Merge "Validate [pci]alias at service startup" into stable/2025.1

Author: Zuul

nova/api/openstack/wsgi_app.py

@@ -26,6 +26,7 @@
 from nova import context
 from nova import exception
 from nova import objects
+from nova.pci import request
 from nova import service
 from nova import utils
 from nova import version
@@ -51,6 +52,11 @@ def _get_config_files(env=None):
 
 
 def _setup_service(host, name):
+
+    # NOTE(gibi): validate the [pci]alias config early to avoid late failures
+    # at instance creation due to config errors.
+    request.get_alias_from_config()
+
     try:
         utils.raise_if_old_compute()
     except exception.TooOldComputeService as e:

nova/compute/manager.py

@@ -1615,6 +1615,10 @@ def init_host(self, service_ref):
             # if the configuration is wrong.
             whitelist.Whitelist(CONF.pci.device_spec)
 
+        # NOTE(gibi): validate the [pci]alias config early to avoid late
+        # failures at instance lifecycle operations due to config errors.
+        pci_req_module.get_alias_from_config()
+
         nova.conf.neutron.register_dynamic_opts(CONF)
         # Even if only libvirt uses them, make it available for all drivers
         nova.conf.devices.register_dynamic_opts(CONF)

nova/pci/request.py

@@ -173,7 +173,7 @@ def _validate_aliases(aliases):
     _validate_required_ids(aliases)
 
 
-def _get_alias_from_config() -> Alias:
+def get_alias_from_config() -> Alias:
     """Parse and validate PCI aliases from the nova config.
 
     :returns: A dictionary where the keys are alias names and the values are
@@ -237,7 +237,7 @@ def _translate_alias_to_requests(
     alias_spec: str, affinity_policy: ty.Optional[str] = None,
 ) -> ty.List['objects.InstancePCIRequest']:
     """Generate complete pci requests from pci aliases in extra_spec."""
-    pci_aliases = _get_alias_from_config()
+    pci_aliases = get_alias_from_config()
 
     pci_requests: ty.List[objects.InstancePCIRequest] = []
     for name, count in [spec.split(':') for spec in alias_spec.split(',')]:

nova/tests/unit/api/openstack/test_wsgi_app.py

@@ -15,6 +15,7 @@
 
 import fixtures
 from oslo_config import fixture as config_fixture
+from oslo_serialization import jsonutils
 from oslotest import base
 
 from nova.api.openstack import wsgi_app
@@ -127,6 +128,14 @@ def test_setup_service_version_workaround(self, mock_check_old, mock_get):
             group='workarounds')
         wsgi_app._setup_service('myhost', 'api')
 
+    def test_setup_service_pci_alias_validation(self):
+        wsgi_app.CONF.set_override(
+            'alias', jsonutils.dumps({'name': 'foo'}),
+            group='pci')
+        self.assertRaises(
+            exception.PciInvalidAlias,
+            wsgi_app._setup_service, 'myhost', 'api')
+
     def test__get_config_files_empty_env(self):
         env = {}
         result = wsgi_app._get_config_files(env)

nova/tests/unit/compute/test_compute_mgr.py

@@ -7025,6 +7025,18 @@ def test_init_host_pci_device_spec_validation_failure(self):
         self.assertRaises(exception.PciDeviceInvalidDeviceName,
                           self.compute.init_host, None)
 
+    def test_init_host_pci_alias_validation_failure(self):
+        # Tests that we fail init_host if the pci.alias is
+        # configured incorrectly.
+        self.flags(
+            alias=[
+                jsonutils.dumps({'name': 'foo'})
+            ],
+            group='pci'
+        )
+        self.assertRaises(
+            exception.PciInvalidAlias, self.compute.init_host, None)
+
     @mock.patch('nova.compute.manager.ComputeManager._instance_update')
     def test_error_out_instance_on_exception_not_implemented_err(self,
                                                         inst_update_mock):

nova/tests/unit/pci/test_request.py

@@ -81,7 +81,7 @@ def setUp(self):
 
     def test_get_alias_from_config_valid(self):
         self.flags(alias=[_fake_alias1], group='pci')
-        result = request._get_alias_from_config()
+        result = request.get_alias_from_config()
         expected_result = (
             'legacy',
             [{
@@ -103,7 +103,7 @@ def test_get_alias_from_config_valid_multispec(self):
         })
 
         self.flags(alias=[_fake_alias1, _fake_alias], group='pci')
-        result = request._get_alias_from_config()
+        result = request.get_alias_from_config()
         expected_result = (
             'legacy',
             [{
@@ -133,7 +133,7 @@ def test_get_alias_from_config_multispec_rejected_pci_in_placement(self):
         self.flags(alias=[_fake_alias1, _fake_alias], group='pci')
 
         ex = self.assertRaises(
-            exception.PciInvalidAlias, request._get_alias_from_config)
+            exception.PciInvalidAlias, request.get_alias_from_config)
         self.assertEqual(
             "The PCI alias(es) QuickAssist have multiple specs but "
             "[filter_scheduler]pci_in_placement is True. The PCI in Placement "
@@ -146,7 +146,7 @@ def _test_get_alias_from_config_invalid(self, alias):
         self.flags(alias=[alias], group='pci')
         self.assertRaises(
             exception.PciInvalidAlias,
-            request._get_alias_from_config)
+            request.get_alias_from_config)
 
     def test_get_alias_from_config_invalid_device_type(self):
         fake_alias = jsonutils.dumps({
@@ -215,7 +215,7 @@ def test_get_alias_from_config_valid_numa_policy(self):
                 "numa_policy": policy,
             })
             self.flags(alias=[fake_alias], group='pci')
-            aliases = request._get_alias_from_config()
+            aliases = request.get_alias_from_config()
             self.assertIsNotNone(aliases)
             self.assertIn("xxx", aliases)
             self.assertEqual(policy, aliases["xxx"][0])
@@ -228,7 +228,7 @@ def test_get_alias_from_config_valid_rc_and_traits(self):
         })
         self.flags(pci_in_placement=True, group='filter_scheduler')
         self.flags(alias=[fake_alias], group='pci')
-        aliases = request._get_alias_from_config()
+        aliases = request.get_alias_from_config()
         self.assertIsNotNone(aliases)
         self.assertIn("xxx", aliases)
         self.assertEqual(
@@ -256,7 +256,7 @@ def test_get_alias_from_config_conflicting_device_type(self):
         self.flags(alias=[fake_alias_a, fake_alias_b], group='pci')
         self.assertRaises(
             exception.PciInvalidAlias,
-            request._get_alias_from_config)
+            request.get_alias_from_config)
 
     def test_get_alias_from_config_conflicting_numa_policy(self):
         """Check behavior when numa_policy conflicts occur."""
@@ -277,7 +277,7 @@ def test_get_alias_from_config_conflicting_numa_policy(self):
         self.flags(alias=[fake_alias_a, fake_alias_b], group='pci')
         self.assertRaises(
             exception.PciInvalidAlias,
-            request._get_alias_from_config)
+            request.get_alias_from_config)
 
     def test_get_alias_from_config_missing_ids(self):
         a1 = jsonutils.dumps({
@@ -304,7 +304,7 @@ def test_get_alias_from_config_missing_ids(self):
         self.flags(alias=[a1, a2, a3, a4, a5], group='pci')
 
         ex = self.assertRaises(
-            exception.PciInvalidAlias, request._get_alias_from_config)
+            exception.PciInvalidAlias, request.get_alias_from_config)
         self.assertEqual(
             "The PCI alias(es) a1,a2,a3,a4 does not have vendor_id and "
             "product_id fields set.",
@@ -336,7 +336,7 @@ def test_get_alias_from_config_missing_ids_or_rc_pci_in_placement(self):
         self.flags(alias=[a1, a2, a3, a4, a5], group='pci')
 
         ex = self.assertRaises(
-            exception.PciInvalidAlias, request._get_alias_from_config)
+            exception.PciInvalidAlias, request.get_alias_from_config)
         self.assertEqual(
             "The PCI alias(es) a1,a2,a3 does not have vendor_id and "
             "product_id fields set or resource_class field set.",