Today, we already leverage distinct scans for $group + $first/$last/$bottom/$top. We can do the same for accumulators like $topN even when N > 1. We prototyped this optimization in SKUNK-30 and saw huge improvements on Query High Value Workloads. When eligible, this is an optimization we should always do.

Required work:

• Extend distinct scan execution to accept parameter N
• Ensure $groupByDistinctScan works for cases where N > 1
• Extend planning logic to pick DISTINCT_N_SCAN in eligible cases