Allow enabling PCI tracking in Placement

This patch introduces the [pci]report_in_placement config option that is
False by default but if set to True will enable reporting of the PCI
passthrough inventories to Placement.

blueprint: pci-device-tracking-in-placement
Change-Id: I49a3dbf4c5708d2d92dedd29a9dc3ef25b6cd66c

Author: gibizer

doc/source/admin/pci-passthrough.rst

@@ -351,6 +351,14 @@ information, refer to :oslo.config:option:`the documentation <pci.alias>`.
 
 PCI tracking in Placement
 -------------------------
+.. note::
+   The feature described below are optional and disabled by default in nova
+   26.0.0. (Zed). The legacy PCI tracker code path is still supported and
+   enabled. The Placement PCI tracking can be enabled via the
+   :oslo.config:option:`pci.report_in_placement` configuration. But please note
+   that once it is enabled on a given compute host it cannot be disabled there
+   any more.
+
 Since nova 26.0.0 (Zed) PCI passthrough device inventories are tracked in
 Placement. If a PCI device exists on the hypervisor and
 matches one of the device specifications configured via

nova/compute/pci_placement_translator.py

@@ -21,6 +21,7 @@
 from oslo_utils import uuidutils
 
 from nova.compute import provider_tree
+import nova.conf
 from nova import exception
 from nova.i18n import _
 from nova.objects import fields
@@ -29,6 +30,7 @@
 from nova.pci import manager as pci_manager
 
 
+CONF = nova.conf.CONF
 LOG = logging.getLogger(__name__)
 
 
@@ -42,14 +44,7 @@
 
 
 def _is_placement_tracking_enabled() -> bool:
-    # This is false to act as a feature flag while we develop the feature
-    # step by step. It will be replaced with a config check when the feature is
-    # ready for production.
-    #
-    # return CONF.pci.report_in_placement
-
-    # Test code will mock this function to enable the feature in the test env
-    return False
+    return CONF.pci.report_in_placement
 
 
 def _normalize_traits(traits: ty.List[str]) -> ty.List[str]:

nova/conf/pci.py

@@ -225,7 +225,21 @@
 
     device_spec = [{"product_id":"0001", "vendor_id":"8086"},
                    {"product_id":"0002", "vendor_id":"8086"}]
-""")
+"""),
+    cfg.BoolOpt('report_in_placement',
+                default=False,
+                help="""
+Enable PCI resource inventory reporting to Placement. If it is enabled then the
+nova-compute service will report PCI resource inventories to Placement
+according to the [pci]device_spec configuration and the PCI devices reported
+by the hypervisor. Once it is enabled it cannot be disabled any more. In a
+future release the default of this config will be change to True.
+
+Related options:
+
+* [pci]device_spec: to define which PCI devices nova are allowed to track and
+  assign to guests.
+"""),
 ]
 
 

nova/tests/functional/libvirt/test_pci_in_placement.py

@@ -61,13 +61,7 @@ class PlacementPCIReportingTests(test_pci_sriov_servers._PCIServersTestBase):
 
     def setUp(self):
         super().setUp()
-        patcher = mock.patch(
-            "nova.compute.pci_placement_translator."
-            "_is_placement_tracking_enabled",
-            return_value=True
-        )
-        self.addCleanup(patcher.stop)
-        self.mock_pci_report_in_placement = patcher.start()
+        self.flags(group="pci", report_in_placement=True)
 
         # These tests should not depend on the host's sysfs
         self.useFixture(
@@ -718,7 +712,7 @@ def _create_one_compute_with_a_pf_consumed_by_an_instance(self):
             ]
         )
         self.flags(group='pci', device_spec=device_spec)
-        self.mock_pci_report_in_placement.return_value = True
+        self.flags(group="pci", report_in_placement=True)
         self.start_compute(hostname="compute1", pci_info=pci_info)
 
         self.assertPCIDeviceCounts("compute1", total=1, free=1)
@@ -893,7 +887,7 @@ def test_reporting_disabled_nothing_is_reported(self):
         # Disable placement reporting so even if there are PCI devices on the
         # hypervisor matching the [pci]device_spec config they are not reported
         # to Placement
-        self.mock_pci_report_in_placement.return_value = False
+        self.flags(group="pci", report_in_placement=False)
         self.start_compute(hostname="compute1", pci_info=pci_info)
 
         self.assert_placement_pci_view(
@@ -931,7 +925,7 @@ def test_reporting_cannot_be_disable_once_it_is_enabled(self):
 
         # Try to disable placement reporting. The compute will refuse to start
         # as there are already PCI device RPs in placement.
-        self.mock_pci_report_in_placement.return_value = False
+        self.flags(group="pci", report_in_placement=False)
         ex = self.assertRaises(
             exception.PlacementPciException,
             self.restart_compute_service,
@@ -988,7 +982,7 @@ def test_heal_single_pci_allocation(self):
         self.flags(group='pci', device_spec=device_spec)
 
         # Start a compute *without* PCI tracking in placement
-        self.mock_pci_report_in_placement.return_value = False
+        self.flags(group="pci", report_in_placement=False)
         self.start_compute(hostname="compute1", pci_info=pci_info)
         self.assertPCIDeviceCounts("compute1", total=1, free=1)
 
@@ -999,7 +993,7 @@ def test_heal_single_pci_allocation(self):
         self.assertPCIDeviceCounts("compute1", total=1, free=0)
 
         # Restart the compute but now with PCI tracking enabled
-        self.mock_pci_report_in_placement.return_value = True
+        self.flags(group="pci", report_in_placement=True)
         self.restart_compute_service("compute1")
         # Assert that the PCI allocation is healed in placement
         self.assertPCIDeviceCounts("compute1", total=1, free=0)
@@ -1057,7 +1051,7 @@ def test_heal_multiple_allocations(self):
         self.flags(group='pci', device_spec=device_spec)
 
         # Start a compute *without* PCI tracking in placement
-        self.mock_pci_report_in_placement.return_value = False
+        self.flags(group="pci", report_in_placement=False)
         self.start_compute(hostname="compute1", pci_info=pci_info)
         # 2 PCI + 1 PF + 4 VFs
         self.assertPCIDeviceCounts("compute1", total=7, free=7)
@@ -1082,7 +1076,7 @@ def test_heal_multiple_allocations(self):
         self.assertPCIDeviceCounts("compute1", total=7, free=1)
 
         # Restart the compute but now with PCI tracking enabled
-        self.mock_pci_report_in_placement.return_value = True
+        self.flags(group="pci", report_in_placement=True)
         self.restart_compute_service("compute1")
         # Assert that the PCI allocation is healed in placement
         self.assertPCIDeviceCounts("compute1", total=7, free=1)
@@ -1157,7 +1151,7 @@ def test_heal_partial_allocations(self):
         self.flags(group='pci', device_spec=device_spec)
 
         # Start a compute with PCI tracking in placement
-        self.mock_pci_report_in_placement.return_value = True
+        self.flags(group="pci", report_in_placement=True)
         self.start_compute(hostname="compute1", pci_info=pci_info)
         # 2 PCI + 1 PF + 4 VFs
         self.assertPCIDeviceCounts("compute1", total=7, free=7)
@@ -1234,7 +1228,7 @@ def test_heal_partial_allocations_during_resize_downsize(self):
         self.flags(group='pci', device_spec=compute1_device_spec)
 
         # Start a compute with PCI tracking in placement
-        self.mock_pci_report_in_placement.return_value = True
+        self.flags(group="pci", report_in_placement=True)
         self.start_compute(hostname="compute1", pci_info=compute1_pci_info)
         self.assertPCIDeviceCounts("compute1", total=2, free=2)
         compute1_expected_placement_view = {
@@ -1395,7 +1389,7 @@ def test_heal_partial_allocations_during_resize_change_dev_type(self):
         self.flags(group='pci', device_spec=compute1_device_spec)
 
         # Start a compute with PCI tracking in placement
-        self.mock_pci_report_in_placement.return_value = True
+        self.flags(group="pci", report_in_placement=True)
         self.start_compute(hostname="compute1", pci_info=compute1_pci_info)
         self.assertPCIDeviceCounts("compute1", total=1, free=1)
         compute1_expected_placement_view = {
@@ -1458,7 +1452,7 @@ def test_heal_partial_allocations_during_resize_change_dev_type(self):
         self.flags(group='pci', device_spec=compute2_device_spec)
 
         # Start a compute with PCI tracking in placement
-        self.mock_pci_report_in_placement.return_value = True
+        self.flags(group="pci", report_in_placement=True)
         self.start_compute(hostname="compute2", pci_info=compute2_pci_info)
         self.assertPCIDeviceCounts("compute2", total=3, free=3)
         compute2_expected_placement_view = {
@@ -1533,7 +1527,7 @@ def test_heal_allocation_during_same_host_resize(self):
         )
         self.flags(group='pci', device_spec=compute1_device_spec)
         # Start a compute with PCI tracking in placement
-        self.mock_pci_report_in_placement.return_value = True
+        self.flags(group="pci", report_in_placement=True)
         self.start_compute(hostname="compute1", pci_info=compute1_pci_info)
         self.assertPCIDeviceCounts("compute1", total=3, free=3)
         compute1_expected_placement_view = {

nova/tests/functional/libvirt/test_pci_sriov_servers.py

@@ -1742,13 +1742,7 @@ class PCIServersTest(_PCIServersTestBase):
 
     def setUp(self):
         super().setUp()
-        patcher = mock.patch(
-            "nova.compute.pci_placement_translator."
-            "_is_placement_tracking_enabled",
-            return_value=True
-        )
-        self.addCleanup(patcher.stop)
-        patcher.start()
+        self.flags(group="pci", report_in_placement=True)
 
     def test_create_server_with_pci_dev_and_numa(self):
         """Verifies that an instance can be booted with cpu pinning and with an

releasenotes/notes/bp-pci-device-tracking-in-placement-75ee1d20a57662f2.yaml

@@ -0,0 +1,9 @@
+---
+features:
+  - |
+    Nova started tracking PCI devices in Placement. This is an optional feature
+    disable by default while we are implementing inventory tracking and
+    scheduling support for both PCI passthrough devices and SR-IOV devices
+    consumed via Neutron ports. Please read our
+    `documentation <https://docs.openstack.org/nova/latest/admin/pci-passthrough.html#pci-tracking-in-placement>`_
+    for more details on what is supported how this feature can be enabled.