Part of #12994, we would like a stream that contains un-partial-stated events.

Once we un-partial-state an event, we need to tell other workers about it, so that they can:

• flush their cache for is_partial_state_event
• flush their cache for _get_state_group_for_event
• if the event changed rejection status, do _invalidate_local_get_event_cache
• unblock any requests which are waiting for full state at that event

I think this probably "just" wants to be a new replication stream listing the event ids, and the old and new rejection status, of any un-partial-stated events. This can then be used to flush the caches and to call _state_storage_controller.notify_event_un_partial_stated.

I think most of the infrastructure for the stream is already present (in synapse.replication.tcp.streams). We'll need to define a new stream type, and add a new stream table to contain the data.

— #12994 (comment)

It might be interesting to ponder #13456 — getting rid of stream tables — but also that's likely to be too much of a distraction here (and I've had a little ponder and can't convince myself of any clean alternatives, at the moment).

Bear in mind the stream may have multiple writers in the future.