[FLINK-21946] FlinkRelMdUtil.numDistinctVals produces exceptional Double.NaN result when domainSize is in range(0,1)

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 1.11.3, 1.12.2
• Fix Version/s: 1.13.0
• Component/s: Table SQL / Planner
• Labels: pull-request-available
• Assignee: lincoln lee
• Reporter: lincoln lee
• Created: 24/Mar/21 08:45
• Updated: 26/Mar/21 02:04
• Resolved: 26/Mar/21 02:04

Description

CALCITE-4351 introduced a new 'more accurate' formula for estimating the number of distinct values, FLINK-19780 fixed one bad case when number of rows are large, but still there's exception path when domainSize is in range(0,1), the following code in calcite‘s RelMdUtil.numDistinctVals will produce Double.NaN result:

double expo = numSel * Math.log(1.0 - 1.0 / dSize);

Before CALCITE-4351 was fixed, we should continue the fixing in flink.

Issue Links

• is caused by: CALCITE-4351 RelMdUtil#numDistinctVals always returns 0 for large inputs
• links to: GitHub Pull Request #15357