Merge "Introduce live_migration_claim()"
This commit is contained in:
Zuul
Gerrit Code Review
commit
6592716cca
|
|
@ -58,8 +58,8 @@ class Claim(NopClaim):
|
|||
"""
|
||||
|
||||
def __init__(self, context, instance, nodename, tracker, resources,
|
||||
pci_requests, limits=None):
|
||||
super(Claim, self).__init__()
|
||||
pci_requests, migration=None, limits=None):
|
||||
super(Claim, self).__init__(migration=migration)
|
||||
# Stash a copy of the instance at the current point of time
|
||||
self.instance = instance.obj_clone()
|
||||
self.nodename = nodename
|
||||
|
|
@ -160,7 +160,7 @@ class MoveClaim(Claim):
|
|||
Move can be either a migrate/resize, live-migrate or an evacuate operation.
|
||||
"""
|
||||
def __init__(self, context, instance, nodename, instance_type, image_meta,
|
||||
tracker, resources, pci_requests, limits=None):
|
||||
tracker, resources, pci_requests, migration, limits=None):
|
||||
self.context = context
|
||||
self.instance_type = instance_type
|
||||
if isinstance(image_meta, dict):
|
||||
|
|
@ -168,8 +168,7 @@ class MoveClaim(Claim):
|
|||
self.image_meta = image_meta
|
||||
super(MoveClaim, self).__init__(context, instance, nodename, tracker,
|
||||
resources, pci_requests,
|
||||
limits=limits)
|
||||
self.migration = None
|
||||
migration=migration, limits=limits)
|
||||
|
||||
@property
|
||||
def numa_topology(self):
|
||||
|
|
@ -186,3 +185,64 @@ class MoveClaim(Claim):
|
|||
self.instance, self.nodename,
|
||||
instance_type=self.instance_type)
|
||||
self.instance.drop_migration_context()
|
||||
|
||||
def _test_pci(self):
|
||||
"""Test whether this host can accept this claim's PCI requests. For
|
||||
live migration, only Neutron SRIOV PCI requests are supported. Any
|
||||
other type of PCI device would need to be removed and re-added for live
|
||||
migration to work, and there is currently no support for that. For cold
|
||||
migration, all types of PCI requests are supported, so we just call up
|
||||
to normal Claim's _test_pci().
|
||||
"""
|
||||
if self.migration.migration_type != 'live-migration':
|
||||
return super(MoveClaim, self)._test_pci()
|
||||
elif self._pci_requests.requests:
|
||||
for pci_request in self._pci_requests.requests:
|
||||
if (pci_request.source !=
|
||||
objects.InstancePCIRequest.NEUTRON_PORT):
|
||||
return (_('Non-VIF related PCI requests are not '
|
||||
'supported for live migration.'))
|
||||
# TODO(artom) At this point, once we've made sure we only have
|
||||
# NEUTRON_PORT (aka SRIOV) PCI requests, we should check whether
|
||||
# the host can support them, like Claim._test_pci() does. However,
|
||||
# SRIOV live migration is currently being handled separately - see
|
||||
# for example _claim_pci_for_instance_vifs() in the compute
|
||||
# manager. So we do nothing here to avoid stepping on that code's
|
||||
# toes, but ideally MoveClaim would be used for all live migration
|
||||
# resource claims.
|
||||
|
||||
def _test_live_migration_page_size(self):
|
||||
"""Tests that the current page size and the requested page size are the
|
||||
same.
|
||||
|
||||
Must be called after _test_numa_topology() to make sure
|
||||
self.claimed_numa_topology is set.
|
||||
|
||||
This only applies for live migrations when the hw:mem_page_size
|
||||
extra spec has been set to a non-numeric value (like 'large'). That
|
||||
would in theory allow an instance to live migrate from a host with a 1M
|
||||
page size to a host with a 2M page size, for example. This is not
|
||||
something we want to support, so fail the claim if the page sizes are
|
||||
different.
|
||||
"""
|
||||
if (self.migration.migration_type == 'live-migration' and
|
||||
self.instance.numa_topology and
|
||||
# NOTE(artom) We only support a single page size across all
|
||||
# cells, checking cell 0 is sufficient.
|
||||
self.claimed_numa_topology.cells[0].pagesize !=
|
||||
self.instance.numa_topology.cells[0].pagesize):
|
||||
return (_('Requested page size is different from current '
|
||||
'page size.'))
|
||||
|
||||
def _test_numa_topology(self, resources, limit):
|
||||
"""Test whether this host can accept the instance's NUMA topology. The
|
||||
_test methods return None on success, and a string-like Message _()
|
||||
object explaining the reason on failure. So we call up to the normal
|
||||
Claim's _test_numa_topology(), and if we get nothing back we test the
|
||||
page size.
|
||||
"""
|
||||
numa_test_failure = super(MoveClaim,
|
||||
self)._test_numa_topology(resources, limit)
|
||||
if numa_test_failure:
|
||||
return numa_test_failure
|
||||
return self._test_live_migration_page_size()
|
||||
|
|
|
|||
|
|
@ -243,6 +243,27 @@ class ResourceTracker(object):
|
|||
migration, image_meta=image_meta,
|
||||
limits=limits)
|
||||
|
||||
@utils.synchronized(COMPUTE_RESOURCE_SEMAPHORE)
|
||||
def live_migration_claim(self, context, instance, nodename, migration,
|
||||
limits):
|
||||
"""Builds a MoveClaim for a live migration.
|
||||
|
||||
:param context: The request context.
|
||||
:param instance: The instance being live migrated.
|
||||
:param nodename: The nodename of the destination host.
|
||||
:param migration: The Migration object associated with this live
|
||||
migration.
|
||||
:param limits: A SchedulerLimits object from when the scheduler
|
||||
selected the destination host.
|
||||
:returns: A MoveClaim for this live migration.
|
||||
"""
|
||||
# Flavor and image cannot change during a live migration.
|
||||
instance_type = instance.flavor
|
||||
image_meta = instance.image_meta
|
||||
return self._move_claim(context, instance, instance_type, nodename,
|
||||
migration, move_type='live-migration',
|
||||
image_meta=image_meta, limits=limits)
|
||||
|
||||
def _move_claim(self, context, instance, new_instance_type, nodename,
|
||||
migration, move_type=None, image_meta=None, limits=None):
|
||||
"""Indicate that resources are needed for a move to this host.
|
||||
|
|
@ -295,12 +316,17 @@ class ResourceTracker(object):
|
|||
new_pci_requests.requests.append(request)
|
||||
claim = claims.MoveClaim(context, instance, nodename,
|
||||
new_instance_type, image_meta, self, cn,
|
||||
new_pci_requests,
|
||||
limits=limits)
|
||||
new_pci_requests, migration, limits=limits)
|
||||
|
||||
claim.migration = migration
|
||||
claimed_pci_devices_objs = []
|
||||
if self.pci_tracker:
|
||||
# TODO(artom) The second part of this condition should not be
|
||||
# necessary, but since SRIOV live migration is currently handled
|
||||
# elsewhere - see for example _claim_pci_for_instance_vifs() in the
|
||||
# compute manager - we don't do any PCI claims if this is a live
|
||||
# migration to avoid stepping on that code's toes. Ideally,
|
||||
# MoveClaim/this method would be used for all live migration resource
|
||||
# claims.
|
||||
if self.pci_tracker and migration.migration_type != 'live-migration':
|
||||
# NOTE(jaypipes): ComputeNode.pci_device_pools is set below
|
||||
# in _update_usage_from_instance().
|
||||
claimed_pci_devices_objs = self.pci_tracker.claim_instance(
|
||||
|
|
@ -367,7 +393,11 @@ class ResourceTracker(object):
|
|||
migration.dest_compute = self.host
|
||||
migration.dest_node = nodename
|
||||
migration.dest_host = self.driver.get_host_ip_addr()
|
||||
migration.status = 'pre-migrating'
|
||||
# NOTE(artom) Migration objects for live migrations are created with
|
||||
# status 'accepted' by the conductor in live_migrate_instance() and do
|
||||
# not have a 'pre-migrating' status.
|
||||
if migration.migration_type != 'live-migration':
|
||||
migration.status = 'pre-migrating'
|
||||
migration.save()
|
||||
|
||||
def _set_instance_host_and_node(self, instance, nodename):
|
||||
|
|
|
|||
|
|
@ -338,10 +338,10 @@ class MoveClaimTestCase(ClaimTestCase):
|
|||
@mock.patch('nova.db.api.instance_extra_get_by_instance_uuid',
|
||||
return_value=self.db_numa_topology)
|
||||
def get_claim(mock_extra_get, mock_numa_get):
|
||||
return claims.MoveClaim(self.context, self.instance, _NODENAME,
|
||||
instance_type, image_meta, self.tracker,
|
||||
self.resources, requests,
|
||||
limits=limits)
|
||||
return claims.MoveClaim(
|
||||
self.context, self.instance, _NODENAME, instance_type,
|
||||
image_meta, self.tracker, self.resources, requests,
|
||||
objects.Migration(migration_type='migration'), limits=limits)
|
||||
return get_claim()
|
||||
|
||||
@mock.patch('nova.objects.Instance.drop_migration_context')
|
||||
|
|
@ -358,3 +358,69 @@ class MoveClaimTestCase(ClaimTestCase):
|
|||
image_meta = objects.ImageMeta()
|
||||
claim = self._claim(image_meta=image_meta)
|
||||
self.assertIsInstance(claim.image_meta, objects.ImageMeta)
|
||||
|
||||
|
||||
class LiveMigrationClaimTestCase(ClaimTestCase):
|
||||
|
||||
def test_live_migration_claim_bad_pci_request(self):
|
||||
instance_type = self._fake_instance_type()
|
||||
instance = self._fake_instance()
|
||||
instance.numa_topology = None
|
||||
self.assertRaisesRegex(
|
||||
exception.ComputeResourcesUnavailable,
|
||||
'PCI requests are not supported',
|
||||
claims.MoveClaim, self.context, instance, _NODENAME, instance_type,
|
||||
{}, self.tracker, self.resources,
|
||||
objects.InstancePCIRequests(requests=[
|
||||
objects.InstancePCIRequest(alias_name='fake-alias')]),
|
||||
objects.Migration(migration_type='live-migration'), None)
|
||||
|
||||
def test_live_migration_page_size(self):
|
||||
instance_type = self._fake_instance_type()
|
||||
instance = self._fake_instance()
|
||||
instance.numa_topology = objects.InstanceNUMATopology(
|
||||
cells=[objects.InstanceNUMACell(id=1, cpuset=set([1, 2]),
|
||||
memory=512, pagesize=2)])
|
||||
claimed_numa_topology = objects.InstanceNUMATopology(
|
||||
cells=[objects.InstanceNUMACell(id=1, cpuset=set([1, 2]),
|
||||
memory=512, pagesize=1)])
|
||||
with mock.patch('nova.virt.hardware.numa_fit_instance_to_host',
|
||||
return_value=claimed_numa_topology):
|
||||
self.assertRaisesRegex(
|
||||
exception.ComputeResourcesUnavailable,
|
||||
'Requested page size is different',
|
||||
claims.MoveClaim, self.context, instance, _NODENAME,
|
||||
instance_type, {}, self.tracker, self.resources,
|
||||
self.empty_requests,
|
||||
objects.Migration(migration_type='live-migration'), None)
|
||||
|
||||
def test_claim_fails_page_size_not_called(self):
|
||||
instance_type = self._fake_instance_type()
|
||||
instance = self._fake_instance()
|
||||
# This topology cannot fit in self.resources (see _fake_resources())
|
||||
numa_topology = objects.InstanceNUMATopology(
|
||||
cells=[objects.InstanceNUMACell(id=1, cpuset=set([1, 2, 3]),
|
||||
memory=1024)])
|
||||
with test.nested(
|
||||
mock.patch('nova.virt.hardware.numa_get_constraints',
|
||||
return_value=numa_topology),
|
||||
mock.patch(
|
||||
'nova.compute.claims.MoveClaim._test_live_migration_page_size'
|
||||
)) as (mock_test_numa, mock_test_page_size):
|
||||
self.assertRaisesRegex(
|
||||
exception.ComputeResourcesUnavailable,
|
||||
'Requested instance NUMA topology',
|
||||
claims.MoveClaim, self.context, instance, _NODENAME,
|
||||
instance_type, {}, self.tracker, self.resources,
|
||||
self.empty_requests,
|
||||
objects.Migration(migration_type='live-migration'), None)
|
||||
mock_test_page_size.assert_not_called()
|
||||
|
||||
def test_live_migration_no_instance_numa_topology(self):
|
||||
instance_type = self._fake_instance_type()
|
||||
instance = self._fake_instance()
|
||||
instance.numa_topology = None
|
||||
claims.MoveClaim(
|
||||
self.context, instance, _NODENAME, instance_type, {}, self.tracker,
|
||||
self.resources, self.empty_requests,
|
||||
objects.Migration(migration_type='live-migration'), None)
|
||||
|
|
|
|||
|
|
@ -38,6 +38,7 @@ from nova.objects import pci_device
|
|||
from nova.pci import manager as pci_manager
|
||||
from nova.scheduler.client import report
|
||||
from nova import test
|
||||
from nova.tests.unit import fake_instance
|
||||
from nova.tests.unit import fake_notifier
|
||||
from nova.tests.unit.objects import test_pci_device as fake_pci_device
|
||||
from nova.virt import driver
|
||||
|
|
@ -2824,6 +2825,42 @@ class TestRebuild(BaseTestCase):
|
|||
inst_save_mock.assert_called_once_with()
|
||||
|
||||
|
||||
class TestLiveMigration(BaseTestCase):
|
||||
|
||||
def test_live_migration_claim(self):
|
||||
self._setup_rt()
|
||||
self.rt.compute_nodes[_NODENAME] = _COMPUTE_NODE_FIXTURES[0]
|
||||
ctxt = context.get_admin_context()
|
||||
instance = fake_instance.fake_instance_obj(ctxt)
|
||||
instance.pci_requests = None
|
||||
instance.pci_devices = None
|
||||
instance.numa_topology = None
|
||||
migration = objects.Migration(id=42, migration_type='live-migration',
|
||||
status='accepted')
|
||||
image_meta = objects.ImageMeta(properties=objects.ImageMetaProps())
|
||||
self.rt.pci_tracker = pci_manager.PciDevTracker(mock.sentinel.ctx)
|
||||
with test.nested(
|
||||
mock.patch.object(objects.ImageMeta, 'from_instance',
|
||||
return_value=image_meta),
|
||||
mock.patch.object(objects.Migration, 'save'),
|
||||
mock.patch.object(objects.Instance, 'save'),
|
||||
mock.patch.object(self.rt, '_update'),
|
||||
mock.patch.object(self.rt.pci_tracker, 'claim_instance'),
|
||||
) as (mock_from_instance, mock_migration_save, mock_instance_save,
|
||||
mock_update, mock_pci_claim_instance):
|
||||
claim = self.rt.live_migration_claim(ctxt, instance, _NODENAME,
|
||||
migration, limits=None)
|
||||
self.assertEqual(42, claim.migration.id)
|
||||
# Check that we didn't set the status to 'pre-migrating', like we
|
||||
# do for cold migrations, but which doesn't exist for live
|
||||
# migrations.
|
||||
self.assertEqual('accepted', claim.migration.status)
|
||||
self.assertIn('migration_context', instance)
|
||||
mock_update.assert_called_with(
|
||||
mock.ANY, _COMPUTE_NODE_FIXTURES[0])
|
||||
mock_pci_claim_instance.assert_not_called()
|
||||
|
||||
|
||||
class TestUpdateUsageFromMigration(test.NoDBTestCase):
|
||||
@mock.patch('nova.compute.resource_tracker.ResourceTracker.'
|
||||
'_get_instance_type')
|
||||
|
|
|
|||
Loading…
Reference in New Issue