Postgresql源码（105）分区表剪枝代码分析

相关：
《Postgresql源码（105）分区表剪枝代码分析》
《Postgresql源码（106）Generic Plan与Custom Plan的区别（以分区表为例）》

对于分区表，子表剪枝是保证性能最重要的手段，优化器在生成Plan的过程中对子表进行了剪枝，本篇对剪枝流程做简要总结。

1 构造数据

CREATE TABLE measurement (
    city_id         int not null,
    logdate         date not null,
    peaktemp        int,
    unitsales       int
) PARTITION BY RANGE (logdate);

CREATE TABLE measurement_y2006m01 PARTITION OF measurement FOR VALUES FROM ('2006-01-01') TO ('2006-02-01');
CREATE TABLE measurement_y2006m02 PARTITION OF measurement FOR VALUES FROM ('2006-02-01') TO ('2006-03-01');
CREATE TABLE measurement_y2006m03 PARTITION OF measurement FOR VALUES FROM ('2006-03-01') TO ('2006-04-01');
CREATE TABLE measurement_y2006m04 PARTITION OF measurement FOR VALUES FROM ('2006-04-01') TO ('2006-05-01');
CREATE TABLE measurement_y2006m05 PARTITION OF measurement FOR VALUES FROM ('2006-05-01') TO ('2006-06-01');

insert into measurement values (1, '2006-01-03', floor(random() * 100), floor(random() * 100));
insert into measurement values (2, '2006-02-04', floor(random() * 100), floor(random() * 100));
insert into measurement values (3, '2006-03-05', floor(random() * 100), floor(random() * 100));
insert into measurement values (4, '2006-03-06', floor(random() * 100), floor(random() * 100));
insert into measurement values (5, '2006-03-07', floor(random() * 100), floor(random() * 100));
insert into measurement values (6, '2006-03-08', floor(random() * 100), floor(random() * 100));
insert into measurement values (7, '2006-04-09', floor(random() * 100), floor(random() * 100));
insert into measurement values (8, '2006-04-10', floor(random() * 100), floor(random() * 100));
insert into measurement values (9, '2006-05-11', floor(random() * 100), floor(random() * 100));
insert into measurement values (10, '2006-05-12', floor(random() * 100), floor(random() * 100));

2 验证SQL

SQL1
这是一条经过裁剪的SQL，只扫描4月分区：

explain select * from measurement where logdate between '2006-04-05' and '2006-04-20';
                                    QUERY PLAN                                    
----------------------------------------------------------------------------------
 Seq Scan on measurement_y2006m04 measurement  (cost=0.00..37.75 rows=9 width=16)
   Filter: ((logdate >= '2006-04-05'::date) AND (logdate <= '2006-04-20'::date))

SQL2
这是一条无法裁剪的SQL，扫描所有月份的分区。

explain select * from measurement where peaktemp between 1 and 10;
                                        QUERY PLAN                                        
------------------------------------------------------------------------------------------
 Append  (cost=0.00..188.97 rows=45 width=16)
   ->  Seq Scan on measurement_y2006m01 measurement_1  (cost=0.00..37.75 rows=9 width=16)
         Filter: ((peaktemp >= 1) AND (peaktemp <= 10))
   ->  Seq Scan on measurement_y2006m02 measurement_2  (cost=0.00..37.75 rows=9 width=16)
         Filter: ((peaktemp >= 1) AND (peaktemp <= 10))
   ->  Seq Scan on measurement_y2006m03 measurement_3  (cost=0.00..37.75 rows=9 width=16)
         Filter: ((peaktemp >= 1) AND (peaktemp <= 10))
   ->  Seq Scan on measurement_y2006m04 measurement_4  (cost=0.00..37.75 rows=9 width=16)
         Filter: ((peaktemp >= 1) AND (peaktemp <= 10))
   ->  Seq Scan on measurement_y2006m05 measurement_5  (cost=0.00..37.75 rows=9 width=16)
         Filter: ((peaktemp >= 1) AND (peaktemp <= 10))

3 SQL1分析（可剪枝SQL）

select * from measurement where logdate between '2006-04-05' and '2006-04-20';

3.1 语义分析阶段结果：pg_analyze_and_rewrite_fixedparams

p elog_node_display(LOG, "querytree_list", querytree_list, true)

在语义分析阶段未剪枝。

3.2 计划生成阶段结果：pg_plan_queries

p elog_node_display(LOG, "plantree_list", plantree_list, true)

pg_plan_queries
	pg_plan_query
		planner
			standard_planner
				subquery_planner              // 【1】第一步
				fetch_upper_rel               // 【2】第二步
				get_cheapest_fractional_path  // 【3】第三步
				create_plan                   // 【4】第四步

第一步到第四步的输出结果（除了最右面的图用来对比，其他都是当前可剪枝SQL的）：

3.2.1 剪枝发生在subquery_planner

【1】subquery_planner函数中：PlannerInfo初始状态

【2】subquery_planner函数中：进入grouping_planner前都没有剪枝

subquery_planner
	foreach(l, parse->rtable)
		switch (rte->rtekind)
			case RTE_RELATION:
				if (rte->inh)
					rte->inh = has_subclass(rte->relid);   // 确认当前父表存在子表！去pg_class查relhassubclass字段
	...
	...
	grouping_planner  // 进入前

【3】prune_append_rel_partitions函数开始剪枝！

query_planner
	...
	...
	add_other_rels_to_query
		/* If it's marked as inheritable, look for children. */
		if (rte->inh)
			expand_inherited_rtentry	
				expand_partitioned_rtentry
					PartitionDirectoryLookup
						relinfo->live_parts = live_parts = prune_append_rel_partitions(relinfo);
	

剪枝函数：prune_append_rel_partitions
剪枝后只有一个分区：relinfo->live_parts = {nwords = 1, words = 0x173e4c0}
剪枝依据：relinfo：

1. prune_append_rel_partitions构造裁剪步骤steps调用get_matching_partitions
2. get_matching_partitions收到steps：三步（between产生两个比较，和一个固定的combine）

prune_append_rel_partitions
	get_matching_partitions(&context, pruning_steps);    // 收到pruning_steps

pruning_steps:
	{step = {type = T_PartitionPruneStepOp, step_id = 0}, opstrategy = 2, exprs = 0x172cfe0, cmpfns = 0x172cf88, nullkeys = 0x0}
		exprs = Const
		cmpfns = 1092 date_cmp
	{step = {type = T_PartitionPruneStepOp, step_id = 1}, opstrategy = 4, exprs = 0x172d1f0, cmpfns = 0x172d198, nullkeys = 0x0}
		exprs = Const
		cmpfns = 1092 date_cmp
	{step = {type = T_PartitionPruneStepCombine, step_id = 2}, combineOp = PARTPRUNE_COMBINE_INTERSECT, source_stepids = 0x172d350}
		source_stepids = {0, 1}
	

get_matching_partitions 处理上面steps，产生结果：results[]

PruneStepResult = {bound_offsets = 0x172d558, scan_default = false, scan_null = false}
	bound_offsets = Bitmapset{0001 1111}
PruneStepResult = {bound_offsets = 0x172d5a8, scan_default = false, scan_null = false}
	bound_offsets = Bitmapset{0111 0000}
PruneStepResult = {bound_offsets = 0x172d5f8, scan_default = false, scan_null = false}
	bound_offsets = Bitmapset{0001 0000}   // 只存了4

expand_partitioned_rtentry函数中，使用4找到4-1=3号子表：

expand_partitioned_rtentry
	while ((i = bms_next_member(live_parts, i)) >= 0)
		Oid                     childOID = partdesc->oids[i];    // i = 3 找到3号子表

流程总结：（左侧流程）
《Postgresql源码（106）Generic Plan与Custom Plan的区别（以分区表为例）》

剪枝流程完成，后续细节不再展开。

3.3 执行阶段

按照裁剪计划数执行，只扫描4月表measurement_y2006m04

explain select * from measurement where logdate between '2006-04-05' and '2006-04-20';
                                    QUERY PLAN                                    
----------------------------------------------------------------------------------
 Seq Scan on measurement_y2006m04 measurement  (cost=0.00..37.75 rows=9 width=16)
   Filter: ((logdate >= '2006-04-05'::date) AND (logdate <= '2006-04-20'::date))

4 SQL2分析（无法裁剪）

select * from measurement where peaktemp between 1 and 10;

4.1 计划生成阶段

p elog_node_display(LOG, "plantree_list", plantree_list, true)

4.2 执行阶段

按照计划数执行会自动把所有子表扫一遍。

explain select * from measurement where peaktemp between 1 and 10;
                                        QUERY PLAN                                        
------------------------------------------------------------------------------------------
 Append  (cost=0.00..188.97 rows=45 width=16)
   ->  Seq Scan on measurement_y2006m01 measurement_1  (cost=0.00..37.75 rows=9 width=16)
         Filter: ((peaktemp >= 1) AND (peaktemp <= 10))
   ->  Seq Scan on measurement_y2006m02 measurement_2  (cost=0.00..37.75 rows=9 width=16)
         Filter: ((peaktemp >= 1) AND (peaktemp <= 10))
   ->  Seq Scan on measurement_y2006m03 measurement_3  (cost=0.00..37.75 rows=9 width=16)
         Filter: ((peaktemp >= 1) AND (peaktemp <= 10))
   ->  Seq Scan on measurement_y2006m04 measurement_4  (cost=0.00..37.75 rows=9 width=16)
         Filter: ((peaktemp >= 1) AND (peaktemp <= 10))
   ->  Seq Scan on measurement_y2006m05 measurement_5  (cost=0.00..37.75 rows=9 width=16)
         Filter: ((peaktemp >= 1) AND (peaktemp <= 10))