While diagnosing certain flows, we came across some clusters with _hundreds_ of aliases per index. We need to stop collecting large numbers of aliases.