PostgreSQL查询优化器之grouping_planner

grouping_planner主要做了3个工作：

1. 对集合进行处理
2. 对非SPJ函数进行优化
3. 对SQL查询语句进行物理优化

grouping_planner实现代码如下：

static void
grouping_planner(PlannerInfo *root, bool inheritance_update,
				 double tuple_fraction)
{
    /* 如果存在limit，offset，元组片段因子要改小 */
    if (parse->limitCount || parse->limitOffset)
    {
    	tuple_fraction = preprocess_limit(root, tuple_fraction,
    				  &offset_est, &count_est);
    }
    /* Make tuple_fraction accessible to lower-level routines */
    root->tuple_fraction = tuple_fraction;

        //判断是否存在集合操作，如何存在，则处理集合运算。
    if (parse->setOperations)
    {
        //会把集合语句按照集合操作符(差，并，交)分割SQL语句，
        //然后调用为每一个独立的部分调用subquery_planner，
        //所以Postgresql几乎不支持集合优化
        //current_rel = plan_set_operations(root);
        //顺便求出路径排序
    	root->sort_pathkeys = make_pathkeys_for_sortclauses(root,
    			parse->sortClause,
    			tlist);
    }
    else//非集合操作
    {
        /* ORDER BY和GROUP BY同时存在，先GROUP BY，在ORDER BY */
        if (parse->groupingSets)
        {
                groupclause = preprocess_groupclause(root,
                     linitial(current_sets));
        }

        /* 对目标列进行处理*/
        tlist = preprocess_targetlist(root, tlist);

        //提前执行带有max/min的聚合函数子句
        if (parse->hasAggs)
        	preprocess_minmax_aggregates(root, tlist);
        }
    	/*最优路径*/
    	current_rel = query_planner(root, tlist,
				standard_qp_callback, &qp_extra);

		//为max/min生成执行计划
        if (parse->hasAggs)
            preprocess_minmax_aggregates(root, tlist);
    }
}

query_planner生成最优查询路径

产生两个最优查询路径，主要是cheatest_path(未排序)和sorted_path(排序)

RelOptInfo *
query_planner(PlannerInfo *root, List *tlist,
			  query_pathkeys_callback qp_callback, void *qp_extra)
{
    /*
     * If the query has an empty join tree, then it's something easy like
     * "SELECT 2+2;" or "INSERT ... VALUES()".  Fall through quickly.
     */
    if (parse->jointree->fromlist == NIL)
    {
    	/*
    	 * We still are required to call qp_callback, in case it's something
    	 * like "SELECT 2+2 ORDER BY 1".标准化其他排序键，例如ORDER BY，GROUP BY
    	 */
    	root->canon_pathkeys = NIL;
    	(*qp_callback) (root, qp_extra);

    	return final_rel;
    }
    //初始化ROOT成员
    /*找出所有基本表，放入simple_rte_array */
    setup_simple_rel_arrays(root);
    /*找出所有基本表，放入生成基本关系*/
    add_base_rels_to_query(root, (Node *) parse->jointree);

    //分解where和join中的约束条件，构建连接树
    joinlist = deconstruct_jointree(root);

    /*检查外连接子句，把外连接的约束条件分发到对应关系上
    * ，看源码好像没有推到join关系上，而是推到join关系的子关系上
    */
    reconsider_outer_join_clauses(root);

    /*处理隐含约束条件*/
    generate_base_implied_equalities(root);

    /*去除无用连接*/
    joinlist = remove_useless_joins(root, joinlist);

    /*完成多表链接，采用动态规划和遗传算法 */
    final_rel = make_one_rel(root, joinlist);

    return final_rel;
}

deconstruct_jointree构造连接树函数

deconstruct_jointree用于分解树上的连接结构，分解方式为：把where和join中每个子句加入一个list中，然后把约束条件分配到每个关系上。一是把限制条件分配到基本关系上，二是把连接条件分配到连接关系上。这些本质上是逻辑优化阶段的“谓词下推操作”。但是由于此时还没有构造join关系，所以不能推到join关系上

static List *
deconstruct_recurse(PlannerInfo *root, Node *jtnode, bool below_outer_join,
				Relids *qualscope, Relids *inner_join_rels,
				List **postponed_qual_list)
{
    if (IsA(jtnode, RangeTblRef))
    {
        //构造只有一个节点的关系
    	joinlist = list_make1(jtnode);
    }
    else if (IsA(jtnode, FromExpr))
    {
        //递归构造每一个From子句，然后把结果下推
    	/*
    	 * Now process the top-level quals.
    	 */
    	foreach(l, (List *) f->quals)
    	{
    	    //还构建了RestrictInfo
    		distribute_qual_to_rels(root, qual,
				false, below_outer_join, JOIN_INNER,
				*qualscope, NULL, NULL, NULL,
				postponed_qual_list);
    	}
    }
    else if (IsA(jtnode, JoinExpr))
    {
        //递归构造join两边
    	switch (j->jointype)
    	{
    		case JOIN_INNER:
    		case JOIN_ANTI:
    		case JOIN_FULL:
    		default:
    	}

    	/*处理join下推*/
    	foreach(l, my_quals)
    	{
    		Node	   *qual = (Node *) lfirst(l);

    		distribute_qual_to_rels(root, qual,
				false, below_outer_join, j->jointype,
				*qualscope,
				ojscope, nonnullable_rels, NULL,
				postponed_qual_list);
    	}
    }
    return joinlist;
}

reconsider_outer_join_clauses

分发外连接子句的约束条件

generate_base_implied_equalites

找出隐含条件，进一步谓词下推

make_one_rel 构造多表连接路径并选择最优路径的函数

RelOptInfo *
make_one_rel(PlannerInfo *root, List *joinlist)
{
    /* Mark base rels as to whether we care about fast-start plans */
    set_base_rel_consider_startup(root);

    //为每个基本关系估计大小
    set_base_rel_sizes(root);
    //为每个基本关系生成RelOptInfo结构，并且生成访问路径放在path，这是单表/子查询的最佳扫描方式.
    set_base_rel_pathlists(root);

    /*返回一个最终的连接所有表的RelOptInfo */
    rel = make_rel_from_joinlist(root, joinlist);

    /*
     * The result should join all and only the query's base rels.
     */
    Assert(bms_equal(rel->relids, root->all_baserels));

    return rel;
}

make_rel_from_joinlist

joinlist是从where和join on子句找出能做连接操作的对象

static RelOptInfo *
make_rel_from_joinlist(PlannerInfo *root, List *joinlist)
{
    /*
     * Construct a list of rels corresponding to the child joinlist nodes.
     * This may contain both base rels and rels constructed according to
     * sub-joinlists.
     */
    initial_rels = NIL;
    foreach(jl, joinlist)
    {
        if (IsA(jlnode, RangeTblRef))//范围表直接找出要连接的关系
    	{
    		int			varno = ((RangeTblRef *) jlnode)->rtindex;

    		thisrel = find_base_rel(root, varno);
    	}
    	else if (IsA(jlnode, List))//遍历子查询
    	{
    		/* Recurse to handle subproblem */
    		thisrel = make_rel_from_joinlist(root, (List *) jlnode);
    	}

    	initial_rels = lappend(initial_rels, thisrel);
    }

    if (levels_needed == 1)
    {
    }
    else
    {
    	root->initial_rels = initial_rels;

    	if (join_search_hook)
    		return (*join_search_hook) (root, levels_needed, initial_rels);//用户自定义
    	else if (enable_geqo && levels_needed >= geqo_threshold)
    		return geqo(root, levels_needed, initial_rels);//遗传算法
    	else
    		return standard_join_search(root, levels_needed, initial_rels);//动态规划
    }
}

动态规划算法

例如：有一条SQL语句

SELECT * FROM A,B,C,D where A.a=B.a and ...

每层的关系如下：

1. 第四层：ABCD
2. 第三层：ABC,ACD,BCD
3. 第二层：AB,AC,AD,BC,BD...
4. 第一层：A,B,C,D

RelOptInfo *
standard_join_search(PlannerInfo *root, int levels_needed, List *initial_rels)
{
    int			lev;
    RelOptInfo *rel;

    /* root->join_rel_level[j]存放的是第j层的连接路径，
     * 如果有n个关系，最大链接层数就是n。
     */
    root->join_rel_level = (List **) palloc0((levels_needed + 1) * sizeof(List *));

    root->join_rel_level[1] = initial_rels;//初始层数

    for (lev = 2; lev <= levels_needed; lev++)
    {
    	ListCell   *lc;

    	/*使用动态规划求第lev层的所有关系，采用左深树和紧密熟的方式。N=N-1 +1;N=N-k + k */
    	join_search_one_level(root, lev);

    	/*
    	 * Run generate_gather_paths() for each just-processed joinrel.  We
    	 * could not do this earlier because both regular and partial paths
    	 * can get added to a particular joinrel at multiple times within
    	 * join_search_one_level.  After that, we're done creating paths for
    	 * the joinrel, so run set_cheapest().
    	 */
    	foreach(lc, root->join_rel_level[lev])
    	{
    		rel = (RelOptInfo *) lfirst(lc);
            //为lev层每个关系求最优路径
    		set_cheapest(rel);
    	}
    }

    rel = (RelOptInfo *) linitial(root->join_rel_level[levels_needed]);

    root->join_rel_level = NULL;

    return rel;
}