How Does TestCase's Async-to-Sync Conversion Cause Unreleased ChannelLayer Locks in Django Tests?

I recently encountered an issue using Django’s built-in TestCase for asynchronous Channels consumers tests

I added several complex tests that simulate multiple clients (up to 3) interacting concurrently. Suddenly, my test suite began intermittently failing with the following RuntimeError:

PYTHONCopied
RuntimeError: <asyncio.locks.Lock object at 0x77eba260dae0 [locked]> is bound to a different event loop

Upon further investigation, I discovered that this error appears only in a specific test-case (containing 12 tests involving consumers/channels) and is much more frequent when running only that test-case (approximately 4 out of 5 cases) compared to running the entire suite (around 1 out of 10 cases). When tests are run individually, the error does not occur.

By adding debug logs to the acquire and release methods in the ChannelLock class, I noticed that occasionally a test completes with an unreleased lock, which isn’t cleaned up by the channel layer.

I suspect the problem is related to Django TestCase wrapping each async test in async_to_sync, thereby creating a new event loop from a synchronous context, potentially causing the channel layer to hold onto stale locks. I already attempted using RedisChannelLayer.flush in the setUp and tearDown methods to address the issue, but the problem persisted. Currently, I’m forcing the layer to rebuild in setUp to reset the locks.

Has anyone experienced similar behavior or have suggestions for identifying the root cause, which might be related to timing issues?