planner: fix bugs and make it more effective in outer join elimination rule.

[lzmhhh123]

What problem does this PR solve?

Now, the rule outer join elimination keeps a bug as the following cases:

tidb(localhost:4000) > drop table if exists t1, t2;
Query OK, 0 rows affected (0.00 sec)

tidb(localhost:4000) > create table t1(a int, b int, primary key(a));
Query OK, 0 rows affected (0.01 sec)

tidb(localhost:4000) > create table t2(a int, b int, c int, index ac(a, c));
Query OK, 0 rows affected (0.00 sec)

tidb(localhost:4000) > desc select t2.a sa, t2.c sc from t2 left join t1 on t2.a = t1.a;
+--------------------------+----------+------+------------------------------------------------------------------------+
| id                       | count    | task | operator info                                                          |
+--------------------------+----------+------+------------------------------------------------------------------------+
| Projection_5             | 12500.00 | root | test.t2.a, test.t2.c                                                   |
| └─MergeJoin_6            | 12500.00 | root | left outer join, left key:test.t2.a, right key:test.t1.a               |
|   ├─IndexReader_12       | 10000.00 | root | index:IndexScan_11                                                     |
|   │ └─IndexScan_11       | 10000.00 | cop  | table:t2, index:a, c, range:[NULL,+inf], keep order:true, stats:pseudo |
|   └─TableReader_14       | 10000.00 | root | data:TableScan_13                                                      |
|     └─TableScan_13       | 10000.00 | cop  | table:t1, range:[-inf,+inf], keep order:true, stats:pseudo             |
+--------------------------+----------+------+------------------------------------------------------------------------+
6 rows in set (0.00 sec)

It can't regard the alias column as the same as origin column. After this PR, this case of outer join can be normally eliminated.

tidb(localhost:4000) > desc select t2.a sa, t2.c sc from t2 left join t1 on t2.a = t1.a;
+-------------------+----------+------+-------------------------------------------------------------+                                           │
| id                | count    | task | operator info                                               |                                           │
+-------------------+----------+------+-------------------------------------------------------------+                                           │ 
| TableReader_7     | 10000.00 | root | data:TableScan_6                                            |                                           │
| └─TableScan_6     | 10000.00 | cop  | table:t2, range:[-inf,+inf], keep order:false, stats:pseudo |                                           │
+-------------------+----------+------+-------------------------------------------------------------+

Other enhancements and bugs fix are referred to issue: #9536, #11167

What is changed and how it works?

Use col.OrigColName instead of col.ColName. Record columns in schema between LogicalAggregation and LogicalJoin. Substitue columns for LogicalProjection and LogicalSelection.

Check List

Tests

• Unit test
• Integration test

Related changes

• Need to cherry-pick to the release branch

[winoros]

You can change

	// check the duplicate agnostic aggregate functions
	if ok, newCols := o.isDuplicateAgnosticAgg(p); ok {
		aggCols = newCols
	}

	newChildren := make([]LogicalPlan, 0, len(p.Children()))

to

	// check the duplicate agnostic aggregate functions
	if ok, newCols := o.isDuplicateAgnosticAgg(p); ok {
		aggCols = newCols
	}
	if proj, ok := p.(*LogicalProjection); ok {
		call expression.ColumnSubstitute to remap the aggCols.
	}
	newChildren := make([]LogicalPlan, 0, len(p.Children()))

Like what projection did when meet column prune and predicate push down.

Also, don't use projection's schema directly. Use its expression to substitute the columns in schema instead.

[codecov]

Codecov Report

Merging #11160 into master will decrease coverage by 0.2083%.
The diff coverage is 86.6666%.

@@               Coverage Diff               @@
##             master    #11160        +/-   ##
===============================================
- Coverage   81.4843%   81.276%   -0.2084%     
===============================================
  Files           423       423                
  Lines         91285     90125      -1160     
===============================================
- Hits          74383     73250      -1133     
+ Misses        11603     11576        -27     
  Partials       5299      5299

[winoros]

This diff patch can also pass the current tests in planner.

diff.txt

[lzmhhh123]

PTAL. @winoros @lamxTyler

[lzmhhh123]

/run-all-tests

[winoros]

I think it's not expected change?

[zz-jason]

@lzmhhh123 you can use make tidy to clean up the go.sum

[lzmhhh123]

That's the result with make tidy. ╮(╯_╰)╭

[lzmhhh123]

PTAL. @zz-jason

[zz-jason]

LGTM

[alivxxx]

LGTM

[lzmhhh123]

/run-all-tests

[sre-bot]

cherry pick to release-2.1 failed

[sre-bot]

cherry pick to release-3.0 failed