[OVN] Use the Chassis_Private table for agents healthcheck

The core OVN team has introduced a new table called Chassis_Private to
avoid nb_cfg flooding when checking for the Chassis' status. The OVN
driver does rely on that mechanism for the agent liveness mechanism.

This patch makes use of this new table but it's also backward
compatible.

For more information, check the core OVN changes at:
https://patchwork.ozlabs.org/patch/1254394.

Closes-Bug: #1892477
Change-Id: Iea4263b852d1e3f81eb2557918ea3cbb7adb8016
Signed-off-by: Lucas Alvares Gomes <lucasagomes@gmail.com>

neutron/agent/ovn/metadata/agent.py

@@ -129,7 +129,7 @@ class PortBindingChassisDeletedEvent(PortBindingChassisEvent):
             return False
 
 
-class ChassisCreateEvent(row_event.RowEvent):
+class ChassisCreateEventBase(row_event.RowEvent):
     """Row create event - Chassis name == our_chassis.
 
     On connection, we get a dump of all chassis so if we catch a creation
@@ -137,14 +137,14 @@ class ChassisCreateEvent(row_event.RowEvent):
     to do a full sync to make sure that we capture all changes while the
     connection to OVSDB was down.
     """
+    table = None
 
     def __init__(self, metadata_agent):
         self.agent = metadata_agent
         self.first_time = True
-        table = 'Chassis'
         events = (self.ROW_CREATE,)
-        super(ChassisCreateEvent, self).__init__(
+        super(ChassisCreateEventBase, self).__init__(
-            events, table, (('name', '=', self.agent.chassis),))
+            events, self.table, (('name', '=', self.agent.chassis),))
         self.event_name = self.__class__.__name__
 
     def run(self, event, row, old):
@@ -159,6 +159,14 @@ class ChassisCreateEvent(row_event.RowEvent):
             self.agent.sync()
 
 
+class ChassisCreateEvent(ChassisCreateEventBase):
+    table = 'Chassis'
+
+
+class ChassisPrivateCreateEvent(ChassisCreateEventBase):
+    table = 'Chassis_Private'
+
+
 class SbGlobalUpdateEvent(row_event.RowEvent):
     """Row update event on SB_Global table."""
 
@@ -170,8 +178,12 @@ class SbGlobalUpdateEvent(row_event.RowEvent):
         self.event_name = self.__class__.__name__
 
     def run(self, event, row, old):
-        self.agent.sb_idl.update_metadata_health_status(
+        table = ('Chassis_Private' if self.agent.has_chassis_private
-            self.agent.chassis, row.nb_cfg).execute()
+                 else 'Chassis')
+        self.agent.sb_idl.db_set(
+            table, self.agent.chassis, ('external_ids', {
+                ovn_const.OVN_AGENT_METADATA_SB_CFG_KEY:
+                    str(row.nb_cfg)})).execute()
 
 
 class MetadataAgent(object):
@@ -207,13 +219,26 @@ class MetadataAgent(object):
         proxy = metadata_server.UnixDomainMetadataProxy(self.conf)
         proxy.run()
 
+        tables = ('Encap', 'Port_Binding', 'Datapath_Binding', 'SB_Global',
+                  'Chassis')
+        events = (PortBindingChassisCreatedEvent(self),
+                  PortBindingChassisDeletedEvent(self),
+                  SbGlobalUpdateEvent(self))
+
+        # TODO(lucasagomes): Remove this in the future. Try to register
+        # the Chassis_Private table, if not present, fallback to the normal
+        # Chassis table.
         # Open the connection to OVN SB database.
-        self.sb_idl = ovsdb.MetadataAgentOvnSbIdl(
+        self.has_chassis_private = False
-            chassis=self.chassis,
+        try:
-            events=[PortBindingChassisCreatedEvent(self),
+            self.sb_idl = ovsdb.MetadataAgentOvnSbIdl(
-                    PortBindingChassisDeletedEvent(self),
+                chassis=self.chassis, tables=tables + ('Chassis_Private', ),
-                    ChassisCreateEvent(self),
+                events=events + (ChassisPrivateCreateEvent(self), )).start()
-                    SbGlobalUpdateEvent(self)]).start()
+            self.has_chassis_private = True
+        except AssertionError:
+            self.sb_idl = ovsdb.MetadataAgentOvnSbIdl(
+                chassis=self.chassis, tables=tables,
+                events=events + (ChassisCreateEvent(self), )).start()
 
         # Do the initial sync.
         self.sync()
@@ -226,9 +251,10 @@ class MetadataAgent(object):
     def register_metadata_agent(self):
         # NOTE(lucasagomes): db_add() will not overwrite the UUID if
         # it's already set.
+        table = ('Chassis_Private' if self.has_chassis_private else 'Chassis')
         ext_ids = {
             ovn_const.OVN_AGENT_METADATA_ID_KEY: uuidutils.generate_uuid()}
-        self.sb_idl.db_add('Chassis', self.chassis, 'external_ids',
+        self.sb_idl.db_add(table, self.chassis, 'external_ids',
                            ext_ids).execute(check_error=True)
 
     def _get_own_chassis_name(self):

neutron/agent/ovn/metadata/ovsdb.py

@@ -38,8 +38,10 @@ class MetadataAgentOvnSbIdl(ovsdb_monitor.OvnIdl):
             helper.register_table(table)
         super(MetadataAgentOvnSbIdl, self).__init__(
             None, connection_string, helper)
-        if chassis and 'Chassis' in tables:
+        if chassis:
-            self.tables['Chassis'].condition = [['name', '==', chassis]]
+            table = ('Chassis_Private' if 'Chassis_Private' in tables
+                     else 'Chassis')
+            self.tables[table].condition = [['name', '==', chassis]]
         if events:
             self.notify_handler.watch_events(events)
 

neutron/plugins/ml2/drivers/ovn/agent/neutron_agent.py

@@ -28,8 +28,13 @@ class NeutronAgent(abc.ABC):
         for _type in cls.types:
             NeutronAgent.types[_type] = cls
 
-    def __init__(self, chassis):
+    def __init__(self, chassis_private):
-        self.chassis = chassis
+        self.chassis_private = chassis_private
+        try:
+            self.chassis = self.chassis_private.chassis[0]
+        except (AttributeError, IndexError):
+            # No Chassis_Private support, just use Chassis
+            self.chassis = self.chassis_private
 
     @property
     def updated_at(self):
@@ -58,8 +63,8 @@ class NeutronAgent(abc.ABC):
             'admin_state_up': True}
 
     @classmethod
-    def from_type(cls, _type, chassis):
+    def from_type(cls, _type, chassis_private):
-        return cls.types[_type](chassis)
+        return cls.types[_type](chassis_private)
 
     @staticmethod
     def agent_types():
@@ -85,15 +90,15 @@ class ControllerAgent(NeutronAgent):
 
     @property
     def nb_cfg(self):
-        return self.chassis.nb_cfg
+        return self.chassis_private.nb_cfg
 
     @property
     def agent_id(self):
-        return self.chassis.name
+        return self.chassis_private.name
 
     @property
     def description(self):
-        return self.chassis.external_ids.get(
+        return self.chassis_private.external_ids.get(
             ovn_const.OVN_AGENT_DESC_KEY, '')
 
 
@@ -105,15 +110,15 @@ class MetadataAgent(NeutronAgent):
 
     @property
     def nb_cfg(self):
-        return int(self.chassis.external_ids.get(
+        return int(self.chassis_private.external_ids.get(
             ovn_const.OVN_AGENT_METADATA_SB_CFG_KEY, 0))
 
     @property
     def agent_id(self):
-        return self.chassis.external_ids.get(
+        return self.chassis_private.external_ids.get(
             ovn_const.OVN_AGENT_METADATA_ID_KEY)
 
     @property
     def description(self):
-        return self.chassis.external_ids.get(
+        return self.chassis_private.external_ids.get(
             ovn_const.OVN_AGENT_METADATA_DESC_KEY, '')

neutron/plugins/ml2/drivers/ovn/mech_driver/mech_driver.py

@@ -120,6 +120,9 @@ class OVNMechanismDriver(api.MechanismDriver):
         self.subscribe()
         self.qos_driver = qos_driver.OVNQosDriver.create(self)
         self.trunk_driver = trunk_driver.OVNTrunkDriver.create(self)
+        # The agent_chassis_table will be changed to Chassis_Private if it
+        # exists, we need to have a connection in order to check that.
+        self.agent_chassis_table = 'Chassis'
 
     @property
     def _plugin(self):
@@ -262,6 +265,9 @@ class OVNMechanismDriver(api.MechanismDriver):
         self._nb_ovn, self._sb_ovn = impl_idl_ovn.get_ovn_idls(
             self, trigger, binding_events=not is_maintenance)
 
+        if self._sb_ovn.is_table_present('Chassis_Private'):
+            self.agent_chassis_table = 'Chassis_Private'
+
         # AGENTS must be populated after fork so if ovn-controller is stopped
         # before a worker handles a get_agents request, we still show agents
         populate_agents(self)
@@ -1105,14 +1111,15 @@ class OVNMechanismDriver(api.MechanismDriver):
     def mark_agent_alive(self, agent):
         # Update the time of our successful check
         value = timeutils.utcnow(with_timezone=True).isoformat()
-        self._sb_ovn.db_set('Chassis', agent.chassis.uuid,
+        self._sb_ovn.db_set(
-                            ('external_ids', {agent.key: value})).execute(
+            self.agent_chassis_table, agent.chassis_private.uuid,
-            check_error=True)
+            ('external_ids', {agent.key: value})).execute(check_error=True)
 
-    def agents_from_chassis(self, chassis, update_db=True):
+    def agents_from_chassis(self, chassis_private, update_db=True):
         agent_dict = {}
         # Iterate over each unique Agent subclass
-        for agent in [a(chassis) for a in n_agent.NeutronAgent.agent_types()]:
+        for agent in [a(chassis_private)
+                      for a in n_agent.NeutronAgent.agent_types()]:
             if not agent.agent_id:
                 continue
             alive = self.agent_alive(agent, update_db)
@@ -1186,7 +1193,7 @@ class OVNMechanismDriver(api.MechanismDriver):
 
 
 def populate_agents(driver):
-    for ch in driver._sb_ovn.tables['Chassis'].rows.values():
+    for ch in driver._sb_ovn.tables[driver.agent_chassis_table].rows.values():
         # update the cache, rows are hashed on uuid but it is the name that
         # stays consistent across ovn-controller restarts
         AGENTS.update({ch.name: ch})
@@ -1207,11 +1214,12 @@ def get_agents(self, context, filters=None, fields=None, _driver=None):
 def get_agent(self, context, id, fields=None, _driver=None):
     chassis = None
     try:
-        # look up Chassis by *name*, which the id attribte is
+        # look up Chassis by *name*, which the id attribute is
-        chassis = _driver._sb_ovn.lookup('Chassis', id)
+        chassis = _driver._sb_ovn.lookup(_driver.agent_chassis_table, id)
     except idlutils.RowNotFound:
         # If the UUID is not found, check for the metadata agent ID
-        for ch in _driver._sb_ovn.tables['Chassis'].rows.values():
+        for ch in _driver._sb_ovn.tables[
+                _driver.agent_chassis_table].rows.values():
             metadata_agent_id = ch.external_ids.get(
                 ovn_const.OVN_AGENT_METADATA_ID_KEY)
             if id == metadata_agent_id:

neutron/plugins/ml2/drivers/ovn/mech_driver/ovsdb/impl_idl_ovn.py

@@ -836,12 +836,6 @@ class OvsdbSbOvnIdl(sb_impl_idl.OvnSbApiIdlImpl, Backend):
                 if (r.mac and str(r.datapath.uuid) == network) and
                 ip_address in r.mac[0].split(' ')]
 
-    def update_metadata_health_status(self, chassis, nb_cfg):
-        return cmd.UpdateChassisExtIdsCommand(
-            self, chassis,
-            {ovn_const.OVN_AGENT_METADATA_SB_CFG_KEY: str(nb_cfg)},
-            if_exists=True)
-
     def set_port_cidrs(self, name, cidrs):
         # TODO(twilson) add if_exists to db commands
         return self.db_set('Port_Binding', name, 'external_ids',

neutron/plugins/ml2/drivers/ovn/mech_driver/ovsdb/ovsdb_monitor.py

@@ -528,6 +528,8 @@ class OvnSbIdl(OvnIdlDistributedLock):
     def from_server(cls, connection_string, schema_name, driver):
         _check_and_set_ssl_files(schema_name)
         helper = idlutils.get_schema_helper(connection_string, schema_name)
+        if 'Chassis_Private' in helper.schema_json['tables']:
+            helper.register_table('Chassis_Private')
         helper.register_table('Chassis')
         helper.register_table('Encap')
         helper.register_table('Port_Binding')

neutron/tests/unit/fake_resources.py

@@ -163,6 +163,8 @@ class FakeOvsdbSbOvnIdl(object):
         self.db_set = mock.Mock()
         self.lookup = mock.MagicMock()
         self.chassis_list = mock.MagicMock()
+        self.is_table_present = mock.Mock()
+        self.is_table_present.return_value = False
 
 
 class FakeOvsdbTransaction(object):

neutron/tests/unit/plugins/ml2/drivers/ovn/mech_driver/test_mech_driver.py

@@ -1615,18 +1615,21 @@ class TestOVNMechanismDriver(test_plugin.Ml2PluginV2TestCase):
 
     def _add_chassis_agent(self, nb_cfg, agent_type, updated_at=None):
         updated_at = updated_at or datetime.datetime.utcnow()
-        chassis = mock.Mock()
+        chassis_private = mock.Mock()
-        chassis.nb_cfg = nb_cfg
+        chassis_private.nb_cfg = nb_cfg
-        chassis.uuid = uuid.uuid4()
+        chassis_private.uuid = uuid.uuid4()
-        chassis.external_ids = {ovn_const.OVN_LIVENESS_CHECK_EXT_ID_KEY:
+        chassis_private.external_ids = {
-                                datetime.datetime.isoformat(updated_at)}
+            ovn_const.OVN_LIVENESS_CHECK_EXT_ID_KEY:
+                datetime.datetime.isoformat(updated_at)}
         if agent_type == ovn_const.OVN_METADATA_AGENT:
-            chassis.external_ids.update({
+            chassis_private.external_ids.update({
                 ovn_const.OVN_AGENT_METADATA_SB_CFG_KEY: nb_cfg,
                 ovn_const.METADATA_LIVENESS_CHECK_EXT_ID_KEY:
                 datetime.datetime.isoformat(updated_at)})
+        chassis_private.chassis = [chassis_private]
 
-        return neutron_agent.NeutronAgent.from_type(agent_type, chassis)
+        return neutron_agent.NeutronAgent.from_type(
+            agent_type, chassis_private)
 
     def test_agent_alive_true(self):
         for agent_type in (ovn_const.OVN_CONTROLLER_AGENT,