Merge pull request #106 from stackhpc/upstream/2023.1-2024-09-09

Synchronise 2023.1 with upstream

nova/tests/unit/virt/libvirt/test_driver.py

@@ -894,6 +894,16 @@ def test_driver_capabilities_secure_boot(self, mock_supports):
         )
         mock_supports.assert_called_once_with()
 
+    @mock.patch.object(hardware, 'get_cpu_dedicated_set',
+                       return_value=set([0, 42, 1337]))
+    @mock.patch.object(libvirt_driver.LibvirtDriver,
+                       '_register_all_undefined_instance_details')
+    def test_init_host_topology(self, mock_get_cpu_dedicated_set, _):
+        driver = libvirt_driver.LibvirtDriver(fake.FakeVirtAPI(), False)
+        with mock.patch.object(driver.cpu_api, 'power_up') as mock_power_up:
+            driver.init_host('goat')
+            mock_power_up.assert_called_with(set([0, 42, 1337]))
+
     @mock.patch.object(
         libvirt_driver.LibvirtDriver,
         '_register_all_undefined_instance_details',

nova/tests/unit/virt/test_virt_drivers.py

@@ -21,6 +21,7 @@
 import netaddr
 import os_resource_classes as orc
 import os_vif
+from oslo_concurrency import processutils
 from oslo_log import log as logging
 from oslo_serialization import jsonutils
 from oslo_utils import importutils
@@ -239,8 +240,16 @@ def test_snapshot_not_running(self):
     def test_snapshot_running(self):
         img_ref = self.image_service.create(self.ctxt, {'name': 'snap-1'})
         instance_ref, network_info = self._get_running_instance()
-        self.connection.snapshot(self.ctxt, instance_ref, img_ref['id'],
-                                 lambda *args, **kwargs: None)
+        # this test depends on qemu-img
+        # being installed and in the path,
+        # if it is not installed, skip
+        try:
+            self.connection.snapshot(self.ctxt, instance_ref, img_ref['id'],
+                                     lambda *args, **kwargs: None)
+        except processutils.ProcessExecutionError as e:
+            if 'qemu-img' in e.stderr and 'No such file' in e.stderr:
+                self.skipTest("qemu-img not installed")
+            raise e
 
     @catch_notimplementederror
     def test_reboot(self):

nova/virt/libvirt/cpu/api.py

@@ -91,7 +91,7 @@ def core(self, i):
         """
         return Core(i)
 
-    def _power_up(self, cpus: ty.Set[int]) -> None:
+    def power_up(self, cpus: ty.Set[int]) -> None:
         if not CONF.libvirt.cpu_power_management:
             return
         cpu_dedicated_set = hardware.get_cpu_dedicated_set_nozero() or set()
@@ -111,7 +111,7 @@ def power_up_for_instance(self, instance: objects.Instance) -> None:
             return
         pcpus = instance.numa_topology.cpu_pinning.union(
             instance.numa_topology.cpuset_reserved)
-        self._power_up(pcpus)
+        self.power_up(pcpus)
 
     def power_up_for_migration(
         self, dst_numa_info: objects.LibvirtLiveMigrateNUMAInfo
@@ -121,7 +121,7 @@ def power_up_for_migration(
             pcpus = dst_numa_info.emulator_pins
         for pins in dst_numa_info.cpu_pins.values():
             pcpus = pcpus.union(pins)
-        self._power_up(pcpus)
+        self.power_up(pcpus)
 
     def _power_down(self, cpus: ty.Set[int]) -> None:
         if not CONF.libvirt.cpu_power_management:

nova/virt/libvirt/driver.py

@@ -735,12 +735,31 @@ def _handle_conn_event(self, enabled, reason):
                  {'enabled': enabled, 'reason': reason})
         self._set_host_enabled(enabled, reason)
 
+    def _init_host_topology(self):
+        """To work around a bug in libvirt that reports offline CPUs as always
+        being on socket 0 regardless of their real socket, power up all
+        dedicated CPUs (the only ones whose socket we actually care about),
+        then call get_capabilities() to initialize the topology with the
+        correct socket values. get_capabilities()'s implementation will reuse
+        these initial socket value, and avoid clobbering them with 0 for
+        offline CPUs.
+        """
+        cpus = hardware.get_cpu_dedicated_set()
+        if cpus:
+            self.cpu_api.power_up(cpus)
+            self._host.get_capabilities()
+
     def init_host(self, host):
         self._host.initialize()
 
-        self._update_host_specific_capabilities()
-
+        # NOTE(artom) Do this first to make sure our first call to
+        # get_capabilities() happens with all dedicated CPUs online and caches
+        # their correct socket ID. Unused dedicated CPUs will be powered down
+        # further down in this method.
         self._check_cpu_set_configuration()
+        self._init_host_topology()
+
+        self._update_host_specific_capabilities()
 
         self._do_quality_warnings()