diff --git a/pykafka/broker.py b/pykafka/broker.py
index bfcdfd15c..e06e01fcb 100644
--- a/pykafka/broker.py
+++ b/pykafka/broker.py
@@ -294,7 +294,16 @@ def _get_unique_req_handler(self, connection_id):
             handler = RequestHandler(self._handler, conn)
             handler.start()
             self._req_handlers[connection_id] = handler
-        return self._req_handlers[connection_id]
+
+        handler = self._req_handlers[connection_id]
+
+        # Ensure that we're returning a handler with a connected connection.
+        # If the connection is disconnected, it will raise SocketDisconnectedError
+        if not handler.shared.connection.connected:
+            log.warn('Attempting to reconnect for connection id %s..', connection_id)
+            handler.shared.connection.connect(self._socket_timeout_ms)
+
+        return handler
 
     @_check_handler
     def fetch_messages(self,
diff --git a/pykafka/managedbalancedconsumer.py b/pykafka/managedbalancedconsumer.py
index fb4613ec8..c2081ee95 100644
--- a/pykafka/managedbalancedconsumer.py
+++ b/pykafka/managedbalancedconsumer.py
@@ -324,10 +324,10 @@ def _update_member_assignment(self):
                 log.debug("Successfully rebalanced consumer '%s'", self._consumer_id)
                 break
             except Exception as ex:
+                log.exception(ex)
                 if i == self._rebalance_max_retries - 1:
                     log.warning('Failed to rebalance s after %d retries.', i)
                     raise
-                log.exception(ex)
                 log.info('Unable to complete rebalancing. Retrying')
                 self._cluster.handler.sleep(i * (self._rebalance_backoff_ms / 1000))
         self._raise_worker_exceptions()