I’ve been thinking of these ‘opportunities to get unlucky’ as bad luck tickets. On a database with 64 shards, the query is executed on each shard and then aggregated. But the latency of the aggregate query is the maximum latency of the 64 individual shard queries.