Range estimate not to go below equivalent point get estimate

[terry1purcell]

Enhancement

Range predicate estimate can be much lower than an equivalent point get estimate if the high/low range of a histogram covers a broader range of values. This is because range estimation uses an interpolation formula whereas point get estimates will default to 1/NDV. These should be relatively consistent.

Example below shows the same query written as between vs an IN predicate. The example used comes from mysql_test/t/select_all.test. And any predicate values can be used where the histograms show a large range.

tidb> explain select fld1 from t2 where fld1 between 228313 and 228314;
+------------------------+---------+-----------+----------------------------+-----------------------------------------+
| id                     | estRows | task      | access object              | operator info                           |
+------------------------+---------+-----------+----------------------------+-----------------------------------------+
| IndexReader_6          | 0.05    | root      |                            | index:IndexRangeScan_5                  |
| └─IndexRangeScan_5    | 0.05    | cop[tikv] | table:t2, index:fld1(fld1) | range:[228313,228314], keep order:false |
+------------------------+---------+-----------+----------------------------+-----------------------------------------+

tidb> explain select fld1 from t2 where fld1 in (228313, 228314);
+-------------------+---------+------+----------------------------+------------------------------+
| id                | estRows | task | access object              | operator info                |
+-------------------+---------+------+----------------------------+------------------------------+
| Batch_Point_Get_1 | 2.00    | root | table:t2, index:fld1(fld1) | keep order:false, desc:false |
+-------------------+---------+------+----------------------------+------------------------------+