Cache database and message queue connection objects

Recently in the gate we have seen a trace on some work-in-progress
patches:

  OperationalError: (pymysql.err.OperationalError)
                    (1040, u'Too many connections')

and at least one operator has reported that the number of database
connections increased significantly going from Mitaka to Newton.

It was suspected that the increase was caused by creating new oslo.db
transaction context managers on-the-fly when switching database
connections for cells. Comparing the dstat --tcp output of runs of the
gate-tempest-dsvm-neutron-full-ubuntu-xenial job with and without
caching of the database connections showed a difference of 445 active
TCP connections and 1495 active TCP connections, respectively [1].

This adds caching of the oslo.db transaction context managers and the
oslo.messaging transports to avoid creating a large number of objects
that are not being garbage-collected as expected.

Closes-Bug: #1691545

[1] https://docs.google.com/spreadsheets/d/1DIfFfX3kaA_SRoCM-aO7BN4IBEShChXLztOBFeKryt4/edit?usp=sharing

 Conflicts:
	nova/test.py
	nova/tests/unit/test_context.py

NOTE(melwitt): Conflicts caused by the fact that no other global cache
resets exist in nova.test.py in Newton and the get_context function
doesn't exist in Newton.

Change-Id: I17e0eb836dd87aac5859f506e7d771d42753d31a
(cherry picked from commit f4159d1755)

nova/context.py

@@ -34,6 +34,10 @@ from nova import policy
 from nova import utils
 
 LOG = logging.getLogger(__name__)
+# TODO(melwitt): This cache should be cleared whenever WSGIService receives a
+# SIGHUP and periodically based on an expiration time. Currently, none of the
+# cell caches are purged, so neither is this one, for now.
+CELL_CACHE = {}
 
 
 class _ContextAuthPlugin(plugin.BaseAuthPlugin):
@@ -349,11 +353,27 @@ def target_cell(context, cell_mapping):
     :param context: The RequestContext to add connection information
     :param cell_mapping: A objects.CellMapping object
     """
+    global CELL_CACHE
+
     original_db_connection = context.db_connection
     # avoid circular import
     from nova import db
-    db_connection_string = cell_mapping.database_connection
+
-    context.db_connection = db.create_context_manager(db_connection_string)
+    # Synchronize access to the cache by multiple API workers.
+    @utils.synchronized(cell_mapping.uuid)
+    def get_or_set_cached_cell_and_set_connections():
+        try:
+            cell_db_conn = CELL_CACHE[cell_mapping.uuid]
+        except KeyError:
+            db_connection_string = cell_mapping.database_connection
+            context.db_connection = db.create_context_manager(
+                db_connection_string)
+            CELL_CACHE[cell_mapping.uuid] = context.db_connection
+        else:
+            context.db_connection = cell_db_conn
+
+    get_or_set_cached_cell_and_set_connections()
+
     try:
         yield context
     finally:

nova/test.py

@@ -211,6 +211,9 @@ class TestCase(testtools.TestCase):
         self.useFixture(conf_fixture.ConfFixture(CONF))
         self.useFixture(nova_fixtures.RPCFixture('nova.test'))
 
+        # NOTE(melwitt): Reset the cached db connection objects
+        context.CELL_CACHE = {}
+
         if self.USES_DB:
             self.useFixture(nova_fixtures.Database())
             self.useFixture(nova_fixtures.Database(database='api'))

nova/tests/unit/test_context.py

@@ -20,6 +20,7 @@ from nova import context
 from nova import exception
 from nova import objects
 from nova import test
+from nova.tests import uuidsentinel as uuids
 
 
 class ContextTestCase(test.NoDBTestCase):
@@ -286,7 +287,26 @@ class ContextTestCase(test.NoDBTestCase):
                                       roles=['admin', 'weasel'])
         # Verify the existing db_connection, if any, is restored
         ctxt.db_connection = mock.sentinel.db_conn
-        mapping = objects.CellMapping(database_connection='fake://')
+        mapping = objects.CellMapping(database_connection='fake://',
+                                      transport_url='fake://',
+                                      uuid=uuids.cell)
         with context.target_cell(ctxt, mapping):
             self.assertEqual(ctxt.db_connection, mock.sentinel.cm)
         self.assertEqual(mock.sentinel.db_conn, ctxt.db_connection)
+
+    @mock.patch('nova.db.create_context_manager')
+    def test_target_cell_caching(self, mock_create_cm):
+        mock_create_cm.return_value = mock.sentinel.db_conn_obj
+        ctxt = context.RequestContext('111', '222')
+        mapping = objects.CellMapping(database_connection='fake://db',
+                                      transport_url='fake://mq',
+                                      uuid=uuids.cell)
+        # First call should create new connection objects.
+        with context.target_cell(ctxt, mapping):
+            self.assertEqual(mock.sentinel.db_conn_obj, ctxt.db_connection)
+        mock_create_cm.assert_called_once_with('fake://db')
+        # Second call should use cached objects.
+        mock_create_cm.reset_mock()
+        with context.target_cell(ctxt, mapping):
+            self.assertEqual(mock.sentinel.db_conn_obj, ctxt.db_connection)
+        mock_create_cm.assert_not_called()

releasenotes/notes/bug-1691545-1acd6512effbdffb.yaml

@@ -0,0 +1,10 @@
+---
+fixes:
+  - |
+    Fixes `bug 1691545`_ in which there was a significant increase in database
+    connections because of the way connections to cell databases were being
+    established. With this fix, objects related to database connections are
+    cached in the API service and reused to prevent new connections being
+    established for every communication with cell databases.
+
+    .. _bug 1691545: https://bugs.launchpad.net/nova/+bug/1691545