Why do we need to provide the pubsub reference here? If a failover has occurred, shouldn’t _active_pubsub already contain all the necessary information? By passing in this reference, don’t we risk holding onto a pubsub object from a failed instance and potentially trying to work with it?

We need currently active pub/sub reference to be able to execute any commands on it in any point of time (f.e subscribe to more channels if needed or unsubscribe), if MultiDbClient is in Pub/Sub mode. Also we keep it to be able to re-initialise new pub/sub during failover with the same configuration and re-subscribe to the same channels.

We always update reference when failover happens, so we do not risk to keep stale reference.

https://github.com/redis/redis-py/pull/3722/files#diff-77af8b3b216196f2b0569279fe1c6d9db146e355eba207eeb84021ccf54c3c0eR54

Yes, and it handles the state of self._active_pubsub for which we call run_in_thread, but then a pubsub object is also provided as an argument - I was asking about this one.

What do we expect to have happened in the pubsub_thread? Would it be worth validating whether it completed successfully or failed?

In pubsub_thread we're validating if we receive expected message from subscribed channel

https://github.com/redis/redis-py/pull/3722/files/bee15d943e3c381ad1b0e9cebe6061cbce973d16#diff-eccf1b0c83ba8b8e9d8dc7e837ba57e529ec23df09f3d6f43f354fdf5584fcf7R281

Do you think we need to verify something else? Like counter to prove that it was called > 0 times?

Maybe a check of the thread object results. The thread might have raised an exception due to this assert failure, but unless the results of the thread object are checked, we might never see that something has failed.