First of all this would be a breaking change. Second, I think it is too late to make this change. This behavior is now part of the command's contract.

@hpatro @madolson thoughts?

We discussed this in Redis and we were planning on breaking it for Redis 8.

redis/redis#12228

My comment at the time was that I viewed this more as a bug because cluster routing wouldn't work for some clients. If you sent PUBSUB SHARDNUMBSUB <foo>, some clients, like go, wouldn't know where to send it. My tenet is we should be okay with breaking changes that are actively harming end users, and I think that is the case here.

First of all this would be a breaking change. Second, I think it is too late to make this change. This behavior is now part of the command's contract.

@hpatro @madolson thoughts?

Sharded pubsub isn't that well adopted yet. We still have some room to smooth rough edges until the wrong behavior becomes the defacto. I'm cool with fixing the behavior for multi channel input.

SGTM - thanks for the context

Sharded pubsub isn't that well adopted yet. We still have some room to smooth rough edges until the wrong behavior becomes the defacto. I'm cool with fixing the behavior for multi channel input.

This is very much my view too.

That's why I'm OK with other things for sharded pubsub, like using replication link, but I don't think it's OK to do such changes for regular pubsub.