namespace=self.config.namespace,
                 field_manager="cluster-manager",
-                force_conflicts=force  # Allow forcing conflicts if needed
+                force_conflicts=force,  # Allow forcing conflicts if needed
             )
             print(f"Cluster '{self.config.name}' applied successfully.")
         except ApiException as e:
@@ -234,7 +238,9 @@ def apply(self, force=False):
                     "To force the patch, set 'force=True' in the apply() method."
                 )
             elif e.status == 404:
-                print(f"Namespace '{self.config.namespace}' or resource '{resource_name}' not found. Verify the namespace or CRD.")
+                print(
+                    f"Namespace '{self.config.namespace}' or resource '{resource_name}' not found. Verify the namespace or CRD."
+                )
             else:
                 raise RuntimeError(f"Failed to apply cluster: {e.reason}")
 
@@ -266,7 +272,7 @@ def down(self):
         resource_name = self.config.name
         self._throw_for_no_raycluster()
         try:
-            config_check()
+            self.config_check()
             api_instance = client.CustomObjectsApi(get_api_client())
             if self.config.appwrapper:
                 api_instance.delete_namespaced_custom_object(

src/codeflare_sdk/ray/cluster/test_cluster.py

@@ -20,6 +20,7 @@
 )
 from codeflare_sdk.common.utils.unit_test_support import (
     createClusterWithConfig,
+    createClusterWithConfigAndNumWorkers,
     arg_check_del_effect,
     ingress_retrieval,
     arg_check_apply_effect,
@@ -29,6 +30,7 @@
     get_obj_none,
     get_ray_obj_with_status,
     get_aw_obj_with_status,
+    mock_server_side_apply,
 )
 from codeflare_sdk.ray.cluster.cluster import _is_openshift_cluster
 from pathlib import Path
@@ -67,11 +69,48 @@ def test_cluster_up_down(mocker):
         "kubernetes.client.CustomObjectsApi.list_namespaced_custom_object",
         return_value=get_local_queue("kueue.x-k8s.io", "v1beta1", "ns", "localqueues"),
     )
-    cluster = cluster = createClusterWithConfig(mocker)
+    cluster = createClusterWithConfig(mocker)
     cluster.up()
     cluster.down()
 
 
+def test_cluster_apply_scale_up_scale_down(mocker):
+    mock_dynamic_client = mocker.Mock()
+    mocker.patch("codeflare_sdk.ray.cluster.cluster.Cluster._throw_for_no_raycluster")
+    mocker.patch(
+        "kubernetes.dynamic.DynamicClient.resources", new_callable=mocker.PropertyMock
+    )
+    mocker.patch(
+        "codeflare_sdk.ray.cluster.cluster.Cluster.create_resource",
+        return_value="./tests/test_cluster_yamls/ray/default-ray-cluster.yaml",
+    )
+
+    # Initialize test
+    initial_num_workers = 1
+    scaled_up_num_workers = 2
+
+    # Step 1: Create cluster with initial workers
+    cluster = createClusterWithConfigAndNumWorkers(
+        mocker, initial_num_workers, dynamic_client=mock_dynamic_client
+    )
+    cluster.apply()
+
+    # Step 2: Scale up the cluster
+    cluster = createClusterWithConfigAndNumWorkers(
+        mocker, scaled_up_num_workers, dynamic_client=mock_dynamic_client
+    )
+    cluster.apply()
+
+    # Step 3: Scale down the cluster
+    cluster = createClusterWithConfigAndNumWorkers(
+        mocker, initial_num_workers, dynamic_client=mock_dynamic_client
+    )
+    cluster.apply()
+
+    # Tear down
+    cluster.down()
+
+
 def test_cluster_up_down_no_mcad(mocker):
     mocker.patch("codeflare_sdk.ray.cluster.cluster.Cluster._throw_for_no_raycluster")
     mocker.patch("kubernetes.config.load_kube_config", return_value="ignore")
@@ -117,7 +156,7 @@ def test_cluster_uris(mocker):
         "kubernetes.client.CustomObjectsApi.list_namespaced_custom_object",
         return_value=get_local_queue("kueue.x-k8s.io", "v1beta1", "ns", "localqueues"),
     )
-    cluster = cluster = createClusterWithConfig(mocker)
+    cluster = createClusterWithConfig(mocker)
     mocker.patch(
         "kubernetes.client.NetworkingV1Api.list_namespaced_ingress",
         return_value=ingress_retrieval(
@@ -159,7 +198,7 @@ def ray_addr(self, *args):
         "kubernetes.client.CustomObjectsApi.list_namespaced_custom_object",
         return_value=get_local_queue("kueue.x-k8s.io", "v1beta1", "ns", "localqueues"),
     )
-    cluster = cluster = createClusterWithConfig(mocker)
+    cluster = createClusterWithConfig(mocker)
     mocker.patch(
         "ray.job_submission.JobSubmissionClient._check_connection_and_version_with_url",
         return_value="None",

tests/e2e/cluster_apply_test.py renamed to tests/e2e/cluster_apply_kind_test.py

@@ -62,7 +62,7 @@ def test_cluster_apply(self):
         updated_config = ClusterConfiguration(
             name=cluster_name,
             namespace=namespace,
-            num_workers=3,
+            num_workers=2,
             head_cpu_requests="500m",
             head_cpu_limits="1",
             head_memory_requests="1Gi",
@@ -82,12 +82,14 @@ def test_cluster_apply(self):
         # Wait for the updated cluster to be ready
         cluster.wait_ready()
         updated_status = cluster.status()
-        assert updated_status["ready"], f"Cluster {cluster_name} is not ready after update: {updated_status}"
+        assert updated_status[
+            "ready"
+        ], f"Cluster {cluster_name} is not ready after update: {updated_status}"
 
         # Verify the cluster is updated
         updated_ray_cluster = get_ray_cluster(cluster_name, namespace)
         assert (
-            updated_ray_cluster["spec"]["workerGroupSpecs"][0]["replicas"] == 3
+            updated_ray_cluster["spec"]["workerGroupSpecs"][0]["replicas"] == 2
         ), "Worker count was not updated"
 
         # Clean up
@@ -152,4 +154,3 @@ def test_apply_invalid_update(self):
 
         # Clean up
         cluster.down()
-

tests/e2e/support.py

@@ -11,6 +11,22 @@
 )
 
 
+def get_ray_cluster(cluster_name, namespace):
+    api = client.CustomObjectsApi()
+    try:
+        return api.get_namespaced_custom_object(
+            group="ray.io",
+            version="v1",
+            namespace=namespace,
+            plural="rayclusters",
+            name=cluster_name,
+        )
+    except client.exceptions.ApiException as e:
+        if e.status == 404:
+            return None
+        raise
+
+
 def get_ray_image():
     default_ray_image = "quay.io/modh/ray@sha256:0d715f92570a2997381b7cafc0e224cfa25323f18b9545acfd23bc2b71576d06"
     return os.getenv("RAY_IMAGE", default_ray_image)