GaussDB Stream方式的Hint
功能描述
指明stream使用的方法,可以为broadcast和redistribute,或者直接指定生成gather计划。
语法格式
[no] broadcast|redistribute( [@queryblock] table_list) gather( [@queryblock] REL|JOIN|ALL)
参数说明
@queryblock 见指定Hint所处的查询块Queryblock章节,可省略,表示在当前查询块生效。 broadcast和redistribute no表示hint的stream方式不使用。 table_list为进行stream操作的单表或多表join结果集,见参数说明。 gather gather hint可以指定三种计划生成方式: REL:只生成基于基表的gather路径,然后再在CN上执行剩余计划。 JOIN:尽可能生成基于join的gather路径,在能下推的join子计划上面(join下面不包含重分布节点)添加gather路径,剩余计划在CN上执行。对于需要重分布节点的join计划则生成不出这种基于join的gather路径,会回退生成基于基表的gather路径。 注意: 在指定Hint(JOIN)后,对于分布表和复制表做连接的情况会导致生成不出来Hint(JOIN)期望的计划,因为优化器已经寻找更优的计划进行替代。
ALL:基于最优方式选择Gather Rel或Gather Join路径。
示例
对示例中原语句使用如下hint:
explain
select /*+ no redistribute(store_sales store_returns item store) leading(((store_sales store_returns item store) customer)) */ i_product_name product_name ...
原计划中,(store_sales store_returns item store)和customer做join时,前者做了重分布,此hint表示禁止前者混合表做重分布,但仍然保持join顺序,则生成计划如下所示:
点击放大
对语句进行Gather Hint指定:
生成基表Gather计划 /* +GATHER(REL)*/。
gaussdb=# explain select /*+ GATHER(REL)*/* from t1, t2, t3 where t1.c2 = t2.c2 and t2.c2 = t3.c2;
id | operation | E-rows | E-width | E-costs
----+---------------------------------------+--------+---------+---------
1 | -> Hash Join (2,8) | 20 | 36 | 44.10
2 | -> Hash Join (3,5) | 20 | 24 | 29.22
3 | -> Streaming (type: GATHER) | 20 | 12 | 14.35
4 | -> Seq Scan on t1 | 20 | 12 | 13.13
5 | -> Hash | 20 | 12 | 14.35
6 | -> Streaming (type: GATHER) | 20 | 12 | 14.35
7 | -> Seq Scan on t2 | 20 | 12 | 13.13
8 | -> Hash | 20 | 12 | 14.35
9 | -> Streaming (type: GATHER) | 20 | 12 | 14.35
10 | -> Seq Scan on t3 | 20 | 12 | 13.13
(10 rows)
Predicate Information (identified by plan id)
-----------------------------------------------
1 --Hash Join (2,8)
Hash Cond: (t1.c2 = t3.c2)
2 --Hash Join (3,5)
Hash Cond: (t1.c2 = t2.c2)
(4 rows)
2、生成可下推计划的Join Gather计划 /+ GATHER(REL)/。
gaussdb=# explain select /*+ GATHER(JOIN)*/* from t1, t2, t3 where t1.c1 = t2.c1 and t2.c2 = t3.c2;
id | operation | E-rows | E-width | E-costs
----+------------------------------------+--------+---------+---------
1 | -> Hash Join (2,7) | 20 | 36 | 42.37
2 | -> Streaming (type: GATHER) | 20 | 24 | 27.49
3 | -> Hash Join (4,5) | 20 | 24 | 26.56
4 | -> Seq Scan on t1 | 20 | 12 | 13.13
5 | -> Hash | 21 | 12 | 13.13
6 | -> Seq Scan on t2 | 20 | 12 | 13.13
7 | -> Hash | 20 | 12 | 14.35
8 | -> Streaming (type: GATHER) | 20 | 12 | 14.35
9 | -> Seq Scan on t3 | 20 | 12 | 13.13
(9 rows)
Predicate Information (identified by plan id)
-----------------------------------------------
1 --Hash Join (2,7)
Hash Cond: (t2.c2 = t3.c2)
3 --Hash Join (4,5)
Hash Cond: (t1.c1 = t2.c1)
(4 rows)
3、生成最优方式的Gather计划 /+ GATHER(ALL)/。 会基于最优方式及规则选择GATHER(REL)或者GATHER(JOIN)路径。
gaussdb=# explain select /*+ GATHER(ALL)*/* from t1, t2, t3 where t1.c1 = t2.c1 and t2.c2 = t3.c2;
id | operation | E-rows | E-width | E-costs
----+------------------------------------+--------+---------+---------
1 | -> Hash Join (2,7) | 20 | 36 | 42.37
2 | -> Streaming (type: GATHER) | 20 | 24 | 27.49
3 | -> Hash Join (4,5) | 20 | 24 | 26.56
4 | -> Seq Scan on t1 | 20 | 12 | 13.13
5 | -> Hash | 21 | 12 | 13.13
6 | -> Seq Scan on t2 | 20 | 12 | 13.13
7 | -> Hash | 20 | 12 | 14.35
8 | -> Streaming (type: GATHER) | 20 | 12 | 14.35
9 | -> Seq Scan on t3 | 20 | 12 | 13.13
(9 rows)
Predicate Information (identified by plan id)
-----------------------------------------------
1 --Hash Join (2,7)
Hash Cond: (t2.c2 = t3.c2)
3 --Hash Join (4,5)
Hash Cond: (t1.c1 = t2.c1)
(4 rows)