plan, expression: support is null in hash partition pruning (#17281)

[sre-bot]

cherry-pick #17281 to release-4.0

---

What problem does this PR solve?

Problem Summary:

SQL like select * from t where a is null will produce a full table scan plan in hash partition. For example:

create table t (a int, b int) partition by hash(a + b) partitions 10;
explain select * from t where a is null and b = 2;

This SQL will produce following result:

MySQL [test]> explain select * from t where a is null and b = 2;
+------------------------------+----------+-----------+-----------------------+-----------------------------------+
| id                           | estRows  | task      | access object         | operator info                     |
+------------------------------+----------+-----------+-----------------------+-----------------------------------+
| Union_16                     | 0.10     | root      |                       |                                   |
| ├─TableReader_19             | 0.01     | root      |                       | data:Selection_18                 |
| │ └─Selection_18             | 0.01     | cop[tikv] |                       | eq(test.t.b, 2), isnull(test.t.a) |
| │   └─TableFullScan_17       | 10000.00 | cop[tikv] | table:t, partition:p0 | keep order:false, stats:pseudo    |
| ├─TableReader_22             | 0.01     | root      |                       | data:Selection_21                 |
| │ └─Selection_21             | 0.01     | cop[tikv] |                       | eq(test.t.b, 2), isnull(test.t.a) |
| │   └─TableFullScan_20       | 10000.00 | cop[tikv] | table:t, partition:p1 | keep order:false, stats:pseudo    |
| ├─TableReader_25             | 0.01     | root      |                       | data:Selection_24                 |
| │ └─Selection_24             | 0.01     | cop[tikv] |                       | eq(test.t.b, 2), isnull(test.t.a) |
| │   └─TableFullScan_23       | 10000.00 | cop[tikv] | table:t, partition:p2 | keep order:false, stats:pseudo    |
| ├─TableReader_28             | 0.01     | root      |                       | data:Selection_27                 |
| │ └─Selection_27             | 0.01     | cop[tikv] |                       | eq(test.t.b, 2), isnull(test.t.a) |
| │   └─TableFullScan_26       | 10000.00 | cop[tikv] | table:t, partition:p3 | keep order:false, stats:pseudo    |
| ├─TableReader_31             | 0.01     | root      |                       | data:Selection_30                 |
| │ └─Selection_30             | 0.01     | cop[tikv] |                       | eq(test.t.b, 2), isnull(test.t.a) |
| │   └─TableFullScan_29       | 10000.00 | cop[tikv] | table:t, partition:p4 | keep order:false, stats:pseudo    |
| ├─TableReader_34             | 0.01     | root      |                       | data:Selection_33                 |
| │ └─Selection_33             | 0.01     | cop[tikv] |                       | eq(test.t.b, 2), isnull(test.t.a) |
| │   └─TableFullScan_32       | 10000.00 | cop[tikv] | table:t, partition:p5 | keep order:false, stats:pseudo    |
| ├─TableReader_37             | 0.01     | root      |                       | data:Selection_36                 |
| │ └─Selection_36             | 0.01     | cop[tikv] |                       | eq(test.t.b, 2), isnull(test.t.a) |
| │   └─TableFullScan_35       | 10000.00 | cop[tikv] | table:t, partition:p6 | keep order:false, stats:pseudo    |
| ├─TableReader_40             | 0.01     | root      |                       | data:Selection_39                 |
| │ └─Selection_39             | 0.01     | cop[tikv] |                       | eq(test.t.b, 2), isnull(test.t.a) |
| │   └─TableFullScan_38       | 10000.00 | cop[tikv] | table:t, partition:p7 | keep order:false, stats:pseudo    |
| ├─TableReader_43             | 0.01     | root      |                       | data:Selection_42                 |
| │ └─Selection_42             | 0.01     | cop[tikv] |                       | eq(test.t.b, 2), isnull(test.t.a) |
| │   └─TableFullScan_41       | 10000.00 | cop[tikv] | table:t, partition:p8 | keep order:false, stats:pseudo    |
| └─TableReader_46             | 0.01     | root      |                       | data:Selection_45                 |
|   └─Selection_45             | 0.01     | cop[tikv] |                       | eq(test.t.b, 2), isnull(test.t.a) |
|     └─TableFullScan_44       | 10000.00 | cop[tikv] | table:t, partition:p9 | keep order:false, stats:pseudo    |
+------------------------------+----------+-----------+-----------------------+-----------------------------------+

What is changed and how it works?

Proposal: xxx

What's Changed:

During partition pruning, if a column is equal to NULL, all calculation relating to it will produce null result and set a NULL flag.

How it Works:

If pruning result with a NULL flag, it will be point to p0.

Related changes

• Need to cherry-pick to the release branch

Check List

Tests

• Unit test

Side effects

• No

Release note

• planner: support is null filter condition in hash partition pruning

[sre-bot]

/run-all-tests

[XuHuaiyu]

LGTM

[jackysp]

LGTM

[jebter]

/merge

[sre-bot]

/run-all-tests