Yesterday I've noticed that, in a Cortex cluster with 3 ingesters and a replication factor of 3, during the ingesters rollout there are gaps in queried series (for the series not flushed to storage yet). My gut feeling is that this is a bug in the ring replication strategy, but further investigation needs to be done.