Reproduce bug 1981813 in func env

There stable/yoga only change in test_pci_sriov_servers.py due to
unittest.mock switch[1] only happened in zed.

[1] https://review.opendev.org/q/topic:unittest.mock+status:merged+project:openstack/nova

Related-Bug: #1981813
Change-Id: I9367b7ed475917bdb05eb3f209ce1a4e646534e2
(cherry picked from commit f8c91eb75fc5504a37fc3b4be1d65d33dbc9b511)

2 files changed, 80 insertions, 3 deletions

diff --git a/nova/tests/fixtures/libvirt.py b/nova/tests/fixtures/libvirt.py
index 0684bae7dd..5ccf01e40f 100644
--- a/nova/tests/fixtures/libvirt.py
+++ b/nova/tests/fixtures/libvirt.py
@@ -2225,9 +2225,12 @@ class LibvirtFixture(fixtures.Fixture):
 
         # libvirt driver needs to call out to the filesystem to get the
         # parent_ifname for the SRIOV VFs.
-        self.useFixture(fixtures.MockPatch(
-            'nova.pci.utils.get_ifname_by_pci_address',
-            return_value='fake_pf_interface_name'))
+        self.mock_get_ifname_by_pci_address = self.useFixture(
+            fixtures.MockPatch(
+                "nova.pci.utils.get_ifname_by_pci_address",
+                return_value="fake_pf_interface_name",
+            )
+        ).mock
 
         self.useFixture(fixtures.MockPatch(
             'nova.pci.utils.get_mac_by_pci_address',
diff --git a/nova/tests/functional/libvirt/test_pci_sriov_servers.py b/nova/tests/functional/libvirt/test_pci_sriov_servers.py
index c228fb04cf..c1618751a9 100644
--- a/nova/tests/functional/libvirt/test_pci_sriov_servers.py
+++ b/nova/tests/functional/libvirt/test_pci_sriov_servers.py
@@ -28,6 +28,7 @@ from oslo_utils import units
 
 import nova
 from nova import context
+from nova import exception
 from nova.network import constants
 from nova import objects
 from nova.objects import fields
@@ -951,6 +952,79 @@ class SRIOVServersTest(_PCIServersWithMigrationTestBase):
             ],
         )
 
+    def test_change_bound_port_vnic_type_kills_compute_at_restart(self):
+        """Create a server with a direct port and change the vnic_type of the
+        bound port to macvtap. Then restart the compute service.
+
+        As the vnic_type is changed on the port but the vif_type is hwveb
+        instead of macvtap the vif plug logic will try to look up the netdev
+        of the parent VF. Howvere that VF consumed by the instance so the
+        netdev does not exists. This causes that the compute service will fail
+        with an exception during startup
+        """
+        pci_info = fakelibvirt.HostPCIDevicesInfo(num_pfs=1, num_vfs=2)
+        self.start_compute(pci_info=pci_info)
+
+        # create a direct port
+        port = self.neutron.network_4_port_1
+        self.neutron.create_port({'port': port})
+
+        # create a server using the VF via neutron
+        server = self._create_server(networks=[{'port': port['id']}])
+
+        # update the vnic_type of the port in neutron
+        port = copy.deepcopy(port)
+        port['binding:vnic_type'] = 'macvtap'
+        self.neutron.update_port(port['id'], {"port": port})
+
+        compute = self.computes['compute1']
+
+        # Force an update on the instance info cache to ensure nova gets the
+        # information about the updated port
+        with context.target_cell(
+            context.get_admin_context(),
+            self.host_mappings['compute1'].cell_mapping
+        ) as cctxt:
+            compute.manager._heal_instance_info_cache(cctxt)
+
+        def fake_get_ifname_by_pci_address(pci_addr: str, pf_interface=False):
+            # we want to fail the netdev lookup only if the pci_address is
+            # already consumed by our instance. So we look into the instance
+            # definition to see if the device is attached to the instance as VF
+            conn = compute.manager.driver._host.get_connection()
+            dom = conn.lookupByUUIDString(server['id'])
+            dev = dom._def['devices']['nics'][0]
+            lookup_addr = pci_addr.replace(':', '_').replace('.', '_')
+            if (
+                dev['type'] == 'hostdev' and
+                dev['source'] == 'pci_' + lookup_addr
+            ):
+                # nova tried to look up the netdev of an already consumed VF.
+                # So we have to fail
+                raise exception.PciDeviceNotFoundById(id=pci_addr)
+
+        # We need to simulate the actual failure manually as in our functional
+        # environment all the PCI lookup is mocked. In reality nova tries to
+        # look up the netdev of the pci device on the host used by the port as
+        # the parent of the macvtap. However, as the originally direct port is
+        # bound to the instance, the VF pci device is already consumed by the
+        # instance and therefore there is no netdev for the VF.
+        with mock.patch(
+            'nova.pci.utils.get_ifname_by_pci_address',
+            side_effect=fake_get_ifname_by_pci_address,
+        ):
+            # This is bug 1981813 as the compute service fails to start with an
+            # exception.
+            # Nova cannot prevent the vnic_type change on a bound port. Neutron
+            # should prevent that instead. But the nova-compute should still
+            # be able to start up and only log an ERROR for this instance in
+            # inconsistent state.
+            self.assertRaises(
+                exception.PciDeviceNotFoundById,
+                self.restart_compute_service,
+                'compute1',
+            )
+
 
 class SRIOVAttachDetachTest(_PCIServersTestBase):
     # no need for aliases as these test will request SRIOV via neutron