I think this is problem of kubic workarounds in transactional-update tests. I created poo#31519 for this.