More accurate agent restart state transfer

Ovs-agent can be very time-consuming in handling a large number
of ports. At this point, the ovs-agent status report may have
exceeded the set timeout value. Some flows updating operations
will not be triggerred. This results in flows loss during agent
restart, especially for hosts to hosts of vxlan tunnel flow.

This fix will let the ovs-agent explicitly, in the first rpc loop,
indicate that the status is restarted. Then l2pop will be required
to update fdb entries.

Conflicts:
	neutron/plugins/ml2/rpc.py

Conflicts:
	neutron/plugins/ml2/drivers/l2pop/mech_driver.py

Closes-Bug: #1813703
Closes-Bug: #1813714
Closes-Bug: #1813715
Closes-Bug: #1794991
Closes-Bug: #1799178

Change-Id: I8edc2deb509216add1fb21e1893f1c17dda80961
(cherry picked from commit a5244d6d44)
(cherry picked from commit cc49ab5501)
(cherry picked from commit 5ffca49668)

neutron/agent/rpc.py

@@ -139,11 +139,12 @@ class PluginApi(object):
                           agent_id=agent_id, host=host)
 
     def update_device_list(self, context, devices_up, devices_down,
-                           agent_id, host):
+                           agent_id, host, agent_restarted=False):
         cctxt = self.client.prepare(version='1.5')
         return cctxt.call(context, 'update_device_list',
                           devices_up=devices_up, devices_down=devices_down,
-                          agent_id=agent_id, host=host)
+                          agent_id=agent_id, host=host,
+                          agent_restarted=agent_restarted)
 
     def tunnel_sync(self, context, tunnel_ip, tunnel_type=None, host=None):
         cctxt = self.client.prepare(version='1.4')

neutron/plugins/ml2/drivers/l2pop/mech_driver.py

@@ -245,6 +245,19 @@ class L2populationMechanismDriver(api.MechanismDriver):
 
         return agents
 
+    def agent_restarted(self, context):
+        agent_host = context.host
+        session = db_api.get_reader_session()
+        agent = l2pop_db.get_agent_by_host(session, agent_host)
+        if l2pop_db.get_agent_uptime(agent) < cfg.CONF.l2pop.agent_boot_time:
+            LOG.warning(_LW("Agent on host '%s' did not supply "
+                            "'agent_restarted'information in RPC message, "
+                            "determined it restarted based on deprecated "
+                            "'agent_boot_time' config option."),
+                        agent_host)
+            return True
+        return False
+
     def update_port_down(self, context):
         port = context.current
         agent_host = context.host
@@ -265,7 +278,7 @@ class L2populationMechanismDriver(api.MechanismDriver):
             self.L2populationAgentNotify.remove_fdb_entries(
                 self.rpc_ctx, fdb_entries)
 
-    def update_port_up(self, context):
+    def update_port_up(self, context, agent_restarted=None):
         port = context.current
         agent_host = context.host
         session = db_api.get_reader_session()
@@ -291,8 +304,10 @@ class L2populationMechanismDriver(api.MechanismDriver):
         # with high concurrency more than 1 port may be activated on an agent
         # at the same time (like VM port + a DVR port) so checking for 1 or 2
         is_first_port = agent_active_ports in (1, 2)
-        if is_first_port or (l2pop_db.get_agent_uptime(agent) <
-                             cfg.CONF.l2pop.agent_boot_time):
+        if agent_restarted is None:
+            # Only for backport compatibility, will be removed.
+            agent_restarted = self.agent_restarted(context)
+        if is_first_port or agent_restarted:
             # First port(s) activated on current agent in this network,
             # we have to provide it with the whole list of fdb entries
             agent_fdb_entries = self._create_agent_fdb(session,

neutron/plugins/ml2/drivers/openvswitch/agent/ovs_neutron_agent.py

@@ -879,9 +879,15 @@ class OVSNeutronAgent(sg_rpc.SecurityGroupAgentRpcCallbackMixin,
                 LOG.debug("Setting status for %s to DOWN", device)
                 devices_down.append(device)
         if devices_up or devices_down:
+            # When the iter_num == 0, that indicate the ovs-agent is doing
+            # the initialization work. L2 pop needs this precise knowledge
+            # to notify the agent to refresh the tunnel related flows.
+            # Otherwise, these flows will be cleaned as stale due to the
+            # different cookie id.
+            agent_restarted = self.iter_num == 0
             devices_set = self.plugin_rpc.update_device_list(
                 self.context, devices_up, devices_down, self.agent_id,
-                self.conf.host)
+                self.conf.host, agent_restarted=agent_restarted)
             failed_devices = (devices_set.get('failed_devices_up') +
                 devices_set.get('failed_devices_down'))
             if failed_devices:

neutron/plugins/ml2/rpc.py

@@ -241,6 +241,7 @@ class RpcCallbacks(type_tunnel.TunnelRpcCallbackMixin):
         agent_id = kwargs.get('agent_id')
         device = kwargs.get('device')
         host = kwargs.get('host')
+        agent_restarted = kwargs.pop('agent_restarted', None)
         LOG.debug("Device %(device)s up at agent %(agent_id)s",
                   {'device': device, 'agent_id': agent_id})
         plugin = directory.get_plugin()
@@ -267,7 +268,8 @@ class RpcCallbacks(type_tunnel.TunnelRpcCallbackMixin):
         else:
             self.update_port_status_to_active(port, rpc_context, port_id, host)
         self.notify_l2pop_port_wiring(port_id, rpc_context,
-                                      n_const.PORT_STATUS_ACTIVE, host)
+                                      n_const.PORT_STATUS_ACTIVE, host,
+                                      agent_restarted)
 
     def update_port_status_to_active(self, port, rpc_context, port_id, host):
         plugin = directory.get_plugin()
@@ -291,7 +293,7 @@ class RpcCallbacks(type_tunnel.TunnelRpcCallbackMixin):
                 provisioning_blocks.L2_AGENT_ENTITY)
 
     def notify_l2pop_port_wiring(self, port_id, rpc_context,
-                                 status, host):
+                                 status, host, agent_restarted=None):
         """Notify the L2pop driver that a port has been wired/unwired.
 
         The L2pop driver uses this notification to broadcast forwarding
@@ -302,25 +304,38 @@ class RpcCallbacks(type_tunnel.TunnelRpcCallbackMixin):
                 'l2population')
         if not l2pop_driver:
             return
+        port = ml2_db.get_port(rpc_context, port_id)
+        if not port:
+            return
         port_context = plugin.get_bound_port_context(
-                rpc_context, port_id)
+                rpc_context, port_id, host)
         if not port_context:
             # port deleted
             return
+        # NOTE: DVR ports are already handled and updated through l2pop
+        # and so we don't need to update it again here. But, l2pop did not
+        # handle DVR ports while restart neutron-*-agent, we need to handle
+        # it here.
+        if agent_restarted is None:
+            agent_restarted = l2pop_driver.obj.agent_restarted(port_context)
+        if (port['device_owner'] == n_const.DEVICE_OWNER_DVR_INTERFACE and
+                not agent_restarted):
+            return
         port = port_context.current
-        if (status == n_const.PORT_STATUS_ACTIVE and
+        if (port['device_owner'] != n_const.DEVICE_OWNER_DVR_INTERFACE and
+            status == n_const.PORT_STATUS_ACTIVE and
             port[portbindings.HOST_ID] != host and
             not l3_hamode_db.is_ha_router_port(rpc_context,
                                                port['device_owner'],
                                                port['device_id'])):
                 # don't setup ACTIVE forwarding entries unless bound to this
-                # host or if it's an HA port (which is special-cased in the
-                # mech driver)
+                # host or if it's an HA or DVR port (which is special-cased in
+                # the mech driver)
                 return
         port_context.current['status'] = status
         port_context.current[portbindings.HOST_ID] = host
         if status == n_const.PORT_STATUS_ACTIVE:
-            l2pop_driver.obj.update_port_up(port_context)
+            l2pop_driver.obj.update_port_up(port_context, agent_restarted)
         else:
             l2pop_driver.obj.update_port_down(port_context)
 

neutron/tests/functional/agent/l2/base.py

@@ -267,7 +267,7 @@ class OVSAgentTestFramework(base.BaseOVSLinuxTestCase):
         return ports
 
     def _mock_update_device(self, context, devices_up, devices_down, agent_id,
-                            host=None):
+                            host=None, agent_restarted=False):
         dev_up = []
         dev_down = []
         for port in self.ports:
@@ -311,7 +311,8 @@ class OVSAgentTestFramework(base.BaseOVSLinuxTestCase):
     def _prepare_failed_dev_up_trigger(self, agent):
 
         def mock_failed_devices_up(context, devices_up, devices_down,
-                                   agent_id, host=None):
+                                   agent_id, host=None,
+                                   agent_restarted=False):
             failed_devices = []
             devices = list(devices_up)
             # first port fails
@@ -332,7 +333,8 @@ class OVSAgentTestFramework(base.BaseOVSLinuxTestCase):
     def _prepare_failed_dev_down_trigger(self, agent):
 
         def mock_failed_devices_down(context, devices_up, devices_down,
-                                     agent_id, host=None):
+                                     agent_id, host=None,
+                                     agent_restarted=False):
             # first port fails
             failed_port_id = self.ports[0]['id']
             failed_devices_down = []

neutron/tests/unit/plugins/ml2/drivers/l2pop/test_mech_driver.py

@@ -338,6 +338,49 @@ class TestL2PopulationRpcTestCase(test_plugin.Ml2PluginV2TestCase):
                 self.mock_fanout.assert_called_with(
                     mock.ANY, 'remove_fdb_entries', expected)
 
+    def _test_ovs_agent_restarted_with_dvr_port(
+            self, agent_boot_timeout=True, agent_restarted=False):
+        plugin = directory.get_plugin()
+        self._setup_l3()
+        router = self._create_router(distributed=True)
+        with mock.patch.object(l2pop_mech_driver.L2populationMechanismDriver,
+                               'agent_restarted',
+                               return_value=agent_boot_timeout):
+            with self.subnet(network=self._network,
+                             enable_dhcp=False) as snet:
+                with self.port(
+                        subnet=snet,
+                        device_owner=constants.DEVICE_OWNER_DVR_INTERFACE)\
+                            as port:
+                    port_id = port['port']['id']
+                    plugin.update_distributed_port_binding(self.adminContext,
+                        port_id, {'port': {portbindings.HOST_ID: HOST_4,
+                        'device_id': router['id']}})
+                    port = self._show('ports', port_id)
+                    self.assertEqual(portbindings.VIF_TYPE_DISTRIBUTED,
+                                    port['port'][portbindings.VIF_TYPE])
+                    self.callbacks.update_device_up(
+                        self.adminContext,
+                        agent_id=HOST_4,
+                        device=port_id,
+                        host=HOST_4,
+                        agent_restarted=agent_restarted)
+                    fanout_expected = {port['port']['network_id']: {
+                        'network_type': u'vxlan',
+                        'ports': {
+                            u'20.0.0.4': [('00:00:00:00:00:00', '0.0.0.0')]},
+                        'segment_id': 1}}
+                    self.mock_fanout.assert_called_with(mock.ANY,
+                                                        'add_fdb_entries',
+                                                        fanout_expected)
+
+    def test_ovs_agent_restarted_with_dvr_port_boot_config_timeout(self):
+        self._test_ovs_agent_restarted_with_dvr_port()
+
+    def test_ovs_agent_restarted_with_dvr_port_rpc_send_timeout(self):
+        self._test_ovs_agent_restarted_with_dvr_port(
+            agent_boot_timeout=False, agent_restarted=True)
+
     def test_ha_agents_get_other_fdb(self):
         # First network port is added on HOST4, then HA router port is
         # added on HOST and HOST2.

neutron/tests/unit/plugins/ml2/drivers/openvswitch/agent/test_ovs_neutron_agent.py

@@ -720,7 +720,8 @@ class TestOvsNeutronAgent(object):
             self.agent._bind_devices(port_details)
             update_devices.assert_called_once_with(mock.ANY, devices_up,
                                                    devices_down,
-                                                   mock.ANY, mock.ANY)
+                                                   mock.ANY, mock.ANY,
+                                                   agent_restarted=True)
 
     def _test_arp_spoofing(self, enable_prevent_arp_spoofing):
         self.agent.prevent_arp_spoofing = enable_prevent_arp_spoofing

neutron/tests/unit/plugins/ml2/test_rpc.py

@@ -445,6 +445,7 @@ class RpcApiTestCase(base.BaseTestCase):
                            devices_down=['fake_device3', 'fake_device4'],
                            agent_id='fake_agent_id',
                            host='fake_host',
+                           agent_restarted=False,
                            version='1.5')
 
     def test_get_devices_details_list_and_failed_devices(self):

releasenotes/notes/precise-agent-state-transfer-67c771cb1ee04dd0.yaml

@@ -0,0 +1,27 @@
+---
+critical:
+  - |
+    The neutron-openvswitch-agent can sometimes spend too much time handling
+    a large number of ports, exceeding its timeout value, ``agent_boot_time``,
+    for L2 population. Because of this, some flow update operations will not
+    be triggerred, resulting in lost flows during agent restart, especially
+    for host-to-host vxlan tunnel flows, causing the original tunnel flows to
+    be treated as stale due to the different cookie IDs. The agent's first
+    RPC loop will also do a stale flow clean-up procedure and delete them,
+    leading to a loss of connectivity.
+    Please ensure that all neutron-server and neutron-openvswitch-agent
+    binaries are upgraded for the changes to take effect, after which
+    the L2 population ``agent_boot_time`` config option will no longer
+    be used.
+fixes:
+  - |
+    The neutron-openvswitch-agent was changed to notify the neutron-server
+    in its first RPC loop that it has restarted. This signals neutron-server
+    to provide updated L2 population information to correctly program FDB
+    entries, ensuring connectivity to instances is not interrupted.
+    This fixes the following bugs:
+    `1794991 <https://bugs.launchpad.net/neutron/+bug/1794991>`_,
+    `1799178 <https://bugs.launchpad.net/neutron/+bug/1799178>`_,
+    `1813703 <https://bugs.launchpad.net/neutron/+bug/1813703>`_,
+    `1813714 <https://bugs.launchpad.net/neutron/+bug/1813714>`_,
+    `1813715 <https://bugs.launchpad.net/neutron/+bug/1813715>`_.