Support same host resize with PCI in placement
Id02e445c55fc956965b7d725f0260876d42422f2 added special case in the healing logic for same host resize. Now that the scheduler also creates allocation on the destination host during resize we need to make sure that the drop_move_claim code that runs during revert and confirm drops the tracked migration from the resource tracker only after the healing logic run as these migrations being confirmed / reverted are still affecting PciDevices at this point. blueprint: pci-device-tracking-in-placement Change-Id: I6241965fe6c1cc1f2560fcce65d5e32ef308d502
This commit is contained in:
Balazs Gibizer
parent
b387401187
commit
fa4832c660
|
|
@ -619,18 +619,11 @@ class ResourceTracker(object):
|
|||
:param prefix: Prefix to use when accessing migration context
|
||||
attributes. 'old_' or 'new_', with 'new_' being the default.
|
||||
"""
|
||||
# Remove usage for an instance that is tracked in migrations, such as
|
||||
# on the dest node during revert resize.
|
||||
if instance['uuid'] in self.tracked_migrations:
|
||||
migration = self.tracked_migrations.pop(instance['uuid'])
|
||||
if instance["uuid"] in self.tracked_migrations:
|
||||
if not flavor:
|
||||
flavor = self._get_flavor(instance, prefix, migration)
|
||||
# Remove usage for an instance that is not tracked in migrations (such
|
||||
# as on the source node after a migration).
|
||||
# NOTE(lbeliveau): On resize on the same node, the instance is
|
||||
# included in both tracked_migrations and tracked_instances.
|
||||
elif instance['uuid'] in self.tracked_instances:
|
||||
self.tracked_instances.remove(instance['uuid'])
|
||||
flavor = self._get_flavor(
|
||||
instance, prefix, self.tracked_migrations[instance["uuid"]]
|
||||
)
|
||||
|
||||
if flavor is not None:
|
||||
numa_topology = self._get_migration_context_resource(
|
||||
|
|
@ -646,6 +639,15 @@ class ResourceTracker(object):
|
|||
ctxt = context.elevated()
|
||||
self._update(ctxt, self.compute_nodes[nodename])
|
||||
|
||||
# Remove usage for an instance that is tracked in migrations, such as
|
||||
# on the dest node during revert resize.
|
||||
self.tracked_migrations.pop(instance['uuid'], None)
|
||||
# Remove usage for an instance that is not tracked in migrations (such
|
||||
# as on the source node after a migration).
|
||||
# NOTE(lbeliveau): On resize on the same node, the instance is
|
||||
# included in both tracked_migrations and tracked_instances.
|
||||
self.tracked_instances.discard(instance['uuid'])
|
||||
|
||||
@utils.synchronized(COMPUTE_RESOURCE_SEMAPHORE, fair=True)
|
||||
def update_usage(self, context, instance, nodename):
|
||||
"""Update the resource usage and stats after a change in an
|
||||
|
|
|
|||
|
|
@ -1597,8 +1597,17 @@ class PlacementPCIAllocationHealingTests(PlacementPCIReportingTests):
|
|||
compute1_expected_placement_view["allocations"][server["id"]] = {
|
||||
"0000:81:00.0": {self.VF_RC: 2}
|
||||
}
|
||||
self.assert_placement_pci_view(
|
||||
"compute1", **compute1_expected_placement_view)
|
||||
# NOTE(gibi): This is unfortunate but during same host resize
|
||||
# confirm when the PCI scheduling is not enabled the healing logic
|
||||
# cannot heal the dest host allocation during the claim. It will only
|
||||
# heal it in the next run of the ResourceTracker._update(). This due
|
||||
# to the fact that ResourceTracker.drop_move_claim runs both for
|
||||
# revert (on the dest) and confirm (on the source) and in same host
|
||||
# resize this means that it runs on both the source and the dest as
|
||||
# they are the same.
|
||||
# Anyhow the healing will happen just a bit later. And the end goal is
|
||||
# to make the scheduler support enabled by default and delete the
|
||||
# whole healing logic. So I think this is acceptable.
|
||||
self._run_periodics()
|
||||
self.assert_placement_pci_view(
|
||||
"compute1", **compute1_expected_placement_view)
|
||||
|
|
|
|||
|
|
@ -2356,6 +2356,133 @@ class PCIServersTest(_PCIServersTestBase):
|
|||
self.assert_no_pci_healing("test_compute0")
|
||||
self.assert_no_pci_healing("test_compute1")
|
||||
|
||||
def test_same_host_resize_with_pci(self):
|
||||
"""Start a single compute with 3 PCI devs and resize and instance
|
||||
from one dev to two devs
|
||||
"""
|
||||
self.flags(allow_resize_to_same_host=True)
|
||||
self.start_compute(
|
||||
hostname='test_compute0',
|
||||
pci_info=fakelibvirt.HostPCIDevicesInfo(num_pci=3))
|
||||
self.assertPCIDeviceCounts('test_compute0', total=3, free=3)
|
||||
test_compute0_placement_pci_view = {
|
||||
"inventories": {
|
||||
"0000:81:00.0": {self.PCI_RC: 1},
|
||||
"0000:81:01.0": {self.PCI_RC: 1},
|
||||
"0000:81:02.0": {self.PCI_RC: 1},
|
||||
},
|
||||
"traits": {
|
||||
"0000:81:00.0": [],
|
||||
"0000:81:01.0": [],
|
||||
"0000:81:02.0": [],
|
||||
},
|
||||
"usages": {
|
||||
"0000:81:00.0": {self.PCI_RC: 0},
|
||||
"0000:81:01.0": {self.PCI_RC: 0},
|
||||
"0000:81:02.0": {self.PCI_RC: 0},
|
||||
},
|
||||
"allocations": {},
|
||||
}
|
||||
self.assert_placement_pci_view(
|
||||
"test_compute0", **test_compute0_placement_pci_view)
|
||||
|
||||
# Boot a server with a single PCI device.
|
||||
# To stabilize the test we reserve 81.01 and 81.02 in placement so
|
||||
# we can be sure that the instance will use 81.00, otherwise the
|
||||
# allocation will be random between 00, 01, and 02
|
||||
self._reserve_placement_resource(
|
||||
"test_compute0_0000:81:01.0", self.PCI_RC, 1)
|
||||
self._reserve_placement_resource(
|
||||
"test_compute0_0000:81:02.0", self.PCI_RC, 1)
|
||||
extra_spec = {'pci_passthrough:alias': f'{self.ALIAS_NAME}:1'}
|
||||
pci_flavor_id = self._create_flavor(extra_spec=extra_spec)
|
||||
server = self._create_server(flavor_id=pci_flavor_id, networks='none')
|
||||
|
||||
self.assertPCIDeviceCounts('test_compute0', total=3, free=2)
|
||||
test_compute0_placement_pci_view[
|
||||
"usages"]["0000:81:00.0"][self.PCI_RC] = 1
|
||||
test_compute0_placement_pci_view[
|
||||
"allocations"][server['id']] = {"0000:81:00.0": {self.PCI_RC: 1}}
|
||||
self.assert_placement_pci_view(
|
||||
"test_compute0", **test_compute0_placement_pci_view)
|
||||
# remove the reservations, so we can resize on the same host and
|
||||
# consume 01 and 02
|
||||
self._reserve_placement_resource(
|
||||
"test_compute0_0000:81:01.0", self.PCI_RC, 0)
|
||||
self._reserve_placement_resource(
|
||||
"test_compute0_0000:81:02.0", self.PCI_RC, 0)
|
||||
|
||||
# Resize the server to use 2 PCI devices
|
||||
extra_spec = {'pci_passthrough:alias': f'{self.ALIAS_NAME}:2'}
|
||||
pci_flavor_id = self._create_flavor(extra_spec=extra_spec)
|
||||
with mock.patch(
|
||||
'nova.virt.libvirt.driver.LibvirtDriver'
|
||||
'.migrate_disk_and_power_off',
|
||||
return_value='{}',
|
||||
):
|
||||
self._resize_server(server, pci_flavor_id)
|
||||
|
||||
self.assertPCIDeviceCounts('test_compute0', total=3, free=0)
|
||||
# the source host side of the allocation is now held by the migration
|
||||
# UUID
|
||||
self._move_server_allocation(
|
||||
test_compute0_placement_pci_view["allocations"], server['id'])
|
||||
# but we have the dest host side of the allocations on the same host
|
||||
test_compute0_placement_pci_view[
|
||||
"usages"]["0000:81:01.0"][self.PCI_RC] = 1
|
||||
test_compute0_placement_pci_view[
|
||||
"usages"]["0000:81:02.0"][self.PCI_RC] = 1
|
||||
test_compute0_placement_pci_view["allocations"][server['id']] = {
|
||||
"0000:81:01.0": {self.PCI_RC: 1},
|
||||
"0000:81:02.0": {self.PCI_RC: 1},
|
||||
}
|
||||
self.assert_placement_pci_view(
|
||||
"test_compute0", **test_compute0_placement_pci_view)
|
||||
|
||||
# revert the resize so the instance should go back to use a single
|
||||
# device
|
||||
self._revert_resize(server)
|
||||
self.assertPCIDeviceCounts('test_compute0', total=3, free=2)
|
||||
# the migration allocation is moved back to the instance UUID
|
||||
self._move_server_allocation(
|
||||
test_compute0_placement_pci_view["allocations"],
|
||||
server["id"],
|
||||
revert=True,
|
||||
)
|
||||
# and the "dest" side of the allocation is dropped
|
||||
test_compute0_placement_pci_view[
|
||||
"usages"]["0000:81:01.0"][self.PCI_RC] = 0
|
||||
test_compute0_placement_pci_view[
|
||||
"usages"]["0000:81:02.0"][self.PCI_RC] = 0
|
||||
test_compute0_placement_pci_view["allocations"][server['id']] = {
|
||||
"0000:81:00.0": {self.PCI_RC: 1},
|
||||
}
|
||||
self.assert_placement_pci_view(
|
||||
"test_compute0", **test_compute0_placement_pci_view)
|
||||
|
||||
# resize again but now confirm the same host resize and assert that
|
||||
# only the new flavor usage remains
|
||||
with mock.patch(
|
||||
'nova.virt.libvirt.driver.LibvirtDriver'
|
||||
'.migrate_disk_and_power_off',
|
||||
return_value='{}',
|
||||
):
|
||||
self._resize_server(server, pci_flavor_id)
|
||||
self._confirm_resize(server)
|
||||
|
||||
self.assertPCIDeviceCounts('test_compute0', total=3, free=1)
|
||||
test_compute0_placement_pci_view["usages"] = {
|
||||
"0000:81:01.0": {self.PCI_RC: 1},
|
||||
"0000:81:02.0": {self.PCI_RC: 1},
|
||||
}
|
||||
test_compute0_placement_pci_view["allocations"][
|
||||
server['id']] = {self.PCI_RC: 1}
|
||||
test_compute0_placement_pci_view["allocations"][server['id']] = {
|
||||
"0000:81:01.0": {self.PCI_RC: 1},
|
||||
"0000:81:02.0": {self.PCI_RC: 1},
|
||||
}
|
||||
self.assert_no_pci_healing("test_compute0")
|
||||
|
||||
def _confirm_resize(self, server, host='host1'):
|
||||
# NOTE(sbauza): Unfortunately, _cleanup_resize() in libvirt checks the
|
||||
# host option to know the source hostname but given we have a global
|
||||
|
|
|
|||
Loading…
Reference in New Issue