It would be nice if the Search API supported ?refresh like the Index/Update/Delete APIs, so that requests could block until a consistent result could be returned. Has this ever been discussed / proposed? Any other suggestions on how this might be implemented? I guess in theory you could "wait one second" and the near-realtime search would be in play, but looking for something more deterministic.
Also wondering what the boundary scenarios for the "1 second" refresh are - how dependent on the current load on the cluster e.g.