Postgresql源码（116）提升子查询案例分析

0 总结

对于SQL：select * from student, (select * from score where sno > 2) s where student.sno = s.sno;

pullup在pull_up_subqueries函数内递归完成，分几步：

1. 将内层rte score追加到上层rtbable中：rte1是student、rte2带subquery是子查询、rte3是score。
2. 调整所有var的varno（从1指向3）、varlevelsup（本例不涉及）；还有其他调整本例不涉及。
3. 将上层代表子查询的rte2的subquery清空，但rte2不删除。
4. 将上层jointree中，指向子查询的rte替换为 子查询中的FromExpr（sno > 2）。

1 待分析场景

drop table student;
create table student(sno int, sname varchar(10), ssex int);
insert into student values(1, 'stu1', 0);
insert into student values(2, 'stu2', 1);
insert into student values(3, 'stu3', 1);
insert into student values(4, 'stu4', 0);drop table course;
create table course(cno int, cname varchar(10), tno int);
insert into course values(10, 'meth', 1);
insert into course values(11, 'english', 2);drop table teacher;
create table teacher(tno int, tname varchar(10), tsex int);
insert into teacher values(1, 'te1', 1);
insert into teacher values(2, 'te2', 0);drop table score;
create table score (sno int, cno int, degree int);
insert into score values (1, 10, 100);
insert into score values (1, 11, 89);
insert into score values (2, 10, 99);
insert into score values (2, 11, 90);
insert into score values (3, 10, 87);
insert into score values (3, 11, 20);
insert into score values (4, 10, 60);
insert into score values (4, 11, 70);

带子查询的语句：select * from student, (select * from score where sno > 2) s where student.sno = s.sno;

set enable_hashjoin to off;
set enable_mergejoin to off;explain select * from student, (select * from score where sno > 2) s where student.sno = s.sno;QUERY PLAN
---------------------------------------------------------------------Nested Loop  (cost=0.00..11278.20 rows=3740 width=58)Join Filter: (student.sno = score.sno)->  Seq Scan on student  (cost=0.00..21.00 rows=1100 width=46)->  Materialize  (cost=0.00..38.90 rows=680 width=12)->  Seq Scan on score  (cost=0.00..35.50 rows=680 width=12)Filter: (sno > 2)||||||||||
||等价写法||
vvvvvvvvvvvexplain select * from student, score where score.sno > 2 and student.sno = score.sno;QUERY PLAN
---------------------------------------------------------------------Nested Loop  (cost=0.00..11278.20 rows=3740 width=58)Join Filter: (student.sno = score.sno)->  Seq Scan on student  (cost=0.00..21.00 rows=1100 width=46)->  Materialize  (cost=0.00..38.90 rows=680 width=12)->  Seq Scan on score  (cost=0.00..35.50 rows=680 width=12)Filter: (sno > 2)

---

pull_up_subqueries做的事情就是帮我们把子查询上拉了，下面分析上拉是如何做的。

select * from student, (select * from score where sno > 2) s where student.sno = s.sno;

---

Plannerinfo完整结构

注意pull_up_subqueries只对jointree做处理。

void
pull_up_subqueries(PlannerInfo *root)
{/* Top level of jointree must always be a FromExpr */Assert(IsA(root->parse->jointree, FromExpr));/* Recursion starts with no containing join nor appendrel */root->parse->jointree = (FromExpr *)pull_up_subqueries_recurse(root, (Node *) root->parse->jointree,NULL, NULL);/* We should still have a FromExpr */Assert(IsA(root->parse->jointree, FromExpr));
}

2 pull_up_subqueries流程分析

2.1 处理FromExpr下面挂的第一个RANGETBLREF（student表）

FromExpr的第一张表是student表，指向一个rtekind = RTE_RELATION普通表类型，无需做任何处理。

2.2 处理FromExpr下面挂的第二个RANGETBLREF（子查询）

FromExpr的第二个rte是子查询(select * from score where sno > 2) s，可以看到引用的rte结构的subquery指向了内层query：

开始进入pull_up_simple_subquery内部处理，进入路径：

在分析pull_up_simple_subquery前有两个准入条件：

1. rte->rtekind == RTE_SUBQUERY
2. is_simple_subquery：不全部列举了，其中重要的是子查询不能带有一些特殊的语法：

is_simple_subquery...if (subquery->hasAggs ||subquery->hasWindowFuncs ||subquery->hasTargetSRFs ||subquery->groupClause ||subquery->groupingSets ||subquery->havingQual ||subquery->sortClause ||subquery->distinctClause ||subquery->limitOffset ||subquery->limitCount ||subquery->hasForUpdate ||subquery->cteList)return false;...

2.3 进入pull_up_simple_subquery开始处理子查上拉

第一步：拿到rte指向的子查询的Query树，构造PlannerInfo开始处理。

static Node *
pull_up_simple_subquery(PlannerInfo *root, Node *jtnode, RangeTblEntry *rte,JoinExpr *lowest_outer_join,AppendRelInfo *containing_appendrel)
{Query	   *subquery;Query	   *parse = root->parse;PlannerInfo *subroot;subquery = copyObject(rte->subquery);subroot = makeNode(PlannerInfo);subroot->parse = subquery;......

第二步：递归的处理sublink、subquery等。

	replace_empty_jointree(subquery);if (subquery->hasSubLinks)pull_up_sublinks(subroot);preprocess_function_rtes(subroot);pull_up_subqueries(subroot);

第三步：开始pull up

到这里subroot就是rte2的subquery子查询的结构还没有任何调整：

	/** Adjust level-0 varnos in subquery so that we can append its rangetable* to upper query's.  We have to fix the subquery's append_rel_list as* well.*/rtoffset = list_length(parse->rtable);  // 2OffsetVarNodes((Node *) subquery, rtoffset, 0);OffsetVarNodes((Node *) subroot->append_rel_list, rtoffset, 0);

parse是上层查询的，上层有两个rtable。因为要把子查询拉平，所以把子查询的varno的指向调整一下，因为是要append到父查询，所以直接加上父查询rte的个数就好了（这里是2）。加完了应该指向父查询rte的3的位置（现在父查询只有两个rte，3位置是空的）。

OffsetVarNodes((Node *) subquery, 2, 0);

1. 调整var→varno：1→3。
2. 调整rangetblref→rindex：1→3。
   

	/** Upper-level vars in subquery are now one level closer to their parent* than before.*/IncrementVarSublevelsUp((Node *) subquery, -1, 1);IncrementVarSublevelsUp((Node *) subroot->append_rel_list, -1, 1);

• 这一步调整的目的：因为varlevelsup=1表示引用上一层的列（相当于距离）这里拉平后，varlevelsup就需要-1了，因为距离少了1。
• 在当前SQL中select * from student, (select * from score where sno > 2) s where student.sno = s.sno;，开始调整var→varlevelsup字段，注意这个字段表示当前查询中使用了上层的变量，但上面子查询中(select * from score where sno > 2)没有引用上层的任何列，所以子查询中的var→varlevelsup都是0。这一步调整不会有影响。

	/** Now append the adjusted rtable entries and their perminfos to upper* query. (We hold off until after fixing the upper rtable entries; no* point in running that code on the subquery ones too.)*/CombineRangeTables(&parse->rtable, &parse->rteperminfos,subquery->rtable, subquery->rteperminfos);

开始把子查询的RTE拷贝到上层，现在子查询里面的varno=3指向就对了。

	/** We no longer need the RTE's copy of the subquery's query tree.  Getting* rid of it saves nothing in particular so far as this level of query is* concerned; but if this query level is in turn pulled up into a parent,* we'd waste cycles copying the now-unused query tree.*/rte->subquery = NULL;

删除子查询RTE带的Query，注意现在还缺一个条件。

pull_up_simple_subqueryreturn (Node *) subquery->jointree;

返回一个jointree带着条件。

返回去后，在这里把fromlist指向的第二个rte（子查询）换成 上面计算好的jointree。

然后就拉平了。

pull_up_subqueries_recurse...else if (IsA(jtnode, FromExpr)){FromExpr   *f = (FromExpr *) jtnode;ListCell   *l;Assert(containing_appendrel == NULL);/* Recursively transform all the child nodes */foreach(l, f->fromlist){lfirst(l) = pull_up_subqueries_recurse(root, lfirst(l),lowest_outer_join,NULL);}}

最终效果对比

pullup前 vs pullup后