Merge "Fix inactive session error in compute node creation" into stable/wallaby

Author: Zuul

nova/db/sqlalchemy/api.py

@@ -688,16 +688,29 @@ def compute_node_search_by_hypervisor(context, hypervisor_match):
 
 
 @pick_context_manager_writer
-def compute_node_create(context, values):
+def _compute_node_create(context, values):
     """Creates a new ComputeNode and populates the capacity fields
     with the most recent data.
     """
     convert_objects_related_datetimes(values)
 
     compute_node_ref = models.ComputeNode()
     compute_node_ref.update(values)
+    compute_node_ref.save(context.session)
+    return compute_node_ref
+
+
+# NOTE(mgoddard): We avoid decorating this with @pick_context_manager_writer,
+# so that we get a separate transaction in the exception handler. This avoids
+# an error message about inactive DB sessions during a transaction rollback.
+# See https://bugs.launchpad.net/nova/+bug/1853159.
+def compute_node_create(context, values):
+    """Creates a new ComputeNode and populates the capacity fields
+    with the most recent data. Will restore a soft deleted compute node if a
+    UUID has been explicitly requested.
+    """
     try:
-        compute_node_ref.save(context.session)
+        compute_node_ref = _compute_node_create(context, values)
     except db_exc.DBDuplicateEntry:
         with excutils.save_and_reraise_exception(logger=LOG) as err_ctx:
             # Check to see if we have a (soft) deleted ComputeNode with the

nova/tests/functional/regressions/test_bug_1853009.py

@@ -85,10 +85,6 @@ def test_node_rebalance_deleted_compute_node_race(self):
 
         # host_b[1]: Finds no compute record in RT. Tries to create one
         # (_init_compute_node).
-        # FIXME(mgoddard): This shows a traceback with SQL rollback due to
-        # soft-deleted node. The create seems to succeed but breaks the RT
-        # update for this node. See
-        # https://bugs.launchpad.net/nova/+bug/1853159.
         host_b.manager.update_available_resource(self.ctxt)
         self._assert_hypervisor_api(self.nodename, expected_host='host_b')
         # There should only be one resource provider (fake-node).
@@ -164,41 +160,12 @@ def test_node_rebalance_deleted_compute_node_race(self):
             self.ctxt, cn, cascade=True)
 
         # host_b[3]: Should recreate compute node and resource provider.
-        # FIXME(mgoddard): Resource provider not recreated here, due to
-        # https://bugs.launchpad.net/nova/+bug/1853159.
         host_b.manager.update_available_resource(self.ctxt)
 
         # Verify that the node was recreated.
         self._assert_hypervisor_api(self.nodename, 'host_b')
 
-        # But due to https://bugs.launchpad.net/nova/+bug/1853159 the compute
-        # node is not cached in the RT.
-        self.assertNotIn(self.nodename, host_b.manager.rt.compute_nodes)
-
-        # There is no RP.
-        rps = self._get_all_providers()
-        self.assertEqual(0, len(rps), rps)
-
-        # But the RP exists in the provider tree.
-        self.assertFalse(host_b.manager.rt.reportclient._provider_tree.exists(
-            self.nodename))
-
-        # host_b[1]: Should add compute node to RT cache and recreate resource
-        # provider.
-        host_b.manager.update_available_resource(self.ctxt)
-
-        # Verify that the node still exists.
-        self._assert_hypervisor_api(self.nodename, 'host_b')
-
-        # And it is now in the RT cache.
-        self.assertIn(self.nodename, host_b.manager.rt.compute_nodes)
-
         # The resource provider has now been created.
         rps = self._get_all_providers()
         self.assertEqual(1, len(rps), rps)
         self.assertEqual(self.nodename, rps[0]['name'])
-
-        # This fails due to the lack of a resource provider.
-        self.assertIn(
-            'Skipping removal of allocations for deleted instances',
-            self.stdlog.logger.output)

nova/tests/unit/db/test_db_api.py

@@ -5257,6 +5257,22 @@ def test_compute_node_create_duplicate_host_hypervisor_hostname(self):
         self.assertRaises(db_exc.DBDuplicateEntry,
                           db.compute_node_create, self.ctxt, other_node)
 
+    def test_compute_node_create_duplicate_uuid(self):
+        """Tests to make sure that no exception is raised when trying to create
+        a compute node with the same host, hypervisor_hostname and uuid values
+        as another compute node that was previously soft-deleted.
+        """
+        # Prior to fixing https://bugs.launchpad.net/nova/+bug/1853159, this
+        # raised the following error:
+        # sqlalchemy.exc.InvalidRequestError: This session is in 'inactive'
+        # state, due to the SQL transaction being rolled back; no further SQL
+        # can be emitted within this transaction.
+        constraint = db.constraint(host=db.equal_any(self.item['host']))
+        sqlalchemy_api.compute_node_delete(
+            self.ctxt, self.item['id'], constraint=constraint)
+        new_node = db.compute_node_create(self.ctxt, self.compute_node_dict)
+        self.assertEqual(self.item['uuid'], new_node['uuid'])
+
     def test_compute_node_get_all(self):
         nodes = db.compute_node_get_all(self.ctxt)
         self.assertEqual(1, len(nodes))