openstack/neutron
Code Issues Proposed changes

Merge "Add ALLOCATING state to routers"

Browse Source
This commit is contained in:
Jenkins 2016-04-13 23:36:51 +00:00 committed by Gerrit Code Review
commit 81c61a9939
9 changed files with 160 additions and 97 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
neutron/common/constants.py
View File

@ -30,6 +30,11 @@ ROUTER_INTERFACE_OWNERS = lib_constants.ROUTER_INTERFACE_OWNERS + \
ROUTER_INTERFACE_OWNERS_SNAT = lib_constants.ROUTER_INTERFACE_OWNERS_SNAT + \
(DEVICE_OWNER_HA_REPLICATED_INT,)
ROUTER_STATUS_ACTIVE = 'ACTIVE'
# NOTE(kevinbenton): a BUILD status for routers could be added in the future
# for agents to indicate when they are wiring up the ports. The following is
# to indicate when the server is busy building sub-components of a router
ROUTER_STATUS_ALLOCATING = 'ALLOCATING'
L3_AGENT_MODE_DVR = 'dvr'
L3_AGENT_MODE_DVR_SNAT = 'dvr_snat'
L3_AGENT_MODE_LEGACY = 'legacy'

12
neutron/db/l3_agentschedulers_db.py
View File

@ -347,11 +347,13 @@ class L3AgentSchedulerDbMixin(l3agentscheduler.L3AgentSchedulerPluginBase,
router_ids):
if n_utils.is_extension_supported(self,
constants.L3_HA_MODE_EXT_ALIAS):
return self.get_ha_sync_data_for_host(context, host, agent,
router_ids=router_ids,
active=True)
return self.get_sync_data(context, router_ids=router_ids, active=True)
routers = self.get_ha_sync_data_for_host(context, host, agent,
router_ids=router_ids,
active=True)
else:
routers = self.get_sync_data(context, router_ids=router_ids,
active=True)
return self.filter_allocating_and_missing_routers(context, routers)
def list_router_ids_on_host(self, context, host, router_ids=None):
agent = self._get_agent_by_type_and_host(

26
neutron/db/l3_db.py
View File

@ -176,17 +176,41 @@ class L3_NAT_dbonly_mixin(l3.RouterPluginBase,
self._apply_dict_extend_functions(l3.ROUTERS, res, router)
return self._fields(res, fields)
def filter_allocating_and_missing_routers(self, context, routers):
"""Filter out routers that shouldn't go to the agent.
Any routers in the ALLOCATING state will be excluded by
this query because this indicates that the server is still
building necessary dependent sub-resources for the router and it
is not ready for consumption by the agent. It will also filter
out any routers that no longer exist to prevent conditions where
only part of a router's information was populated in sync_routers
due to it being deleted during the sync.
"""
router_ids = set(r['id'] for r in routers)
query = (context.session.query(Router.id).
filter(
Router.id.in_(router_ids),
Router.status != l3_constants.ROUTER_STATUS_ALLOCATING))
valid_routers = set(r.id for r in query)
if router_ids - valid_routers:
LOG.debug("Removing routers that were either concurrently "
"deleted or are in the ALLOCATING state: %s",
(router_ids - valid_routers))
return [r for r in routers if r['id'] in valid_routers]
def _create_router_db(self, context, router, tenant_id):
"""Create the DB object."""
with context.session.begin(subtransactions=True):
# pre-generate id so it will be available when
# configuring external gw port
status = router.get('status', l3_constants.ROUTER_STATUS_ACTIVE)
router_db = Router(id=(router.get('id') or
uuidutils.generate_uuid()),
tenant_id=tenant_id,
name=router['name'],
admin_state_up=router['admin_state_up'],
status="ACTIVE",
status=status,
description=router.get('description'))
context.session.add(router_db)
return router_db

58
neutron/db/l3_hamode_db.py
View File

@ -413,9 +413,11 @@ class L3_HA_NAT_db_mixin(l3_dvr_db.L3_NAT_with_dvr_db_mixin,
self._core_plugin.delete_port(admin_ctx, port_id,
l3_port_check=False)
def _notify_ha_interfaces_updated(self, context, router_id):
def _notify_ha_interfaces_updated(self, context, router_id,
schedule_routers=True):
self.l3_rpc_notifier.routers_updated(
context, [router_id], shuffle_agents=True)
context, [router_id], shuffle_agents=True,
schedule_routers=schedule_routers)
@classmethod
def _is_ha(cls, router):
@ -450,6 +452,11 @@ class L3_HA_NAT_db_mixin(l3_dvr_db.L3_NAT_with_dvr_db_mixin,
def create_router(self, context, router):
is_ha = self._is_ha(router['router'])
router['router']['ha'] = is_ha
if is_ha:
# we set the allocating status to hide it from the L3 agents
# until we have created all of the requisite interfaces/networks
router['router']['status'] = constants.ROUTER_STATUS_ALLOCATING
router_dict = super(L3_HA_NAT_db_mixin,
self).create_router(context, router)
if is_ha:
@ -461,11 +468,17 @@ class L3_HA_NAT_db_mixin(l3_dvr_db.L3_NAT_with_dvr_db_mixin,
context, router_db)[1]
self._set_vr_id(context, router_db, ha_network)
self._notify_ha_interfaces_updated(context, router_db.id)
router_dict['ha_vr_id'] = router_db.extra_attributes.ha_vr_id
self.schedule_router(context, router_dict['id'])
router_dict['status'] = self._update_router_db(
context, router_dict['id'],
{'status': constants.ROUTER_STATUS_ACTIVE})['status']
self._notify_ha_interfaces_updated(context, router_db.id,
schedule_routers=False)
except Exception:
with excutils.save_and_reraise_exception():
self.delete_router(context, router_dict['id'])
router_dict['ha_vr_id'] = router_db.extra_attributes.ha_vr_id
return router_dict
def _update_router_db(self, context, router_id, data):
@ -500,19 +513,26 @@ class L3_HA_NAT_db_mixin(l3_dvr_db.L3_NAT_with_dvr_db_mixin,
if requested_ha_state is False:
raise l3_ha.HAmodeUpdateOfDvrHaNotSupported()
with context.session.begin(subtransactions=True):
router_db = super(L3_HA_NAT_db_mixin, self)._update_router_db(
context, router_id, data)
ha_not_changed = (requested_ha_state is None or
requested_ha_state == original_ha_state)
if ha_not_changed:
return router_db
ha_changed = (requested_ha_state is not None and
requested_ha_state != original_ha_state)
if ha_changed:
if router_db.admin_state_up:
msg = _('Cannot change HA attribute of active routers. Please '
'set router admin_state_up to False prior to upgrade.')
raise n_exc.BadRequest(resource='router', msg=msg)
# set status to ALLOCATING so this router is no longer
# provided to agents while its interfaces are being re-configured.
# Keep in mind that if we want conversion to be hitless, this
# status cannot be used because agents treat hidden routers as
# deleted routers.
data['status'] = constants.ROUTER_STATUS_ALLOCATING
with context.session.begin(subtransactions=True):
router_db = super(L3_HA_NAT_db_mixin, self)._update_router_db(
context, router_id, data)
if not ha_changed:
return router_db
ha_network = self.get_ha_network(context,
router_db.tenant_id)
@ -532,10 +552,14 @@ class L3_HA_NAT_db_mixin(l3_dvr_db.L3_NAT_with_dvr_db_mixin,
ha_network = self._create_ha_interfaces_and_ensure_network(
context, router_db)[1]
self._set_vr_id(context, router_db, ha_network)
self._notify_ha_interfaces_updated(context, router_db.id)
else:
self._delete_ha_interfaces(context, router_db.id)
self._notify_ha_interfaces_updated(context, router_db.id)
self.schedule_router(context, router_id)
router_db = super(L3_HA_NAT_db_mixin, self)._update_router_db(
context, router_id, {'status': constants.ROUTER_STATUS_ACTIVE})
self._notify_ha_interfaces_updated(context, router_db.id,
schedule_routers=False)
return router_db
@ -666,9 +690,7 @@ class L3_HA_NAT_db_mixin(l3_dvr_db.L3_NAT_with_dvr_db_mixin,
if interface:
self._populate_mtu_and_subnets_for_ports(context, [interface])
# we don't want to return HA routers without HA interfaces created yet
return [r for r in list(routers_dict.values())
if not r.get('ha') or r.get(constants.HA_INTERFACE_KEY)]
return list(routers_dict.values())
@log_helpers.log_method_call
def get_ha_sync_data_for_host(self, context, host, agent,

6
neutron/scheduler/l3_agent_scheduler.py
View File

@ -87,6 +87,8 @@ class L3Scheduler(object):
l3_db.Router.id ==
l3_agentschedulers_db.RouterL3AgentBinding.router_id)
query = context.session.query(l3_db.Router.id).filter(no_agent_binding)
query = query.filter(l3_db.Router.status ==
constants.ROUTER_STATUS_ACTIVE)
unscheduled_router_ids = [router_id_[0] for router_id_ in query]
if unscheduled_router_ids:
return plugin.get_routers(
@ -102,7 +104,9 @@ class L3Scheduler(object):
:returns: the list of routers to be scheduled
"""
if router_ids is not None:
routers = plugin.get_routers(context, filters={'id': router_ids})
filters = {'id': router_ids,
'status': [constants.ROUTER_STATUS_ACTIVE]}
routers = plugin.get_routers(context, filters=filters)
return self._filter_unscheduled_routers(context, plugin, routers)
else:
return self._get_unscheduled_routers(context, plugin)

2
neutron/tests/fullstack/test_l3_agent.py
View File

@ -15,7 +15,6 @@
import functools
from oslo_utils import uuidutils
import testtools
from neutron.agent.l3 import agent as l3_agent
from neutron.agent.l3 import namespaces
@ -73,7 +72,6 @@ class TestHAL3Agent(base.BaseFullStackTestCase):
return (
agents['agents'][0]['ha_state'] != agents['agents'][1]['ha_state'])
@testtools.skip('bug/1550886')
def test_ha_router(self):
# TODO(amuller): Test external connectivity before and after a
# failover, see: https://review.openstack.org/#/c/196393/

1
neutron/tests/functional/scheduler/test_l3_agent_scheduler.py
View File

@ -284,6 +284,7 @@ class L3AZSchedulerBaseTest(test_db_base_plugin_v2.NeutronDbPluginV2TestCase):
super(L3AZSchedulerBaseTest, self).setUp(plugin=core_plugin)
self.l3_plugin = l3_router_plugin.L3RouterPlugin()
self.l3_plugin.router_scheduler = None
self.adminContext = context.get_admin_context()
self.adminContext.tenant_id = '_func_test_tenant_'

134
neutron/tests/unit/db/test_l3_hamode_db.py
View File

@ -96,15 +96,6 @@ class L3HATestFramework(testlib_api.SqlTestCase):
data['admin_state_up'] = admin_state
return self.plugin._update_router_db(ctx, router_id, data)
def _bind_router(self, router_id):
with self.admin_ctx.session.begin(subtransactions=True):
agents_db = self.plugin.get_agents_db(self.admin_ctx)
self.plugin.router_scheduler._bind_ha_router_to_agents(
self.plugin,
self.admin_ctx,
router_id,
agents_db)
class L3HATestCase(L3HATestFramework):
@ -144,7 +135,6 @@ class L3HATestCase(L3HATestFramework):
def test_get_ha_router_port_bindings(self):
router = self._create_router()
self._bind_router(router['id'])
bindings = self.plugin.get_ha_router_port_bindings(
self.admin_ctx, [router['id']])
binding_dicts = [{'router_id': binding['router_id'],
@ -157,7 +147,6 @@ class L3HATestCase(L3HATestFramework):
def test_get_l3_bindings_hosting_router_with_ha_states_ha_router(self):
router = self._create_router()
self._bind_router(router['id'])
self.plugin.update_routers_states(
self.admin_ctx, {router['id']: 'active'}, self.agent1['host'])
bindings = self.plugin.get_l3_bindings_hosting_router_with_ha_states(
@ -167,8 +156,10 @@ class L3HATestCase(L3HATestFramework):
self.assertIn((self.agent2['id'], 'standby'), agent_ids)
def test_get_l3_bindings_hosting_router_with_ha_states_agent_none(self):
router = self._create_router()
# Do not bind router to leave agents as None
with mock.patch.object(self.plugin, 'schedule_router'):
# Do not bind router to leave agents as None
router = self._create_router()
res = self.admin_ctx.session.query(
l3_hamode_db.L3HARouterAgentPortBinding).filter(
l3_hamode_db.L3HARouterAgentPortBinding.router_id == router['id']
@ -194,7 +185,6 @@ class L3HATestCase(L3HATestFramework):
def test_get_l3_bindings_hosting_router_with_ha_states_active_and_dead(
self):
router = self._create_router()
self._bind_router(router['id'])
with mock.patch.object(agents_db.Agent, 'is_active',
new_callable=mock.PropertyMock,
return_value=False):
@ -206,6 +196,67 @@ class L3HATestCase(L3HATestFramework):
agent_ids = [(agent[0]['id'], agent[1]) for agent in bindings]
self.assertIn((self.agent1['id'], 'standby'), agent_ids)
def test_router_created_in_active_state(self):
router = self._create_router()
self.assertEqual(constants.ROUTER_STATUS_ACTIVE, router['status'])
def test_router_update_stay_active(self):
router = self._create_router()
router['name'] = 'test_update'
router_updated = self.plugin._update_router_db(self.admin_ctx,
router['id'], router)
self.assertEqual(constants.ROUTER_STATUS_ACTIVE,
router_updated['status'])
def test_allocating_router_hidden_from_sync(self):
r1, r2 = self._create_router(), self._create_router()
r1['status'] = constants.ROUTER_STATUS_ALLOCATING
self.plugin._update_router_db(self.admin_ctx, r1['id'], r1)
# store shorter name for readability
get_method = self.plugin._get_active_l3_agent_routers_sync_data
# r1 should be hidden
expected = [self.plugin.get_router(self.admin_ctx, r2['id'])]
self.assertEqual(expected, get_method(self.admin_ctx, None, None,
[r1['id'], r2['id']]))
# but once it transitions back, all is well in the world again!
r1['status'] = constants.ROUTER_STATUS_ACTIVE
self.plugin._update_router_db(self.admin_ctx, r1['id'], r1)
expected.append(self.plugin.get_router(self.admin_ctx, r1['id']))
# just compare ids since python3 won't let us sort dicts
expected = sorted([r['id'] for r in expected])
result = sorted([r['id'] for r in get_method(
self.admin_ctx, None, None, [r1['id'], r2['id']])])
self.assertEqual(expected, result)
def test_router_ha_update_allocating_then_active(self):
router = self._create_router()
_orig = self.plugin._delete_ha_interfaces
def check_state(context, router_id):
self.assertEqual(
constants.ROUTER_STATUS_ALLOCATING,
self.plugin._get_router(context, router_id)['status'])
return _orig(context, router_id)
with mock.patch.object(self.plugin, '_delete_ha_interfaces',
side_effect=check_state) as ha_mock:
router = self._migrate_router(router['id'], ha=False)
self.assertTrue(ha_mock.called)
self.assertEqual(constants.ROUTER_STATUS_ACTIVE,
router['status'])
def test_router_created_allocating_state_during_interface_create(self):
_orig = self.plugin._create_ha_interfaces
def check_state(context, router_db, ha_network):
self.assertEqual(constants.ROUTER_STATUS_ALLOCATING,
router_db.status)
return _orig(context, router_db, ha_network)
with mock.patch.object(self.plugin, '_create_ha_interfaces',
side_effect=check_state) as ha_mock:
router = self._create_router()
self.assertTrue(ha_mock.called)
self.assertEqual(constants.ROUTER_STATUS_ACTIVE, router['status'])
def test_ha_router_create(self):
router = self._create_router()
self.assertTrue(router['ha'])
@ -330,7 +381,6 @@ class L3HATestCase(L3HATestFramework):
def test_unbind_ha_router(self):
router = self._create_router()
self._bind_router(router['id'])
bound_agents = self.plugin.get_l3_agents_hosting_routers(
self.admin_ctx, [router['id']])
@ -365,7 +415,6 @@ class L3HATestCase(L3HATestFramework):
def test_l3_agent_routers_query_interface(self):
router = self._create_router()
self._bind_router(router['id'])
routers = self.plugin.get_ha_sync_data_for_host(self.admin_ctx,
self.agent1['host'],
self.agent1)
@ -396,7 +445,6 @@ class L3HATestCase(L3HATestFramework):
def _deployed_router_change_ha_flag(self, to_ha):
router1 = self._create_router(ha=not to_ha)
self._bind_router(router1['id'])
routers = self.plugin.get_ha_sync_data_for_host(
self.admin_ctx, self.agent1['host'], self.agent1)
router = routers[0]
@ -427,31 +475,6 @@ class L3HATestCase(L3HATestFramework):
self._create_router()
self.assertTrue(self.notif_m.called)
def test_allocating_router_hidden_from_sync(self):
self.plugin.supported_extension_aliases = [
constants.L3_HA_MODE_EXT_ALIAS]
# simulate a router that is being allocated during
# the agent's synchronization
r1, r2 = self._create_router(), self._create_router()
self.plugin._delete_ha_interfaces(self.admin_ctx, r1['id'])
# store shorter name for readability
get_method = self.plugin._get_active_l3_agent_routers_sync_data
# r1 should be hidden
self.assertEqual([r2['id']],
[r['id'] for r in get_method(self.admin_ctx,
None, self.agent1,
[r1['id'], r2['id']])])
# but once it transitions back, all is well in the world again!
rdb = self.plugin._get_router(self.admin_ctx, r1['id'])
self.plugin._create_ha_interfaces(
self.admin_ctx, rdb, self.plugin.get_ha_network(
self.admin_ctx, rdb.tenant_id))
# just compare ids since python3 won't let us sort dicts
expected = sorted([r1['id'], r2['id']])
result = sorted([r['id'] for r in get_method(
self.admin_ctx, None, self.agent1, [r1['id'], r2['id']])])
self.assertEqual(expected, result)
def test_update_router_to_ha_notifies_agent(self):
router = self._create_router(ha=False)
self.notif_m.reset_mock()
@ -459,10 +482,8 @@ class L3HATestCase(L3HATestFramework):
self.assertTrue(self.notif_m.called)
def test_unique_vr_id_between_routers(self):
router1 = self._create_router()
router2 = self._create_router()
self._bind_router(router1['id'])
self._bind_router(router2['id'])
self._create_router()
self._create_router()
routers = self.plugin.get_ha_sync_data_for_host(
self.admin_ctx, self.agent1['host'], self.agent1)
self.assertEqual(2, len(routers))
@ -474,10 +495,8 @@ class L3HATestCase(L3HATestFramework):
@mock.patch('neutron.db.l3_hamode_db.VR_ID_RANGE', new=set(range(1, 2)))
def test_vr_id_unique_range_per_tenant(self):
router1 = self._create_router()
router2 = self._create_router(tenant_id=_uuid())
self._bind_router(router1['id'])
self._bind_router(router2['id'])
self._create_router()
self._create_router(tenant_id=_uuid())
routers = self.plugin.get_ha_sync_data_for_host(
self.admin_ctx, self.agent1['host'], self.agent1)
self.assertEqual(2, len(routers))
@ -527,10 +546,8 @@ class L3HATestCase(L3HATestFramework):
self.assertEqual(allocs_before, allocs_after)
def test_one_ha_router_one_not(self):
router1 = self._create_router(ha=False)
router2 = self._create_router()
self._bind_router(router1['id'])
self._bind_router(router2['id'])
self._create_router(ha=False)
self._create_router()
routers = self.plugin.get_ha_sync_data_for_host(
self.admin_ctx, self.agent1['host'], self.agent1)
@ -695,7 +712,6 @@ class L3HATestCase(L3HATestFramework):
def test_get_active_host_for_ha_router(self):
router = self._create_router()
self._bind_router(router['id'])
self.assertEqual(
None,
self.plugin.get_active_host_for_ha_router(
@ -709,9 +725,7 @@ class L3HATestCase(L3HATestFramework):
def test_update_routers_states(self):
router1 = self._create_router()
self._bind_router(router1['id'])
router2 = self._create_router()
self._bind_router(router2['id'])
routers = self.plugin.get_ha_sync_data_for_host(
self.admin_ctx, self.agent1['host'], self.agent1)
@ -731,9 +745,7 @@ class L3HATestCase(L3HATestFramework):
def test_set_router_states_handles_concurrently_deleted_router(self):
router1 = self._create_router()
self._bind_router(router1['id'])
router2 = self._create_router()
self._bind_router(router2['id'])
bindings = self.plugin.get_ha_router_port_bindings(
self.admin_ctx, [router1['id'], router2['id']])
self.plugin.delete_router(self.admin_ctx, router1['id'])
@ -746,7 +758,6 @@ class L3HATestCase(L3HATestFramework):
def test_update_routers_states_port_not_found(self):
router1 = self._create_router()
self._bind_router(router1['id'])
port = {'id': 'foo', 'device_id': router1['id']}
with mock.patch.object(self.core_plugin, 'get_ports',
return_value=[port]):
@ -891,7 +902,6 @@ class L3HATestCase(L3HATestFramework):
def test_update_port_status_port_bingding_deleted_concurrently(self):
router1 = self._create_router()
self._bind_router(router1['id'])
states = {router1['id']: 'active'}
with mock.patch.object(self.plugin, 'get_ha_router_port_bindings'):
(self.admin_ctx.session.query(
@ -961,7 +971,6 @@ class L3HAModeDbTestCase(L3HATestFramework):
interface_info = {'subnet_id': subnet['id']}
router = self._create_router()
self._bind_router(router['id'])
self.plugin.add_router_interface(self.admin_ctx,
router['id'],
interface_info)
@ -983,7 +992,6 @@ class L3HAModeDbTestCase(L3HATestFramework):
interface_info = {'subnet_id': subnet['id']}
router = self._create_router()
self._bind_router(router['id'])
self.plugin.add_router_interface(self.admin_ctx,
router['id'],
interface_info)

13
neutron/tests/unit/scheduler/test_l3_agent_scheduler.py
View File

@ -1479,11 +1479,11 @@ class L3_HA_scheduler_db_mixinTestCase(L3HATestCaseMixin):
# Mock scheduling so that the test can control it explicitly
mock.patch.object(l3_hamode_db.L3_HA_NAT_db_mixin,
'_notify_ha_interfaces_updated').start()
router1 = self._create_ha_router()
router2 = self._create_ha_router()
router3 = self._create_ha_router(ha=False)
router4 = self._create_ha_router(ha=False)
with mock.patch.object(self.plugin, 'schedule_router'):
router1 = self._create_ha_router()
router2 = self._create_ha_router()
router3 = self._create_ha_router(ha=False)
router4 = self._create_ha_router(ha=False)
# Agent 1 will host 0 routers, agent 2 will host 1, agent 3 will
# host 2, and agent 4 will host 3.
@ -1907,12 +1907,11 @@ class L3AgentAZLeastRoutersSchedulerTestCase(L3HATestCaseMixin):
def test_az_scheduler_ha_auto_schedule(self):
cfg.CONF.set_override('max_l3_agents_per_router', 3)
r1 = self._create_ha_router(az_hints=['az1', 'az3'])
self._set_l3_agent_admin_state(self.adminContext, self.agent2['id'],
state=False)
self._set_l3_agent_admin_state(self.adminContext, self.agent6['id'],
state=False)
self.plugin.schedule_router(self.adminContext, r1['id'])
r1 = self._create_ha_router(az_hints=['az1', 'az3'])
agents = self.plugin.get_l3_agents_hosting_routers(
self.adminContext, [r1['id']])
self.assertEqual(2, len(agents))