【GreatSQL优化器-09】make_join_query_block

一、make_join_query_block介绍

GreatSQL优化器对于多张表join的连接顺序在前面的章节介绍过的best_access_path函数已经执行了，接着就是把where条件进行切割然后推给合适的表。这个过程就是由函数make_join_query_block来执行的。

下面用几个简单的例子来说明join连接中条件推送是什么。

CREATE TABLE t1 (c1 INT PRIMARY KEY, c2 INT,date1 DATETIME);
INSERT INTO t1 VALUES (1,10,'2021-03-25 16:44:00.123456'),(2,1,'2022-03-26 16:44:00.123456'),(3,4,'2023-03-27 16:44:00.123456'),(5,5,'2024-03-25 16:44:00.123456'),(7,null,'2020-03-25 16:44:00.123456'),(8,10,'2020-10-25 16:44:00.123456'),(11,16,'2023-03-25 16:44:00.123456');
CREATE TABLE t2 (cc1 INT PRIMARY KEY, cc2 INT);
INSERT INTO t2 VALUES (1,3),(2,1),(3,2),(4,3),(5,15);
CREATE TABLE t3 (ccc1 INT, ccc2 varchar(100));
INSERT INTO t3 VALUES (1,'aa1'),(2,'bb1'),(3,'cc1'),(4,'dd1'),(null,'ee');
CREATE INDEX idx1 ON t1(c2);
CREATE INDEX idx2 ON t1(c2,date1);
CREATE INDEX idx2_1 ON t2(cc2);
CREATE INDEX idx3_1 ON t3(ccc1);

下面这个例子((t1.c1 = t3.ccc1) or (t3.ccc1 < 3))条件推送给t1

greatsql> EXPLAIN FORMAT=TREE SELECT * FROM t1 join t3 ON t1.c1=t3.ccc1 or t3.ccc1<3;
+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| EXPLAIN                                                                                                                                                                                                                                                         |
+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| -> Filter: ((t1.c1 = t3.ccc1) or (t3.ccc1 < 3))  (cost=5.26 rows=35)
    -> Inner hash join (no condition)  (cost=5.26 rows=35)
        -> Index scan on t1 using idx2  (cost=0.34 rows=7)
        -> Hash
            -> Table scan on t3  (cost=0.75 rows=5)
 |
+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1 row in set (0.00 sec)

下面例子(t1.c1 < 3)条件推给t1，(ccc1=t1.c1)条件推给t3

greatsql> EXPLAIN FORMAT=TREE SELECT * FROM t1 join t3 ON t1.c1=t3.ccc1 and t3.ccc1<3;
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| EXPLAIN                                                                                                                                                                                                                          |
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| -> Nested loop inner join  (cost=2.40 rows=2)
    -> Filter: (t1.c1 < 3)  (cost=1.70 rows=2)
        -> Index scan on t1 using idx2  (cost=1.70 rows=7)
    -> Index lookup on t3 using idx3_1 (ccc1=t1.c1)  (cost=0.30 rows=1)
 |
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+

下面例子((t3.ccc1 = t1.c2) or (t3.ccc1 is null) or (t3.ccc2 like 'a%'))条件推给t3，(((t3.ccc1 = t1.c2) and (t2.cc1 = t1.c1)) or (t3.ccc1 is null) or (t3.ccc2 like 'a%'))条件推给t2

greatsql> EXPLAIN SELECT * FROM t2,t1,t3 WHERE t1.c1=t2.cc1 AND t1.c2=t3.ccc1 OR t3.ccc1 IS NULL OR t3.ccc2 LIKE 'a%';
| -> Filter: (((t3.ccc1 = t1.c2) and (t2.cc1 = t1.c1)) or (t3.ccc1 is null) or (t3.ccc2 like 'a%'))  (cost=14.27 rows=85)
    -> Inner hash join (no condition)  (cost=14.27 rows=85)
        -> Index scan on t2 using idx2_1  (cost=0.09 rows=5)
        -> Hash
            -> Filter: ((t3.ccc1 = t1.c2) or (t3.ccc1 is null) or (t3.ccc2 like 'a%'))  (cost=4.70 rows=17)
                -> Inner hash join (no condition)  (cost=4.70 rows=17)
                    -> Table scan on t3  (cost=0.07 rows=5)
                    -> Hash
                        -> Index scan on t1 using idx2  (cost=0.95 rows=7)

二、make_join_query_block代码解释

make_join_query_block函数通过join表顺序和每张表的table_map属性以及cond条件的属性来决定cond条件添加到哪张表,并且可能会重新对表的索引进行check找出cost更低的索引，下面是代码解析。

bool JOIN::optimize() {
  make_join_query_block();
}

static bool make_join_query_block(JOIN *join, Item *cond) {
  for (uint i = join->const_tables; i < join->tables; i++) {
      // 这四个变量说明见表一
      JOIN_TAB *const tab = join->best_ref[i];
      const plan_idx first_inner = tab->first_inner();
      const table_map used_tables = tab->prefix_tables();
      const table_map current_map = tab->added_tables();
      if (cond)
        // 这里通过table_map属性决定了是否给这个表添加条件，见下面表二、表四和表五说明
        tmp = make_cond_for_table(thd, cond, used_tables, current_map, false);
     // 如果recheck_reason=true，这里需要重新做一次确认，找出cost最低的索引。见表六
      if (recheck_reason)
        test_if_order_by_key();
        test_if_cheaper_ordering();
        test_quick_select();
      }  
      /* Add conditions added by add_not_null_conds(). */
      if (and_conditions(&tmp, tab->condition())) return true;
      if (join->attach_join_conditions(i)) return true;
  }
}
// 条件添加基本原则是条件带有表列的添加到该表，但是如果属性不一致的话也不会添加，只会添加到最后一张表。具体解释见下面实际例子。

表一：上面四个变量解释

表二：make_cond_for_table()动作

表三：is_expensive_processor()函数

表四：Item的table_map属性

表五：表连接添加的属性

表六：表的索引是否要重新check

三、实际例子说明

接下来看几个例子来说明上面的代码。

首先看一下最后确定的连接顺序，为t1,t3,t2，因为条件不带有RAND_TABLE_BIT的Item，因此最后是按照cond含有的列推送给对应表来实现的。

例子一：

greatsql> EXPLAIN SELECT * FROM t2,t1,t3 WHERE t1.c1=t2.cc1 AND t1.c2=t3.ccc1 OR t3.ccc1 IS NULL OR t3.ccc2 LIKE 'a%';
+----+-------------+-------+------------+-------+-------------------+--------+---------+------+------+----------+---------------------------------------------------------+
| id | select_type | table | partitions | type  | possible_keys     | key    | key_len | ref  | rows | filtered | Extra                                                   |
+----+-------------+-------+------------+-------+-------------------+--------+---------+------+------+----------+---------------------------------------------------------+
|  1 | SIMPLE      | t1    | NULL       | index | PRIMARY,idx1,idx2 | idx2   | 11      | NULL |    7 |   100.00 | Using index                                             |
|  1 | SIMPLE      | t3    | NULL       | ALL   | idx3_1            | NULL   | NULL    | NULL |    5 |    48.80 | Using where; Using join buffer (hash join)              |
|  1 | SIMPLE      | t2    | NULL       | index | PRIMARY           | idx2_1 | 5       | NULL |    5 |   100.00 | Using where; Using index; Using join buffer (hash join) |
+----+-------------+-------+------------+-------+-------------------+--------+---------+------+------+----------+---------------------------------------------------------+

表一：是否把cond条件推送给表

注：这里的中括号代表当前检测表的左连接表，中括号右边就是当前正在检测的表

表二：表的table_map值

注：这里的INNER_TABLE_BIT和OUTER_REF_TABLE_BIT在函数JOIN::set_prefix_tables()默认加上了

看一下结果是否符合预期，确实如上表所述。这里看到又执行了一次test_quick_select()来确定走哪个索引。

"attaching_conditions_to_tables": {
              "original_condition": "(((`t3`.`ccc1` = `t1`.`c2`) and (`t2`.`cc1` = `t1`.`c1`)) or (`t3`.`ccc1` is null) or (`t3`.`ccc2` like 'a%'))",
              "attached_conditions_computation": [
                {
                  "table": "`t2`",
                  "rechecking_index_usage": { 这里对索引重新做了一次check
                    "recheck_reason": "not_first_table",
                    "range_analysis": {
                      "table_scan": {
                        "rows": 5,
                        "cost": 3.6
                      },
                      "potential_range_indexes": [
                        {
                          "index": "PRIMARY",
                          "usable": true,
                          "key_parts": [
                            "cc1"
                          ]
                        },
                        {
                          "index": "idx2_1",
                          "usable": false,
                          "cause": "not_applicable"
                        }
                      ],
                      "best_covering_index_scan": {
                        "index": "idx2_1",
                        "cost": 0.751098,
                        "chosen": true
                      },
                      "setup_range_conditions": [
                      ],
                      "group_index_range": {
                        "chosen": false,
                        "cause": "not_single_table"
                      },
                      "skip_scan_range": {
                        "chosen": false,
                        "cause": "not_single_table"
                      }
                    }
                  }
                }
              ],
              "attached_conditions_summary": [
                {
                  "table": "`t1`",
                  "attached": null
                },
                {
                  "table": "`t3`",
                  "attached": "((`t3`.`ccc1` = `t1`.`c2`) or (`t3`.`ccc1` is null) or (`t3`.`ccc2` like 'a%'))"
                },
                {
                  "table": "`t2`",
                  "attached": "(((`t3`.`ccc1` = `t1`.`c2`) and (`t2`.`cc1` = `t1`.`c1`)) or (`t3`.`ccc1` is null) or (`t3`.`ccc2` like 'a%'))"
                }
              ]
            }
          },
          {
            "finalizing_table_conditions": [
              {
                "table": "`t3`",
                "original_table_condition": "((`t3`.`ccc1` = `t1`.`c2`) or (`t3`.`ccc1` is null) or (`t3`.`ccc2` like 'a%'))",
                "final_table_condition   ": "((`t3`.`ccc1` = `t1`.`c2`) or (`t3`.`ccc1` is null) or (`t3`.`ccc2` like 'a%'))"
              },
              {
                "table": "`t2`",
                "original_table_condition": "(((`t3`.`ccc1` = `t1`.`c2`) and (`t2`.`cc1` = `t1`.`c1`)) or (`t3`.`ccc1` is null) or (`t3`.`ccc2` like 'a%'))",
                "final_table_condition   ": "(((`t3`.`ccc1` = `t1`.`c2`) and (`t2`.`cc1` = `t1`.`c1`)) or (`t3`.`ccc1` is null) or (`t3`.`ccc2` like 'a%'))"
              }
            ]
          },
          {
            "refine_plan": [
              {
                "table": "`t1`"
              },
              {
                "table": "`t3`"
              },
              {
                "table": "`t2`"
              }
            ]
          }
        ]
      }
    }

如果条件带有RAND_TABLE_BIT的Item，那么即使cond带有表的列，也不会推送给对应的表，而是推送到最后一张表。看下面的t1.c1 < rand()这个条件。

例子二：

greatsql> SELECT * FROM t2,t1,t3 WHERE t1.c1=t2.cc1 AND t1.c2=t3.ccc1 OR t3.ccc1 IS NULL AND t1.c1 < rand();
              "attached_conditions_summary": [
                {
                  "table": "`t1`",
                  "attached": null
                },
                {
                  "table": "`t3`",
                  "attached": "((`t3`.`ccc1` = `t1`.`c2`) or (`t3`.`ccc1` is null))"
                },
                {
                  "table": "`t2`",
                  "attached": "(((`t3`.`ccc1` = `t1`.`c2`) and (`t2`.`cc1` = `t1`.`c1`)) or ((`t3`.`ccc1` is null) and (`t1`.`c1` < rand())))"  看到条件t1.c1 < rand()没有推送给t1而是推送到最后一张表t2去了
                }
              ]
            }
          },
          {
            "finalizing_table_conditions": [
              {
                "table": "`t3`",
                "original_table_condition": "((`t3`.`ccc1` = `t1`.`c2`) or (`t3`.`ccc1` is null))",
                "final_table_condition   ": "((`t3`.`ccc1` = `t1`.`c2`) or (`t3`.`ccc1` is null))"
              },
              {
                "table": "`t2`",
                "original_table_condition": "(((`t3`.`ccc1` = `t1`.`c2`) and (`t2`.`cc1` = `t1`.`c1`)) or ((`t3`.`ccc1` is null) and (`t1`.`c1` < rand())))",
                "final_table_condition   ": "(((`t3`.`ccc1` = `t1`.`c2`) and (`t2`.`cc1` = `t1`.`c1`)) or ((`t3`.`ccc1` is null) and (`t1`.`c1` < rand())))"
              }
            ]
          },
          {
            "refine_plan": [
              {
                "table": "`t1`"
              },
              {
                "table": "`t3`"
              },
              {
                "table": "`t2`"
              }

看一下每张表的属性：

四、总结

从上面优化器最早的步骤我们认识了make_join_query_block函数的作用，知道了通过join表顺序和每张表的table_map属性以及cond条件的属性来决定cond条件添加到哪张表,并且可能会重新对表的索引进行check找出cost更低的索引，需要注意的是有的带有表列的条件不会被添加到对应表，因为Item的属性跟表的属性不一致所以最后只会被添加到最后一张join表。