AttributeError in aioredis.RedisCluster when the connection is lost: 'ConnectionError' object has no attribute 'host'

[klymenkosergiy]

I noticed an unexpected behavior for aioredis.RedisCluster after losing a connection to the cluster node (the maintenance had been started by the service provider):

AttributeError:'ConnectionError' object has no attribute 'host'
File "<venv_path>/lib/python3.11/site-packages/redis/asyncio/cluster.py", line 787, in execute_command
raise e
File "<venv_path>/lib/python3.11/site-packages/redis/asyncio/cluster.py", line 748, in execute_command
target_nodes = await self._determine_nodes(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "<venv_path>/lib/python3.11/site-packages/redis/asyncio/cluster.py", line 624, in _determine_nodes
self.nodes_manager.get_node_from_slot(
File "<venv_path>/lib/python3.11/site-packages/redis/asyncio/cluster.py", line 1342, in get_node_from_slot
self._update_moved_slots()
File "<venv_path>/lib/python3.11/site-packages/redis/asyncio/cluster.py", line 1299, in _update_moved_slots
redirected_node = self.get_node(host=e.host, port=e.port)
^^^^^^

Python: 3.11
Redis-py: 6.4.0

[petyaslavova]

Hi @klymenkosergiy, thanks for notifying us! We will investigate the issue.

[petyaslavova]

Hi @klymenkosergiy,

I found a case where we might end up storing a connection error in the _moved_exception property. It seems that in certain scenarios—specifically when using pipelines with transactions—we may update this property even for connection-related errors.

Could you please confirm whether your application uses pipelines with transactions?

[klymenkosergiy]

Hi @petyaslavova,

As far as I can tell from the log, this exception was raised for all commands — GET, SET, and HSCAN — that were called by the execute_command function. The traceback looks the same for all of them.
I also use a pipeline with transaction, but I didn’t see any errors inside transactions in the log — probably because transactions start a bit later by design, and the endpoint execution failed before they began.

[petyaslavova]

@klymenkosergiy, thank you for your quick response!

It seems that the connection for the transaction was successfully established, but a ConnectionError occurred afterward. When the transaction strategy tries to handle it, the shared _moved_exception(

redis-py/redis/asyncio/cluster.py

Line 2208 in fff669d

self._pipe.cluster_client.nodes_manager.update_moved_exception(error)

) property ends up being assigned an unexpected exception type, causing subsequent command executions to start raising errors.

Since the code is asynchronous, one command execution or error handler can easily interfere with others, which likely explains the behavior you’ve observed.