PYTHON-2236 Reset the server pool only after marking the server Unknown

pymongo/monitor.py

@@ -142,10 +142,12 @@ class Monitor(MonitorBase):
             if self._publish:
                 self._listeners.publish_server_heartbeat_failed(
                     address, error_time, error)
-            self._topology.reset_pool(address)
             default = ServerDescription(address, error=error)
+            # Reset the server pool only after marking the server Unknown.
+            self._topology.on_change(default)
+            self._topology.reset_pool(address)
+            self._avg_round_trip_time.reset()
             if not retry:
-                self._avg_round_trip_time.reset()
                 # Server type defaults to Unknown.
                 return default
 

test/test_client.py

@@ -1767,7 +1767,7 @@ class TestMongoClientFailover(MockClientTest):
             host='b:2',  # Pass a secondary.
             replicaSet='rs',
             retryReads=False,
-            serverSelectionTimeoutMS=100,
+            serverSelectionTimeoutMS=1000,
         )
         self.addCleanup(c.close)
 
@@ -1782,9 +1782,6 @@ class TestMongoClientFailover(MockClientTest):
         # ServerSelectionTimeoutError or AutoReconnect (from
         # MockPool.get_socket).
         self.assertRaises(AutoReconnect, c.db.collection.find_one)
-        # The second attempt always raises ServerSelectionTimeoutError.
-        self.assertRaises(ServerSelectionTimeoutError,
-                          c.db.collection.find_one)
 
         # But it can reconnect.
         c.revive_host('a:1')
@@ -1804,7 +1801,7 @@ class TestMongoClientFailover(MockClientTest):
                 replicaSet='rs',
                 connect=False,
                 retryReads=False,
-                serverSelectionTimeoutMS=100)
+                serverSelectionTimeoutMS=1000)
             self.addCleanup(c.close)
 
             # Set host-specific information so we can test whether it is reset.

test/test_topology.py

@@ -730,7 +730,8 @@ class TestTopologyErrors(TopologyTest):
                 if ismaster_count[0] in (1, 3):
                     return IsMaster({'ok': 1, 'maxWireVersion': 6}), 0
                 else:
-                    raise AutoReconnect('mock monitor error')
+                    raise AutoReconnect(
+                        'mock monitor error #%s' % (ismaster_count[0],))
 
         t = create_mock_topology(monitor_class=TestMonitor)
         server = wait_for_master(t)
@@ -738,10 +739,14 @@ class TestTopologyErrors(TopologyTest):
         self.assertEqual(SERVER_TYPE.Standalone,
                          server.description.server_type)
 
-        # Second ismaster call, then immediately the third.
+        # Second ismaster call.
         t.request_check_all()
-        self.assertEqual(3, ismaster_count[0])
+        # The third ismaster call (the immediate retry) happens sometime soon
+        # after the failed check triggered by request_check_all. Wait until
+        # the server becomes known again.
+        t.select_server(writable_server_selector, 0.250)
         self.assertEqual(SERVER_TYPE.Standalone, get_type(t, 'a'))
+        self.assertEqual(3, ismaster_count[0])
 
     def test_internal_monitor_error(self):
         exception = AssertionError('internal error')