Merge pull request #267 from canonical/cherrypick-fix-configure-proxy…

…-env-vars-storage-init-0.11

(cherry-pick to track/0.11) fix: configure proxy env vars storage init container (#257)

charms/kserve-controller/config.yaml

@@ -35,4 +35,15 @@ options:
     description: >
       YAML or JSON formatted input defining images to use in Katib
       For usage details, see https://github.com/canonical/kserve-operators.
-
+  http-proxy:
+    default: ""
+    description: The value of HTTP_PROXY environment variable in the storage-initializer container.
+    type: string
+  https-proxy:
+    default: ""
+    description: The value of HTTPS_PROXY environment variable in the storage-initializer container.
+    type: string
+  no-proxy:
+    default: ""
+    description: The value of NO_PROXY environment variable in the storage-initializer container.
+    type: string

charms/kserve-controller/src/charm.py

@@ -68,6 +68,7 @@
     "src/templates/auth_manifests.yaml.j2",
     "src/templates/serving_runtimes_manifests.yaml.j2",
     "src/templates/webhook_manifests.yaml.j2",
+    "src/templates/cluster_storage_containers.yaml.j2",
 ]
 
 # Values for MinIO manifests https://kserve.github.io/website/0.11/modelserving/storage/s3/s3/
@@ -164,6 +165,9 @@ def _context(self):
             "app_name": self.app.name,
             "namespace": self.model.name,
             "cert": f"'{ca_context.decode('utf-8')}'",
+            "http_proxy": self.model.config["http-proxy"],
+            "https_proxy": self.model.config["https-proxy"],
+            "no_proxy": self.model.config["no-proxy"],
         }
 
     @property

charms/kserve-controller/src/templates/cluster_storage_containers.yaml.j2

@@ -0,0 +1,38 @@
+apiVersion: serving.kserve.io/v1alpha1
+kind: ClusterStorageContainer
+metadata:
+  name: default
+spec:
+  container:
+    image: {{ configmap__storageInitializer }}
+    name: storage-initializer
+    resources:
+      limits:
+        cpu: "1"
+        memory: 1Gi
+      requests:
+        cpu: 100m
+        memory: 100Mi
+  {% if http_proxy or https_proxy or no_proxy %}
+    env:
+  {% if http_proxy %}
+    - name: HTTP_PROXY
+      value: {{ http_proxy }}
+  {% endif %}
+  {% if https_proxy %}
+    - name: HTTPS_PROXY
+      value: {{ https_proxy }}
+  {% endif %}
+  {% if no_proxy %}
+    - name: NO_PROXY
+      value: {{ no_proxy }}
+  {% endif %}
+  {% endif %}
+  supportedUriFormats:
+  - prefix: gs://
+  - prefix: s3://
+  - prefix: hdfs://
+  - prefix: webhdfs://
+  - regex: https://(.+?).blob.core.windows.net/(.+)
+  - regex: https://(.+?).file.core.windows.net/(.+)
+  - regex: https?://(.+)/(.+)

charms/kserve-controller/tests/integration/test_charm.py

@@ -25,6 +25,7 @@
     ServiceAccount,
 )
 from pytest_operator.plugin import OpsTest
+from tenacity import Retrying, stop_after_delay, wait_fixed
 
 logger = logging.getLogger(__name__)
 
@@ -73,6 +74,18 @@
 TESTING_NAMESPACE_NAME = "raw-deployment"
 KSERVE_WORKLOAD_CONTAINER = "kserve-container"
 
+ISVC = lightkube.generic_resource.create_namespaced_resource(
+    group="serving.kserve.io",
+    version="v1beta1",
+    kind="InferenceService",
+    plural="inferenceservices",
+    verbs=None,
+)
+
+SKLEARN_INF_SVC_YAML = yaml.safe_load(Path("./tests/integration/sklearn-iris.yaml").read_text())
+SKLEARN_INF_SVC_OBJECT = lightkube.codecs.load_all_yaml(yaml.dump(SKLEARN_INF_SVC_YAML))[0]
+SKLEARN_INF_SVC_NAME = SKLEARN_INF_SVC_OBJECT.metadata.name
+
 
 def deploy_k8s_resources(template_files: str):
     """Deploy k8s resources from template files."""
@@ -163,15 +176,19 @@ def namespace(lightkube_client: lightkube.Client):
     delete_all_from_yaml(yaml_text, lightkube_client)
 
 
-@pytest.fixture
-def cleanup_namespaces_after_execution(request):
-    """Removes the namespaces used for deploying inferenceservices."""
-    yield
+@pytest.fixture(scope="function")
+def serverless_namespace(lightkube_client):
+    """Create a namespaces used for deploying inferenceservices, cleaning it up afterwards."""
+
+    namespace_name = "serverless-namespace"
+    lightkube_client.create(Namespace(metadata=ObjectMeta(name=namespace_name)))
+
+    yield namespace_name
+
     try:
-        lightkube_client = lightkube.Client()
-        lightkube_client.delete(Namespace, name=request.param)
+        lightkube_client.delete(Namespace, name=namespace_name)
     except ApiError:
-        logger.warning(f"The {request.param} namespace could not be removed.")
+        logger.warning(f"The {namespace_name} namespace could not be removed.")
         pass
 
 
@@ -268,14 +285,8 @@ def test_inference_service_raw_deployment(
     test_namespace: None, lightkube_client: lightkube.Client, inference_file, ops_test: OpsTest
 ):
     """Validates that an InferenceService can be deployed."""
-    # Read InferenceService example and create namespaced resource
-    inference_service_resource = lightkube.generic_resource.create_namespaced_resource(
-        group="serving.kserve.io",
-        version="v1beta1",
-        kind="InferenceService",
-        plural="inferenceservices",
-        verbs=None,
-    )
+    # Read InferenceService example
+
     inf_svc_yaml = yaml.safe_load(Path(inference_file).read_text())
     inf_svc_object = lightkube.codecs.load_all_yaml(yaml.dump(inf_svc_yaml))[0]
     inf_svc_name = inf_svc_object.metadata.name
@@ -296,9 +307,7 @@ def create_inf_svc():
         reraise=True,
     )
     def assert_inf_svc_state():
-        inf_svc = lightkube_client.get(
-            inference_service_resource, inf_svc_name, namespace=TESTING_NAMESPACE_NAME
-        )
+        inf_svc = lightkube_client.get(ISVC, inf_svc_name, namespace=TESTING_NAMESPACE_NAME)
         conditions = inf_svc.get("status", {}).get("conditions")
         logger.info(
             f"INFO: Inspecting InferenceService {inf_svc.metadata.name} in namespace {inf_svc.metadata.namespace}"
@@ -379,40 +388,19 @@ async def test_deploy_knative_dependencies(ops_test: OpsTest):
     )
 
 
-@pytest.mark.parametrize(
-    "cleanup_namespaces_after_execution", ["serverless-namespace"], indirect=True
-)
-def test_inference_service_serverless_deployment(
-    cleanup_namespaces_after_execution, ops_test: OpsTest
-):
+def test_inference_service_serverless_deployment(serverless_namespace, ops_test: OpsTest):
     """Validates that an InferenceService can be deployed."""
     # Instantiate a lightkube client
     lightkube_client = lightkube.Client()
 
-    # Read InferenceService example and create namespaced resource
-    inference_service_resource = lightkube.generic_resource.create_namespaced_resource(
-        group="serving.kserve.io",
-        version="v1beta1",
-        kind="InferenceService",
-        plural="inferenceservices",
-        verbs=None,
-    )
-    inf_svc_yaml = yaml.safe_load(Path("./tests/integration/sklearn-iris.yaml").read_text())
-    inf_svc_object = lightkube.codecs.load_all_yaml(yaml.dump(inf_svc_yaml))[0]
-    inf_svc_name = inf_svc_object.metadata.name
-    serverless_mode_namespace = "serverless-namespace"
-
-    # Create Serverless namespace
-    lightkube_client.create(Namespace(metadata=ObjectMeta(name=serverless_mode_namespace)))
-
     # Create InferenceService from example file
     @tenacity.retry(
         wait=tenacity.wait_exponential(multiplier=1, min=1, max=15),
         stop=tenacity.stop_after_delay(30),
         reraise=True,
     )
     def create_inf_svc():
-        lightkube_client.create(inf_svc_object, namespace=serverless_mode_namespace)
+        lightkube_client.create(SKLEARN_INF_SVC_OBJECT, namespace=serverless_namespace)
 
     # Assert InferenceService state is Available
     @tenacity.retry(
@@ -421,9 +409,7 @@ def create_inf_svc():
         reraise=True,
     )
     def assert_inf_svc_state():
-        inf_svc = lightkube_client.get(
-            inference_service_resource, inf_svc_name, namespace=serverless_mode_namespace
-        )
+        inf_svc = lightkube_client.get(ISVC, SKLEARN_INF_SVC_NAME, namespace=serverless_namespace)
         conditions = inf_svc.get("status", {}).get("conditions")
         for condition in conditions:
             if condition.get("status") == "False":
@@ -560,6 +546,64 @@ async def test_new_user_namespace_has_manifests(
     assert service_account.secrets[0].name == manifests_name
 
 
+RETRY_FOR_THREE_MINUTES = Retrying(
+    stop=stop_after_delay(60 * 3),
+    wait=wait_fixed(5),
+    reraise=True,
+)
+
+
+async def test_inference_service_proxy_envs_configuration(
+    serverless_namespace, ops_test: OpsTest, lightkube_client: lightkube.Client
+):
+    """Changes `http-proxy`, `https-proxy` and `no-proxy` configs and asserts that
+    the InferenceService Pod is using the values from configs as environment variables."""
+
+    # Set Proxy envs by setting the charm configs
+    test_http_proxy = "my_http_proxy"
+    test_https_proxy = "my_https_proxy"
+    test_no_proxy = "no_proxy"
+
+    await ops_test.model.applications["kserve-controller"].set_config(
+        {"http-proxy": test_http_proxy, "https-proxy": test_https_proxy, "no-proxy": test_no_proxy}
+    )
+
+    await ops_test.model.wait_for_idle(
+        ["kserve-controller"],
+        status="active",
+        raise_on_blocked=False,
+        timeout=60 * 1,
+    )
+
+    # Create InferenceService from example file
+    for attempt in RETRY_FOR_THREE_MINUTES:
+        with attempt:
+            lightkube_client.create(SKLEARN_INF_SVC_OBJECT, namespace=serverless_namespace)
+
+    # Assert InferenceService Pod specifies the proxy envs for the initContainer
+    for attempt in RETRY_FOR_THREE_MINUTES:
+        with attempt:
+            pods_list = lightkube_client.list(
+                res=Pod,
+                namespace=serverless_namespace,
+                labels={"serving.kserve.io/inferenceservice": SKLEARN_INF_SVC_NAME},
+            )
+            isvc_pod = next(pods_list)
+            init_env_vars = isvc_pod.spec.initContainers[0].env
+
+            for env_var in init_env_vars:
+                if env_var.name == "HTTP_PROXY":
+                    http_proxy_env = env_var.value
+                elif env_var.name == "HTTPS_PROXY":
+                    https_proxy_env = env_var.value
+                elif env_var.name == "NO_PROXY":
+                    no_proxy_env = env_var.value
+
+            assert http_proxy_env == test_http_proxy
+            assert https_proxy_env == test_https_proxy
+            assert no_proxy_env == test_no_proxy
+
+
 async def test_blocked_on_invalid_config(ops_test: OpsTest):
     """
     Test whether the application is blocked on providing an invalid configuration.