RHOAIENG-32532: Fix broken E2E tests

Author: kryanbeane

.github/workflows/rayjob_e2e_tests.yaml

@@ -7,14 +7,10 @@
 from ..kubernetes_cluster import config_check, _kube_api_error_handling
 
 
-def get_current_namespace():
+def get_current_namespace():  # pragma: no cover
     """
     Retrieves the current Kubernetes namespace.
 
-    This function attempts to detect the current namespace by:
-    1. First checking if running inside a pod (reading from service account namespace file)
-    2. Falling back to reading from the current kubeconfig context
-
     Returns:
         str:
             The current namespace or None if not found.

src/codeflare_sdk/common/utils/k8s_utils.py

@@ -136,7 +136,6 @@ def build_ray_cluster(cluster: "codeflare_sdk.ray.cluster.Cluster"):
                 "enableIngress": False,
                 "rayStartParams": {
                     "dashboard-host": "0.0.0.0",
-                    "dashboard-port": "8265",
                     "block": "true",
                     "num-gpus": str(head_gpu_count),
                     "resources": head_resources,

src/codeflare_sdk/ray/cluster/build_ray_cluster.py

@@ -208,6 +208,10 @@ def apply(self, force=False):
         self._throw_for_no_raycluster()
         namespace = self.config.namespace
         name = self.config.name
+
+        # Regenerate resource_yaml to reflect any configuration changes
+        self.resource_yaml = self.create_resource()
+
         try:
             self.config_check()
             api_instance = client.CustomObjectsApi(get_api_client())
@@ -387,16 +391,24 @@ def is_dashboard_ready(self) -> bool:
             bool:
                 True if the dashboard is ready, False otherwise.
         """
+        dashboard_uri = self.cluster_dashboard_uri()
+        if dashboard_uri is None:
+            return False
+
         try:
             response = requests.get(
-                self.cluster_dashboard_uri(),
+                dashboard_uri,
                 headers=self._client_headers,
                 timeout=5,
                 verify=self._client_verify_tls,
             )
         except requests.exceptions.SSLError:  # pragma no cover
             # SSL exception occurs when oauth ingress has been created but cluster is not up
             return False
+        except Exception:  # pragma no cover
+            # Any other exception (connection errors, timeouts, etc.)
+            return False
+
         if response.status_code == 200:
             return True
         else:
@@ -523,7 +535,24 @@ def cluster_dashboard_uri(self) -> str:
                     elif "route.openshift.io/termination" in annotations:
                         protocol = "https"
                 return f"{protocol}://{ingress.spec.rules[0].host}"
-        return "Dashboard not available yet, have you run cluster.up()?"
+
+        # For local/test environments without ingress controller (e.g., KIND)
+        # Try to find the Ray head service
+        try:
+            api_instance = client.CoreV1Api(get_api_client())
+            services = api_instance.list_namespaced_service(
+                self.config.namespace,
+                label_selector=f"ray.io/cluster={self.config.name},ray.io/node-type=head",
+            )
+            for service in services.items:
+                if service.metadata.name == f"{self.config.name}-head-svc":
+                    # For ClusterIP services in local environments, return a placeholder
+                    # The actual connection would need port-forwarding or NodePort
+                    return f"http://{service.metadata.name}.{self.config.namespace}.svc.cluster.local:8265"
+        except Exception:  # pragma: no cover
+            pass
+
+        return None
 
     def list_jobs(self) -> List:
         """
@@ -783,6 +812,12 @@ def remove_autogenerated_fields(resource):
                 del resource[key]
             else:
                 remove_autogenerated_fields(resource[key])
+
+        # After cleaning, remove empty metadata sections
+        if "metadata" in resource and isinstance(resource["metadata"], dict):
+            if len(resource["metadata"]) == 0:
+                del resource["metadata"]
+
     elif isinstance(resource, list):
         for item in resource:
             remove_autogenerated_fields(item)

src/codeflare_sdk/ray/cluster/cluster.py

@@ -322,10 +322,8 @@ def test_cluster_uris(mocker):
     mocker.patch(
         "kubernetes.client.NetworkingV1Api.list_namespaced_ingress",
     )
-    assert (
-        cluster.cluster_dashboard_uri()
-        == "Dashboard not available yet, have you run cluster.up()?"
-    )
+    # When no ingress/route/service is found, the method should return None
+    assert cluster.cluster_dashboard_uri() is None
 
     mocker.patch(
         "codeflare_sdk.ray.cluster.cluster._is_openshift_cluster", return_value=True

src/codeflare_sdk/ray/cluster/test_cluster.py

@@ -1,4 +1,4 @@
-# Copyright 2022-2025 IBM, Red Hat
+# Copyright 2024 IBM, Red Hat
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

src/codeflare_sdk/ray/cluster/test_config.py

@@ -284,7 +284,6 @@ def _build_head_ray_params(self) -> Dict[str, str]:
         """Build Ray start parameters for head node."""
         params = {
             "dashboard-host": "0.0.0.0",
-            "dashboard-port": "8265",
             "block": "true",
         }