The current approach used by CBR sampling CE to estimate IntervalBounds is to scan the sample and generate keys out of it on each estimation call. This is very inefficient.

A potentially much more efficient approach is to transform each interval into a MatchExpression, and apply the expression directly to the sample. This task is about:

• Create equivalent MatchExpression (ME) from an IntervalBounds.
• Replace the calls to
  SamplingEstimator::estimateRIDs with SamplingEstimator::estimateCardinality.
• Benchmark and compare the performance of interval estimation via SamplingEstimator::estimateRIDs as a baseline, and compare that to the new estimation method based on Interval->ME conversion. Benchmarking can be done at least at two levels:
  • Microbenchmark estimation itself excluding/including the conversion Interval->ME, and
  • Run Genny multi-planner benchmarks (or similar).