Full threadhnov·In this case denormalizing the data right into the runs table would likely give you better, but more importantly more predictable performance.View on HN