Fix coordinator join_future race condition (#1338)

* Fix race condition in coordinator join_future handling

Author: dpkp

kafka/coordinator/base.py

@@ -377,19 +377,23 @@ def ensure_active_group(self):
                 # before the pending rebalance has completed.
                 if self.join_future is None:
                     self.state = MemberState.REBALANCING
-                    self.join_future = self._send_join_group_request()
+                    future = self._send_join_group_request()
+
+                    self.join_future = future  # this should happen before adding callbacks
 
                     # handle join completion in the callback so that the
                     # callback will be invoked even if the consumer is woken up
                     # before finishing the rebalance
-                    self.join_future.add_callback(self._handle_join_success)
+                    future.add_callback(self._handle_join_success)
 
                     # we handle failures below after the request finishes.
                     # If the join completes after having been woken up, the
                     # exception is ignored and we will rejoin
-                    self.join_future.add_errback(self._handle_join_failure)
+                    future.add_errback(self._handle_join_failure)
+
+                else:
+                    future = self.join_future
 
-                future = self.join_future
                 self._client.poll(future=future)
 
                 if future.failed():

test/test_coordinator.py

@@ -620,3 +620,16 @@ def test_lookup_coordinator_failure(mocker, coordinator):
                         return_value=Future().failure(Exception('foobar')))
     future = coordinator.lookup_coordinator()
     assert future.failed()
+
+
+def test_ensure_active_group(mocker, coordinator):
+    coordinator._subscription.subscribe(topics=['foobar'])
+    mocker.patch.object(coordinator, 'coordinator_unknown', return_value=False)
+    mocker.patch.object(coordinator, '_send_join_group_request', return_value=Future().success(True))
+    mocker.patch.object(coordinator, 'need_rejoin', side_effect=[True, True, False])
+    mocker.patch.object(coordinator, '_on_join_complete')
+    mocker.patch.object(coordinator, '_heartbeat_thread')
+
+    coordinator.ensure_active_group()
+
+    coordinator._send_join_group_request.assert_called_once_with()