Merge "[OVN] Hash Ring: Set nodes as offline upon exit" into stable/zed

neutron/cmd/ovn/neutron_ovn_db_sync_util.py

@@ -58,12 +58,12 @@ def post_fork_initialize(self, resource, event, trigger, **kwargs):
     def ovn_client(self):
         return self._ovn_client
 
-    def _clean_hash_ring(self):
-        """Don't clean the hash ring.
+    def _set_hash_ring_nodes_offline(self):
+        """Don't set hash ring nodes as offline.
 
         If this method was not overridden, cleanup would be performed when
-        calling the db sync and running neutron server would lose all the nodes
-        from the ring.
+        calling the db sync and running neutron server would mark all the
+        nodes from the ring as offline.
         """
 
     # Since we are not using the ovn mechanism driver while syncing,

neutron/db/ovn_hash_ring_db.py

@@ -50,10 +50,12 @@ def remove_nodes_from_host(context, group_name):
              CONF.host, group_name)
 
 
-def _touch(context, **filter_args):
+def _touch(context, updated_at=None, **filter_args):
+    if updated_at is None:
+        updated_at = timeutils.utcnow()
     with db_api.CONTEXT_WRITER.using(context):
         context.session.query(ovn_models.OVNHashRing).filter_by(
-            **filter_args).update({'updated_at': timeutils.utcnow()})
+            **filter_args).update({'updated_at': updated_at})
 
 
 def touch_nodes_from_host(context, group_name):
@@ -92,3 +94,9 @@ def get_active_nodes(context, interval, group_name, from_host=False):
 def count_offline_nodes(context, interval, group_name):
     query = _get_nodes_query(context, interval, group_name, offline=True)
     return query.count()
+
+
+def set_nodes_from_host_as_offline(context, group_name):
+    timestamp = datetime.datetime(day=26, month=10, year=1985, hour=9)
+    _touch(context, updated_at=timestamp, hostname=CONF.host,
+           group_name=group_name)

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

@@ -278,15 +278,17 @@ def subscribe(self):
                                resources.SECURITY_GROUP_RULE,
                                events.BEFORE_DELETE)
 
-    def _clean_hash_ring(self, *args, **kwargs):
+    def _set_hash_ring_nodes_offline(self, *args, **kwargs):
         admin_context = n_context.get_admin_context()
-        ovn_hash_ring_db.remove_nodes_from_host(admin_context,
-                                                self.hash_ring_group)
+        ovn_hash_ring_db.set_nodes_from_host_as_offline(
+            admin_context, self.hash_ring_group)
+        LOG.info('Hash Ring nodes from host "%s" marked as offline',
+                 cfg.CONF.host)
 
     def pre_fork_initialize(self, resource, event, trigger, payload=None):
         """Pre-initialize the ML2/OVN driver."""
-        atexit.register(self._clean_hash_ring)
-        signal.signal(signal.SIGTERM, self._clean_hash_ring)
+        atexit.register(self._set_hash_ring_nodes_offline)
+        signal.signal(signal.SIGTERM, self._set_hash_ring_nodes_offline)
         ovn_utils.create_neutron_pg_drop()
 
     @staticmethod
@@ -306,7 +308,9 @@ def _setup_hash_ring(self):
         """
         admin_context = n_context.get_admin_context()
         if not self._hash_ring_probe_event.is_set():
-            self._clean_hash_ring()
+            # Clear existing entries
+            ovn_hash_ring_db.remove_nodes_from_host(admin_context,
+                                                    self.hash_ring_group)
             self.node_uuid = ovn_hash_ring_db.add_node(admin_context,
                                                        self.hash_ring_group)
             self._hash_ring_thread = maintenance.MaintenanceThread()

neutron/tests/functional/base.py

@@ -362,7 +362,8 @@ def trigger(self):
         self.addCleanup(self.stop)
         # NOTE(ralonsoh): do not access to the DB at exit when the SQL
         # connection is already closed, to avoid useless exception messages.
-        mock.patch.object(self.mech_driver, '_clean_hash_ring').start()
+        mock.patch.object(
+            self.mech_driver, '_set_hash_ring_nodes_offline').start()
         self.mech_driver.pre_fork_initialize(
             mock.ANY, mock.ANY, trigger_cls.trigger)
 

neutron/tests/unit/db/test_ovn_hash_ring_db.py

@@ -269,3 +269,17 @@ def test_count_offline_nodes(self):
         # Assert no nodes are considered offline
         self.assertEqual(0, ovn_hash_ring_db.count_offline_nodes(
             self.admin_ctx, interval=60, group_name=HASH_RING_TEST_GROUP))
+
+    def test_set_nodes_from_host_as_offline(self):
+        self._add_nodes_and_assert_exists(count=3)
+
+        active_nodes = ovn_hash_ring_db.get_active_nodes(
+            self.admin_ctx, interval=60, group_name=HASH_RING_TEST_GROUP)
+        self.assertEqual(3, len(active_nodes))
+
+        ovn_hash_ring_db.set_nodes_from_host_as_offline(
+                self.admin_ctx, HASH_RING_TEST_GROUP)
+
+        active_nodes = ovn_hash_ring_db.get_active_nodes(
+            self.admin_ctx, interval=60, group_name=HASH_RING_TEST_GROUP)
+        self.assertEqual(0, len(active_nodes))