From c59224d715a21998f40f72cf4e37efdc990e4d7e Mon Sep 17 00:00:00 2001
From: Balazs Gibizer <balazs.gibizer@est.tech>
Date: Mon, 6 Dec 2021 16:36:41 +0100
Subject: [PATCH] Reproduce bug 1953359

This patch adds a functional test that reproduces a race between
incoming migration and the update_available_resource periodic

Change-Id: I4be429c56aaa15ee12f448978c38214e741eae63
Related-Bug: #1953359
---
 nova/tests/functional/integrated_helpers.py   | 10 ++-
 .../functional/libvirt/test_numa_servers.py   | 81 +++++++++++++++++++
 2 files changed, 90 insertions(+), 1 deletion(-)

diff --git a/nova/tests/functional/integrated_helpers.py b/nova/tests/functional/integrated_helpers.py
index 8710ecd768..70918bc5f5 100644
--- a/nova/tests/functional/integrated_helpers.py
+++ b/nova/tests/functional/integrated_helpers.py
@@ -441,7 +441,15 @@ class InstanceHelperMixin:
         """
         # if forcing the server onto a host, we have to use the admin API
         if not api:
-            api = self.api if not az else getattr(self, 'admin_api', self.api)
+            api = self.api if not az and not host else getattr(
+                self, 'admin_api', self.api)
+
+        if host and not api.microversion:
+            api.microversion = '2.74'
+            # with 2.74 networks param needs to use 'none' instead of None
+            # if no network is needed
+            if networks is None:
+                networks = 'none'
 
         body = self._build_server(
             name, image_uuid, flavor_id, networks, az, host)
diff --git a/nova/tests/functional/libvirt/test_numa_servers.py b/nova/tests/functional/libvirt/test_numa_servers.py
index 3bdf96e93f..2736214e64 100644
--- a/nova/tests/functional/libvirt/test_numa_servers.py
+++ b/nova/tests/functional/libvirt/test_numa_servers.py
@@ -929,6 +929,87 @@ class NUMAServersTest(NUMAServersTestBase):
         self._assert_pinned_cpus(src_host, 2)
         self._assert_pinned_cpus(dst_host, 0)
 
+    def test_resize_dedicated_policy_race_on_dest_bug_1953359(self):
+
+        self.flags(cpu_dedicated_set='0-2', cpu_shared_set=None,
+                   group='compute')
+        self.flags(vcpu_pin_set=None)
+
+        host_info = fakelibvirt.HostInfo(cpu_nodes=1, cpu_sockets=1,
+                                         cpu_cores=2, cpu_threads=1)
+        self.start_compute(host_info=host_info, hostname='compute1')
+
+        extra_spec = {
+            'hw:cpu_policy': 'dedicated',
+        }
+        flavor_id = self._create_flavor(vcpu=1, extra_spec=extra_spec)
+        expected_usage = {'DISK_GB': 20, 'MEMORY_MB': 2048, 'PCPU': 1}
+
+        server = self._run_build_test(flavor_id, expected_usage=expected_usage)
+
+        inst = objects.Instance.get_by_uuid(self.ctxt, server['id'])
+        self.assertEqual(1, len(inst.numa_topology.cells))
+        # assert that the pcpu 0 is used on compute1
+        self.assertEqual({'0': 0}, inst.numa_topology.cells[0].cpu_pinning_raw)
+
+        # start another compute with the same config
+        self.start_compute(host_info=host_info, hostname='compute2')
+
+        # boot another instance but now on compute2 so that it occupies the
+        # pcpu 0 on compute2
+        # NOTE(gibi): _run_build_test cannot be used here as it assumes only
+        # compute1 exists
+        server2 = self._create_server(
+            flavor_id=flavor_id,
+            host='compute2',
+        )
+        inst2 = objects.Instance.get_by_uuid(self.ctxt, server2['id'])
+        self.assertEqual(1, len(inst2.numa_topology.cells))
+        # assert that the pcpu 0 is used
+        self.assertEqual(
+            {'0': 0}, inst2.numa_topology.cells[0].cpu_pinning_raw)
+
+        # migrate the first instance from compute1 to compute2 but stop
+        # migrating at the start of finish_resize. Then start a racing periodic
+        # update_available_resources.
+
+        def fake_finish_resize(*args, **kwargs):
+            # start a racing update_available_resource periodic
+            self._run_periodics()
+            # we expect it that CPU pinning fails on the destination node
+            # as the resource_tracker will use the source node numa_topology
+            # and that does not fit to the dest node as pcpu 0 in the dest
+            # is already occupied.
+
+        # TODO(stephenfin): The mock of 'migrate_disk_and_power_off' should
+        # probably be less...dumb
+        with mock.patch('nova.virt.libvirt.driver.LibvirtDriver'
+                        '.migrate_disk_and_power_off', return_value='{}'):
+            with mock.patch(
+                    'nova.compute.manager.ComputeManager.finish_resize'
+            ) as mock_finish_resize:
+                mock_finish_resize.side_effect = fake_finish_resize
+                post = {'migrate': None}
+                self.admin_api.post_server_action(server['id'], post)
+
+        log = self.stdlog.logger.output
+        # The resize_claim correctly calculates that the inst1 should be pinned
+        # to pcpu id 1 instead of 0
+        self.assertIn(
+            'Computed NUMA topology CPU pinning: usable pCPUs: [[1]], '
+            'vCPUs mapping: [(0, 1)]',
+            log,
+        )
+        # But the periodic fails as it tries to apply the source topology on
+        # the dest. This is bug 1953359.
+        log = self.stdlog.logger.output
+        self.assertIn('Error updating resources for node compute2', log)
+        self.assertIn(
+            'nova.exception.CPUPinningInvalid: CPU set to pin [0] must be '
+            'a subset of free CPU set [1]',
+            log,
+        )
+
 
 class NUMAServerTestWithCountingQuotaFromPlacement(NUMAServersTest):