Merge "add semantic validation for label "intelgpu:enabled"" into r/stx.3.0

sysinv/sysinv/sysinv/sysinv/api/controllers/v1/label.py

@@ -3,6 +3,8 @@
 # SPDX-License-Identifier: Apache-2.0
 #
 
+import json
+import os
 import pecan
 from pecan import rest
 import wsme
@@ -204,6 +206,8 @@ class LabelController(rest.RestController):
 
         _semantic_check_worker_labels(body)
 
+        _semantic_check_k8s_plugins_labels(host, body)
+
         existing_labels = {}
         for label_key in body.keys():
             label = None
@@ -348,3 +352,35 @@ def _semantic_check_worker_labels(body):
                 raise wsme.exc.ClientSideError(
                     _(
                         "Invalid value for %s label." % constants.KUBE_CPU_MANAGER_LABEL))
+
+
+def _get_system_enabled_k8s_plugins():
+    if not os.path.isfile(constants.ENABLED_KUBE_PLUGINS):
+        return None
+
+    with open(constants.ENABLED_KUBE_PLUGINS) as f:
+        return json.loads(f.read())
+
+
+def _semantic_check_intel_gpu_plugins_labels(host):
+    pci_devices = pecan.request.dbapi.pci_device_get_by_host(host.id)
+    for pci_device in pci_devices:
+        if ("VGA" in pci_device.pclass and pci_device.driver == "i915"):
+            return
+
+    raise wsme.exc.ClientSideError("Host %s does not support Intel GPU device plugin." % (host.hostname))
+
+
+def _semantic_check_k8s_plugins_labels(host, body):
+    """
+    Perform hardware checks to ensure k8s plugins labels are valid on particular node.
+    """
+    plugins = _get_system_enabled_k8s_plugins()
+    if plugins is None:
+        return
+
+    for label_key, label_value in body.items():
+        label = label_key + "=" + label_value
+        if label in plugins.values():
+            if label == constants.KUBE_INTEL_GPU_DEVICE_PLUGIN_LABEL:
+                _semantic_check_intel_gpu_plugins_labels(host)

sysinv/sysinv/sysinv/sysinv/common/constants.py

@@ -1549,3 +1549,7 @@ HOST_BM_VALID_TYPE_LIST = [HOST_BM_TYPE_DEPROVISIONED,
 HOST_BM_VALID_PROVISIONED_TYPE_LIST = [HOST_BM_TYPE_DYNAMIC,
                                        HOST_BM_TYPE_IPMI,
                                        HOST_BM_TYPE_REDFISH]
+# K8s device plugins
+DEVICE_PLUGINS_FILE = "enabled_kube_plugins"
+ENABLED_KUBE_PLUGINS = os.path.join(tsc.CONFIG_PATH, DEVICE_PLUGINS_FILE)
+KUBE_INTEL_GPU_DEVICE_PLUGIN_LABEL = "intelgpu=enabled"

sysinv/sysinv/sysinv/sysinv/tests/api/test_label.py

@@ -3,6 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 #
 
+import mock
 import platform
 
 from six.moves import http_client
@@ -157,3 +158,46 @@ class LabelAssignTestCase(LabelTestCase):
             'kube-topology-mgr-policy': 'invalid',
         }
         self.assign_labels_failure(host_uuid, topology_mgr_label)
+
+    def mock_get_system_enabled_k8s_plugins_return_plugins():
+        return {"intel-gpu-plugin": "intelgpu=enabled",
+                "intel-qat-plugin": "intelqat=enabled"}
+
+    def mock_get_system_enabled_k8s_plugins_return_none():
+        return None
+
+    @mock.patch('sysinv.api.controllers.v1.label._get_system_enabled_k8s_plugins',
+                mock_get_system_enabled_k8s_plugins_return_plugins)
+    def test_create_plugin_labels_on_supported_node(self):
+        dbutils.create_test_pci_devices(
+            host_id=self.worker.id,
+            pclass='VGA compatible controller',
+            driver='i915',)
+
+        test_plugin_label = {'intelgpu': 'enabled', }
+        self.assign_labels(self.worker.uuid, test_plugin_label)
+
+        response_data = self.get_host_labels(self.worker.uuid)
+        self.validate_labels(test_plugin_label, response_data)
+
+    @mock.patch('sysinv.api.controllers.v1.label._get_system_enabled_k8s_plugins',
+                mock_get_system_enabled_k8s_plugins_return_plugins)
+    def test_create_plugin_labels_on_unsupported_node(self):
+        dbutils.create_test_pci_devices(
+            host_id=self.worker.id,
+            pclass='VGA compatible controller',
+            driver='',)
+
+        test_plugin_label = {'intelgpu': 'enabled', }
+
+        self.assign_labels_failure(self.worker.uuid, test_plugin_label)
+
+    @mock.patch('sysinv.api.controllers.v1.label._get_system_enabled_k8s_plugins',
+                mock_get_system_enabled_k8s_plugins_return_none)
+    def test_create_plugin_labels_on_non_plugin_system(self):
+        test_plugin_label = {'intelgpu': 'enabled', }
+
+        self.assign_labels(self.worker.uuid, test_plugin_label)
+
+        response_data = self.get_host_labels(self.worker.uuid)
+        self.validate_labels(test_plugin_label, response_data)