Move nmstate namespace check to network sanity checks and nodes_active_nics to return node_physical_nics for cloud clusters

conftest.py

@@ -93,7 +93,7 @@
 }
 NAMESPACE_COLLECTION = {
     "storage": [NamespacesNames.OPENSHIFT_STORAGE],
-    "network": ["openshift-nmstate"],
+    "network": [NamespacesNames.OPENSHIFT_NMSTATE],
     "virt": [],
 }
 MUST_GATHER_IGNORE_EXCEPTION_LIST = [
@@ -839,7 +839,7 @@ def get_inspect_command_namespace_string(node: Node, test_name: str) -> str:
         LOGGER.warning(f"{test_name} does not require special data collection on failure")
     else:
         component = components[0]
-        namespaces_to_collect: list[str] = NAMESPACE_COLLECTION[component]
+        namespaces_to_collect: list[str] = NAMESPACE_COLLECTION[component].copy()
         if component == "virt":
             all_markers = get_all_node_markers(node=node)
             if "gpu" in all_markers:
@@ -863,7 +863,6 @@ def pytest_exception_interact(node: Item | Collector, call: CallInfo[Any], repor
     if node.config.getoption("--data-collector") and not is_skip_must_gather(node=node):
         test_name = f"{node.fspath}::{node.name}"
         LOGGER.info(f"Must-gather collection is enabled for {test_name}.")
-        inspect_str = get_inspect_command_namespace_string(test_name=test_name, node=node)
         if call.excinfo and any([
             isinstance(call.excinfo.value, exception_type) for exception_type in MUST_GATHER_IGNORE_EXCEPTION_LIST
         ]):
@@ -882,7 +881,7 @@ def pytest_exception_interact(node: Item | Collector, call: CallInfo[Any], repor
                     since_time=calculate_must_gather_timer(test_start_time=test_start_time),
                     target_dir=collection_dir,
                 )
-                if inspect_str:
+                if inspect_str := get_inspect_command_namespace_string(test_name=test_name, node=node):
                     target_dir = os.path.join(collection_dir, "inspect_collection")
                     inspect_command = (
                         f"{INSPECT_BASE_COMMAND} {inspect_str} "

tests/after_cluster_deploy_sanity/test_after_cluster_deploy_sanity.py

@@ -64,12 +64,11 @@ def test_node_sanity(admin_client, nodes):
 
 
 @pytest.mark.cluster_health_check
-def test_pod_sanity(admin_client, hco_namespace, nmstate_namespace):
-    for namespace_obj in [hco_namespace, nmstate_namespace]:
-        wait_for_pods_running(
-            admin_client=admin_client,
-            namespace=namespace_obj,
-        )
+def test_pod_sanity(admin_client, hco_namespace):
+    wait_for_pods_running(
+        admin_client=admin_client,
+        namespace=hco_namespace,
+    )
 
 
 @pytest.mark.cluster_health_check

tests/conftest.py

@@ -590,6 +590,7 @@ def nodes_active_nics(
     workers_utility_pods,
     node_physical_nics,
 ):
+    # TODO: Add support for environments that do not have KNMstate installed. e.g: clouds
     # TODO: Reduce cognitive complexity
     def _bridge_ports(node_interface):
         ports = set()
@@ -2860,11 +2861,13 @@ def machine_config_pools():
 
 
 @pytest.fixture(scope="session")
-def nmstate_namespace(admin_client, nmstate_required):
-    if nmstate_required:
-        return Namespace(client=admin_client, name="openshift-nmstate", ensure_exists=True)
+def nmstate_namespace(admin_client):
+    try:
+        return Namespace(client=admin_client, name=NamespacesNames.OPENSHIFT_NMSTATE, ensure_exists=True)
 
-    return None
+    except ResourceNotFoundError:
+        LOGGER.info(f"Namespace '{NamespacesNames.OPENSHIFT_NMSTATE}' not found.")
+        return None
 
 
 @pytest.fixture()
@@ -2890,11 +2893,6 @@ def smbios_from_kubevirt_config(kubevirt_config_scope_module):
     return kubevirt_config_scope_module["smbios"]
 
 
-@pytest.fixture(scope="session")
-def nmstate_required(admin_client):
-    return get_cluster_platform(admin_client=admin_client) in ("BareMetal", "OpenStack")
-
-
 # TODO: Replace this fixture with py_config.get("conformance_tests")
 @pytest.fixture(scope="session")
 def conformance_tests(request):

tests/network/conftest.py

@@ -14,6 +14,7 @@
 from ocp_resources.performance_profile import PerformanceProfile
 from ocp_resources.pod import Pod
 from pytest_testconfig import config as py_config
+from timeout_sampler import TimeoutExpiredError
 
 from tests.network.constants import BRCNV
 from tests.network.utils import get_vlan_index_number, vm_for_brcnv_tests
@@ -25,12 +26,14 @@
     ISTIO_SYSTEM_DEFAULT_NS,
     OVS_BRIDGE,
     VIRT_HANDLER,
+    NamespacesNames,
 )
 from utilities.infra import (
     ExecCommandOnPod,
     exit_pytest_execution,
     get_deployment_by_name,
     get_node_selector_dict,
+    wait_for_pods_running,
 )
 from utilities.network import (
     get_cluster_cni_type,
@@ -216,6 +219,7 @@ def network_operator():
 
 @pytest.fixture(scope="session", autouse=True)
 def network_sanity(
+    admin_client,
     hosts_common_available_ports,
     junitxml_plugin,
     request,
@@ -226,6 +230,7 @@ def network_sanity(
     ipv4_supported_cluster,
     ipv6_supported_cluster,
     conformance_tests,
+    nmstate_namespace,
 ):
     """
     Ensures the test cluster meets network requirements before executing tests.
@@ -235,8 +240,9 @@ def network_sanity(
     failure_msgs = []
     collected_tests = request.session.items
 
-    def _verify_multi_nic(request=request):
-        marker_args = request.config.getoption("-m")
+    def _verify_multi_nic(_request):
+        marker_args = _request.config.getoption("-m")
+        # TODO: add multi_nic marker to tests that require multiple NICs
         if marker_args and "single_nic" in marker_args and "not single_nic" not in marker_args:
             LOGGER.info("Running only single-NIC network cases, no need to verify multi NIC support")
             return
@@ -339,14 +345,30 @@ def _verify_bgp_env_vars():
                 failure_msgs.append(f"BGP tests require the following environment variables: {missing_env_vars}")
                 return
 
-    _verify_multi_nic(request=request)
+    def _verify_nmstate_running_pods(_admin_client, namespace):
+        # TODO: Only test if nmstate is required by the test(s)
+        if not namespace:
+            failure_msgs.append(f"Knmstate namespace '{NamespacesNames.OPENSHIFT_NMSTATE}' does not exist.")
+            return
+
+        LOGGER.info("Verifying all pods in nmstate namespace are running")
+        try:
+            wait_for_pods_running(
+                admin_client=_admin_client,
+                namespace=namespace,
+            )
+        except TimeoutExpiredError:
+            failure_msgs.append(f"Some pods are not running in nmstate namespace '{namespace.name}'")
+
+    _verify_multi_nic(_request=request)
     _verify_dpdk()
     _verify_service_mesh()
     _verify_jumbo_frame()
     _verify_sriov()
     _verify_ip_family(family="ipv4", is_supported_in_cluster=ipv4_supported_cluster)
     _verify_ip_family(family="ipv6", is_supported_in_cluster=ipv6_supported_cluster)
     _verify_bgp_env_vars()
+    _verify_nmstate_running_pods(_admin_client=admin_client, namespace=nmstate_namespace)
 
     if failure_msgs:
         err_msg = "\n".join(failure_msgs)

utilities/constants.py

@@ -700,6 +700,7 @@ class NamespacesNames:
     OPENSHIFT_VIRTUALIZATION_OS_IMAGES = "openshift-virtualization-os-images"
     WASP = "wasp"
     OPENSHIFT_KUBE_DESCHEDULER_OPERATOR = "openshift-kube-descheduler-operator"
+    OPENSHIFT_NMSTATE = "openshift-nmstate"
 
 
 # CNV supplemental-templates