I think start with this.
Later, we might look at doing same at other side of the ringbuffer. Currently it would be complicated to do. On other side of the ringbuffer, we just append the edit. The edit has disappeared into the WAL. The sync has not been called yet. Handlers are currently tied to their sync via the SyncFuture. We'd have to add a listing of a Handlers' seqid to the SyncFuture. It'd be messy currently.
This stuff will change again soon; e.g. the Himanshu switch writers and then multiwal hopefully. Maybe come back to this issue then. There may be more points we can make use of after these refactorings.