PostgreSQL物理优化中的create_index_paths->generate_bitmap_or_paths函数分析

这篇文章主要讲解了“postgresql物理优化中的create_index_paths->generate_bitmap_or_paths函数分析”，文中的讲解内容简单清晰，易于学习与理解，下面请大家跟着小编的思路慢慢深入，一起来研究和学习“Postgresql物理优化中的create_index_paths->generate_bitmap_or_paths函数分析”吧！

该函数从条件子句链表中寻找OR子句,如找到并且可以处理则生成BitmapOrPath。

下面是BitmapOrPath访问路径样例:

testdb=# explain verbose select t1.* from t_dwxx t1 where (dwbh > '10000' and dwbh < '30000') OR (dwdz between 'DWDZ10000' and 'DWDZ20000');
QUERY PLAN                              
                                                       
---------------------------------------------------------------------------------------------
 Bitmap Heap Scan on public.t_dwxx t1  (cost=84.38..216.82 rows=3156 width=20)
   Output: dwmc, dwbh, dwdz
   Recheck Cond: ((((t1.dwbh)::text > '10000'::text) AND ((t1.dwbh)::text < '30000'::text)) OR (((t1.dwdz)::text >= 'DWDZ1000
0'::text) AND ((t1.dwdz)::text <= 'DWDZ20000'::text)))
   ->  BitmapOr  (cost=84.38..84.38 rows=3422 width=0)  -->BitmapOr
         ->  Bitmap Index Scan on t_dwxx_pkey  (cost=0.00..50.52 rows=2223 width=0)
               Index Cond: (((t1.dwbh)::text > '10000'::text) AND ((t1.dwbh)::text < '30000'::text))
         ->  Bitmap Index Scan on idx_dwxx_dwdz  (cost=0.00..32.28 rows=1200 width=0)
               Index Cond: (((t1.dwdz)::text >= 'DWDZ10000'::text) AND ((t1.dwdz)::text <= 'DWDZ20000'::text))
(8 rows)

一、数据结构

Cost相关
注意:实际使用的参数值通过系统配置文件定义,而不是这里的常量定义!

 typedef double Cost; 

 
 
 
 
 #define DEFAULT_SEQ_PAGE_COST  1.0       //顺序扫描page的成本
 #define DEFAULT_RANDOM_PAGE_COST  4.0      //随机扫描page的成本
 #define DEFAULT_CPU_TUPLE_COST  0.01     //处理一个元组的CPU成本
 #define DEFAULT_CPU_INDEX_TUPLE_COST 0.005   //处理一个索引元组的CPU成本
 #define DEFAULT_CPU_OPERATOR_COST  0.0025    //执行一次操作或函数的CPU成本
 #define DEFAULT_PARALLEL_TUPLE_COST 0.1    //并行执行,从一个worker传输一个元组到另一个worker的成本
 #define DEFAULT_PARALLEL_SETUP_COST  1000.0  //构建并行执行环境的成本
 
 #define DEFAULT_EFFECTIVE_CACHE_SIZE  524288    

 double      seq_page_cost = DEFAULT_SEQ_PAGE_COST;
 double      random_page_cost = DEFAULT_RANDOM_PAGE_COST;
 double      cpu_tuple_cost = DEFAULT_CPU_TUPLE_COST;
 double      cpu_index_tuple_cost = DEFAULT_CPU_INDEX_TUPLE_COST;
 double      cpu_operator_cost = DEFAULT_CPU_OPERATOR_COST;
 double      parallel_tuple_cost = DEFAULT_PARALLEL_TUPLE_COST;
 double      parallel_setup_cost = DEFAULT_PARALLEL_SETUP_COST;
 
 int         effective_cache_size = DEFAULT_EFFECTIVE_CACHE_SIZE;
 Cost        disable_cost = 1.0e10;//1后面10个0,通过设置一个巨大的成本,让优化器自动放弃此路径
 
 int         max_parallel_workers_per_gather = 2;//每次gather使用的worker数

二、源码解读

generate_bitmap_or_paths函数
create_index_paths->generate_bitmap_or_paths函数从条件子句链表中寻找OR子句,如找到并且可以处理则生成BitmapOrPath.函数返回生成的链表BitmapOrPaths.

 
 static List *
 generate_bitmap_or_paths(PlannerInfo *root, RelOptInfo *rel,
                          List *clauses, List *other_clauses)
 {
     List       *result = NIL;
     List       *all_clauses;
     ListCell   *lc;
 
     
     all_clauses = list_concat(list_copy(clauses), other_clauses);//合并到链表中
 
     foreach(lc, clauses)//遍历子句链表
     {
         RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc);//约束条件
         List       *pathlist;//路径链表
         Path       *bitmapqual;//
         ListCell   *j;
 
         
         if (!restriction_is_or_clause(rinfo))//不是OR子句,处理下一个子句
             continue;
 
         
         pathlist = NIL;
         foreach(j, ((BoolExpr *) rinfo->orclause)->args)//遍历OR子句参数
         {
             Node       *orarg = (Node *) lfirst(j);//参数节点
             List       *indlist;
 
             
             //OR子句的参数必须是AND子句或者是子约束条件
             if (and_clause(orarg))//如为AND子句
             {
                 List       *andargs = ((BoolExpr *) orarg)->args;//获取AND子句的参数
 
                 indlist = build_paths_for_OR(root, rel,
                                              andargs,
                                              all_clauses);//构建路径
 
                 
                 //递归调用generate_bitmap_or_paths,并添加到访问路径链表中
                 indlist = list_concat(indlist,
                                       generate_bitmap_or_paths(root, rel,
                                                                andargs,
                                                                all_clauses));
             }
             else
             {
                 RestrictInfo *rinfo = castNode(RestrictInfo, orarg);//不是AND,则为约束条件
                 List       *orargs;
 
                 Assert(!restriction_is_or_clause(rinfo));
                 orargs = list_make1(rinfo);
 
                 indlist = build_paths_for_OR(root, rel,
                                              orargs,
                                              all_clauses);//构建访问路径
             }
 
             
             if (indlist == NIL)
             {
                 pathlist = NIL;
                 break;
             }
 
             
             bitmapqual = choose_bitmap_and(root, rel, indlist);
             pathlist = lappend(pathlist, bitmapqual);
         }
 
         
         if (pathlist != NIL)
         {
       //创建BitmapOrPath
             bitmapqual = (Path *) create_bitmap_or_path(root, rel, pathlist);
             result = lappend(result, bitmapqual);
         }
     }
 
     return result;
 }
 
//------------------------------------------------------ build_paths_for_OR
 
 static List *
 build_paths_for_OR(PlannerInfo *root, RelOptInfo *rel,
                    List *clauses, List *other_clauses)
 {
     List       *result = NIL;//返回结果
     List       *all_clauses = NIL;  
     ListCell   *lc;//临时变量
 
     foreach(lc, rel->indexlist)//遍历RelOptInfo上的Index
     {
         IndexOptInfo *index = (IndexOptInfo *) lfirst(lc);//IndexOptInfo
         IndexClauseSet clauseset;//
         List       *indexpaths;
         bool        useful_predicate;
 
         
         if (!index->amhasgetbitmap)//索引不支持BitmapindexScan
             continue;
 
         
         useful_predicate = false;
         if (index->indpred != NIL)
         {
             if (index->predOK)
             {
                 
             }
             else
             {
                 
                 if (all_clauses == NIL)
                     all_clauses = list_concat(list_copy(clauses),
                                               other_clauses);
 
                 if (!predicate_implied_by(index->indpred, all_clauses, false))
                     continue;   
 
                 if (!predicate_implied_by(index->indpred, other_clauses, false))
                     useful_predicate = true;
             }
         }
 
         
         MemSet(&clauseset, 0, sizeof(clauseset));
         match_clauses_to_index(index, clauses, &clauseset);
 
         
         if (!clauseset.nonempty && !useful_predicate)//没有合适的,继续下一个索引
             continue;
 
         
         match_clauses_to_index(index, other_clauses, &clauseset);//添加到clauseset中
 
         
         indexpaths = build_index_paths(root, rel,
                                        index, &clauseset,
                                        useful_predicate,
                                        ST_BITMAPSCAN,
                                        NULL,
                                        NULL);//构建索引访问路径
         result = list_concat(result, indexpaths);
     }
 
     return result;
 }
 

//------------------------------------------------------ create_bitmap_or_path

 
 BitmapOrPath *
 create_bitmap_or_path(PlannerInfo *root,
                       RelOptInfo *rel,
                       List *bitmapquals)
 {
     BitmapOrPath *pathnode = makeNode(BitmapOrPath);
 
     pathnode->path.pathtype = T_BitmapOr;
     pathnode->path.parent = rel;
     pathnode->path.pathtarget = rel->reltarget;
     pathnode->path.param_info = NULL;   
 
     
     pathnode->path.parallel_aware = false;
     pathnode->path.parallel_safe = rel->consider_parallel;
     pathnode->path.parallel_workers = 0;
 
     pathnode->path.pathkeys = NIL;  
 
     pathnode->bitmapquals = bitmapquals;
 
     
     cost_bitmap_or_node(pathnode, root);//计算成本
 
     return pathnode;
 }

//------------------------------------ create_bitmap_or_path

 
 void
 cost_bitmap_or_node(BitmapOrPath *path, PlannerInfo *root)
 {
     Cost        totalCost;
     Selectivity selec;
     ListCell   *l;
 
     
     totalCost = 0.0;//成本
     selec = 0.0;//选择率
     foreach(l, path->bitmapquals)//遍历条件
     {
         Path       *subpath = (Path *) lfirst(l);//路径
         Cost        subCost;//成本
         Selectivity subselec;
 
         cost_bitmap_tree_node(subpath, &subCost, &subselec);//遍历路径获取成本&选择率
 
         selec += subselec;//
 
         totalCost += subCost;
         if (l != list_head(path->bitmapquals) &&
             !IsA(subpath, IndexPath))
             totalCost += 100.0 * cpu_operator_cost;//非单个条件而且不是索引访问路径,则添加运行期成本
     }
     path->bitmapselectivity = Min(selec, 1.0);//选择率
     path->path.rows = 0;        
     path->path.startup_cost = totalCost;
     path->path.total_cost = totalCost;
 }

三、跟踪分析

测试脚本如下

select t1.* 
from t_dwxx t1 
where (dwbh > '10000' and dwbh < '30000') 
      OR (dwdz between 'DWDZ10000' and 'DWDZ20000');

启动gdb跟踪

(gdb) b generate_bitmap_or_paths
Breakpoint 1 at 0x74e6c1: file indxpath.c, line 1266.
(gdb) c
Continuing.

Breakpoint 1, generate_bitmap_or_paths (root=0x2aa6248, rel=0x2aa6658, clauses=0x2aaf138, other_clauses=0x0)
    at indxpath.c:1266
1266    List     *result = NIL;

查看输入参数,clauses是链表,只有一个元素(BoolExpr类型,即OR子句);other_clauses为NULL

(gdb) p *clauses
$1 = {type = T_List, length = 1, head = 0x2aaf118, tail = 0x2aaf118}
(gdb) p *(Node *)clauses->head->data.ptr_value
$2 = {type = T_RestrictInfo}
(gdb) p *(RestrictInfo *)clauses->head->data.ptr_value
$3 = {type = T_RestrictInfo, clause = 0x2aad818, is_pushed_down = true, outerjoin_delayed = false, can_join = false, 
  pseudoconstant = false, leakproof = false, security_level = 0, clause_relids = 0x2aaf100, required_relids = 0x2aae938, 
  outer_relids = 0x0, nullable_relids = 0x0, left_relids = 0x0, right_relids = 0x0, orclause = 0x2aaefc0, parent_ec = 0x0, 
  eval_cost = {startup = 0, per_tuple = 0.01}, nORM_selec = 0.31556115090433856, outer_selec = -1, mergeopfamilies = 0x0, 
  left_ec = 0x0, right_ec = 0x0, left_em = 0x0, right_em = 0x0, scansel_cache = 0x0, outer_is_left = false, 
  hashjoinoperator = 0, left_bucketsize = -1, right_bucketsize = -1, left_mcvfreq = -1, right_mcvfreq = -1}
(gdb) p *((RestrictInfo *)clauses->head->data.ptr_value)->clause
$4 = {type = T_BoolExpr}
(gdb) set $clause=((RestrictInfo *)clauses->head->data.ptr_value)->clause
(gdb) p *(BoolExpr *)$clause
$6 = {xpr = {type = T_BoolExpr}, boolop = OR_EXPR, args = 0x2aad758, location = -1}

遍历clausees子句,rinfo->clause即BoolExpr(OR子句)

...
1276    foreach(lc, clauses)
(gdb) 
1278      RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc);
(gdb) 
1284      if (!restriction_is_or_clause(rinfo))

遍历OR子句的参数

(gdb) 
1292      foreach(j, ((BoolExpr *) rinfo->orclause)->args)

参数的第一个元素,BoolExpr,boolop操作符为AND_EXPR

(gdb) n
1294        Node     *orarg = (Node *) lfirst(j);
(gdb) 
1298        if (and_clause(orarg))
(gdb) p *orarg
$10 = {type = T_BoolExpr}
(gdb) p *(BoolExpr *)orarg
$11 = {xpr = {type = T_BoolExpr}, boolop = AND_EXPR, args = 0x2aaea90, location = -1}

AND子句的参数

(gdb) n
1300          List     *andargs = ((BoolExpr *) orarg)->args;
(gdb) 
1302          indlist = build_paths_for_OR(root, rel,
(gdb) p *andargs
$12 = {type = T_List, length = 2, head = 0x2aada78, tail = 0x2aada98}
(gdb) p *(Node *)andargs->head->data.ptr_value
$13 = {type = T_RestrictInfo}
(gdb) p *(RestrictInfo *)andargs->head->data.ptr_value
$14 = {type = T_RestrictInfo, clause = 0x2aace08, is_pushed_down = true, outerjoin_delayed = false, can_join = false, 
  pseudoconstant = false, leakproof = false, security_level = 0, clause_relids = 0x2aaea78, required_relids = 0x2aaea78, 
  outer_relids = 0x0, nullable_relids = 0x0, left_relids = 0x2aaea60, right_relids = 0x0, orclause = 0x0, parent_ec = 0x0, 
  eval_cost = {startup = 0, per_tuple = 0.0025000000000000001}, norm_selec = 0.99990000000000001, outer_selec = -1, 
  mergeopfamilies = 0x0, left_ec = 0x0, right_ec = 0x0, left_em = 0x0, right_em = 0x0, scansel_cache = 0x0, 
  outer_is_left = false, hashjoinoperator = 0, left_bucketsize = -1, right_bucketsize = -1, left_mcvfreq = -1, 
  right_mcvfreq = -1}
(gdb) p *((RestrictInfo *)andargs->head->data.ptr_value)->clause
$15 = {type = T_OpExpr}
(gdb) set $tmp=((RestrictInfo *)andargs->head->data.ptr_value)->clause
(gdb) p *(OpExpr *)$tmp
$16 = {xpr = {type = T_OpExpr}, opno = 666, opfuncid = 742, opresulttype = 16, opretset = false, opcollid = 0, 
  inputcollid = 100, args = 0x2aacd68, location = 39}
(gdb) set $tmp2=((RestrictInfo *)andargs->head->next->data.ptr_value)->clause
(gdb) p *(OpExpr *)$tmp2
$17 = {xpr = {type = T_OpExpr}, opno = 664, opfuncid = 740, opresulttype = 16, opretset = false, opcollid = 0, 
  inputcollid = 100, args = 0x2aacc78, location = 58}
(gdb)

进入build_paths_for_OR函数

(gdb) step
build_paths_for_OR (root=0x2aa6248, rel=0x2aa6658, clauses=0x2aaea90, other_clauses=0x2aaf598) at indxpath.c:1170
1170    List     *result = NIL;

遍历索引,第一个索引是idx_dwxx_dwdz

1174    foreach(lc, rel->indexlist)
(gdb) 
1176      IndexOptInfo *index = (IndexOptInfo *) lfirst(lc);
(gdb) 
1182      if (!index->amhasgetbitmap)
(gdb) p *index
$18 = {type = T_IndexOptInfo, indexoid = 16753, reltablespace = 0, rel = 0x2aa6658, pages = 40, tuples = 10000, 
  tree_height = 1, ncolumns = 1, nkeycolumns = 1, indexkeys = 0x2aae590, indexcollations = 0x2aae5a8, opfamily = 0x2aae5c0, 
  opcintype = 0x2aae5d8, sortopfamily = 0x2aae5c0, reverse_sort = 0x2aae608, nulls_first = 0x2aae620, 
  canreturn = 0x2aae5f0, relam = 403, indexprs = 0x0, indpred = 0x0, indextlist = 0x2aae6f8, indrestrictinfo = 0x2aaf138, 
  predOK = false, unique = false, immediate = true, hypothetical = false, amcanorderbyop = false, amoptionalkey = true, 
  amsearcharray = true, amsearchnulls = true, amhasgettuple = true, amhasgetbitmap = true, amcanparallel = true, 
  amcostestimate = 0x94f0ad <btcostestimate>}
--
testdb=# select relname from pg_class where oid=16753;
    relname    
---------------
 idx_dwxx_dwdz
(1 row)
--

与约束条件不匹配((dwbh > '10000' and dwbh < '30000')),继续下一个索引

1229      if (!clauseset.nonempty && !useful_predicate)
(gdb) p clauseset
$20 = {nonempty = false, indexclauses = {0x0 <repeats 32 times>}}
(gdb) n
1230        continue;

下一个索引是idx_dwxx_predicate_dwmc/idx_dwxx_expr,同样不匹配,继续寻找索引,直至索引t_dwxx_pkey

(gdb) p *index
$23 = {type = T_IndexOptInfo, indexoid = 16738,...
1223      match_clauses_to_index(index, clauses, &clauseset);
(gdb) 
1229      if (!clauseset.nonempty && !useful_predicate)
(gdb) p clauseset
$24 = {nonempty = true, indexclauses = {0x2aaf638, 0x0 <repeats 31 times>}}

构建索引访问路径

(gdb) 
1246      result = list_concat(result, indexpaths);
(gdb) p *indexpaths
$25 = {type = T_List, length = 1, head = 0x2aafb48, tail = 0x2aafb48}
(gdb) p *(Node *)indexpaths->head->data.ptr_value
$26 = {type = T_IndexPath}
(gdb) p *(IndexPath *)indexpaths->head->data.ptr_value
$27 = {path = {type = T_IndexPath, pathtype = T_IndexScan, parent = 0x2aa6658, pathtarget = 0x2aad8d8, param_info = 0x0, 
    parallel_aware = false, parallel_safe = true, parallel_workers = 0, rows = 3156, startup_cost = 0.28500000000000003, 
    total_cost = 191.46871600907946, pathkeys = 0x0}, indexinfo = 0x2aa6868, indexclauses = 0x2aaf6a8, 
  indexquals = 0x2aaf898, indexqualcols = 0x2aaf8e8, indexorderbys = 0x0, indexorderbycols = 0x0, 
  indexscandir = ForwardScanDirection, indextotalcost = 50.515000000000001, indexselectivity = 0.22227191011235958}

回到generate_bitmap_or_paths函数

1250  }
(gdb) 
generate_bitmap_or_paths (root=0x2aa6248, rel=0x2aa6658, clauses=0x2aaf138, other_clauses=0x0) at indxpath.c:1307
1307          indlist = list_concat(indlist,

递归进入generate_bitmap_or_paths

(gdb) n

Breakpoint 1, generate_bitmap_or_paths (root=0x2aa6248, rel=0x2aa6658, clauses=0x2aaea90, other_clauses=0x2aaf598)
    at indxpath.c:1266
1266    List     *result = NIL;
#直接结束
(gdb) finish
Run till exit from #0  generate_bitmap_or_paths (root=0x2aa6248, rel=0x2aa6658, clauses=0x2aaea90, other_clauses=0x2aaf598)
    at indxpath.c:1266
0x000000000074e7a0 in generate_bitmap_or_paths (root=0x2aa6248, rel=0x2aa6658, clauses=0x2aaf138, other_clauses=0x0)
    at indxpath.c:1307
1307          indlist = list_concat(indlist,
Value returned is $28 = (List *) 0x0

完成第一轮循环

(gdb) n
1329        if (indlist == NIL)
(gdb) n
1339        bitmapqual = choose_bitmap_and(root, rel, indlist);
(gdb) 
1340        pathlist = lappend(pathlist, bitmapqual);
(gdb) p *bitmapqual
$29 = {type = T_IndexPath, pathtype = T_IndexScan, parent = 0x2aa6658, pathtarget = 0x2aad8d8, param_info = 0x0, 
  parallel_aware = false, parallel_safe = true, parallel_workers = 0, rows = 3156, startup_cost = 0.28500000000000003, 
  total_cost = 191.46871600907946, pathkeys = 0x0}

这是第二个AND子句

1292      foreach(j, ((BoolExpr *) rinfo->orclause)->args)
(gdb) 
1294        Node     *orarg = (Node *) lfirst(j);
(gdb) 
1298        if (and_clause(orarg))
(gdb) 
1300          List     *andargs = ((BoolExpr *) orarg)->args;

完成第二轮循环

(gdb) 
1339        bitmapqual = choose_bitmap_and(root, rel, indlist);
(gdb) 
1340        pathlist = lappend(pathlist, bitmapqual);
(gdb) p bitmapqual
$33 = (Path *) 0x2aafd78
(gdb) p *bitmapqual
$34 = {type = T_IndexPath, pathtype = T_IndexScan, parent = 0x2aa6658, pathtarget = 0x2aad8d8, param_info = 0x0, 
  parallel_aware = false, parallel_safe = true, parallel_workers = 0, rows = 3156, startup_cost = 0.28500000000000003, 
  total_cost = 148.08735471522883, pathkeys = 0x0}

结束循环,构建BitmapOrPath

1347      if (pathlist != NIL)
(gdb) 
1349        bitmapqual = (Path *) create_bitmap_or_path(root, rel, pathlist);

进入create_bitmap_or_path,调用函数cost_bitmap_or_node计算成本

(gdb) step
create_bitmap_or_path (root=0x2aa6248, rel=0x2aa6658, bitmapquals=0x2aafbf8) at pathnode.c:1156
1156    BitmapOrPath *pathnode = makeNode(BitmapOrPath);
...
1178    cost_bitmap_or_node(pathnode, root);
(gdb) step
cost_bitmap_or_node (path=0x2ab0278, root=0x2aa6248) at costsize.c:1149
...

计算结果,与执行计划中的信息相匹配"BitmapOr  (cost=84.38..84.38 rows=3422 width=0)"

(gdb) p *path
$37 = {path = {type = T_BitmapOrPath, pathtype = T_BitmapOr, parent = 0x2aa6658, pathtarget = 0x2aad8d8, param_info = 0x0, 
    parallel_aware = false, parallel_safe = true, parallel_workers = 0, rows = 0, startup_cost = 84.378, 
    total_cost = 84.378, pathkeys = 0x0}, bitmapquals = 0x2aafbf8, bitmapselectivity = 0.34222288270157986}

回到generate_bitmap_or_paths

(gdb) n
create_bitmap_or_path (root=0x2aa6248, rel=0x2aa6658, bitmapquals=0x2aafbf8) at pathnode.c:1180
1180    return pathnode;
(gdb) 
1181  }
(gdb) 
generate_bitmap_or_paths (root=0x2aa6248, rel=0x2aa6658, clauses=0x2aaf138, other_clauses=0x0) at indxpath.c:1350
1350        result = lappend(result, bitmapqual);

感谢各位的阅读，以上就是“PostgreSQL物理优化中的create_index_paths->generate_bitmap_or_paths函数分析”的内容了，经过本文的学习后，相信大家对PostgreSQL物理优化中的create_index_paths->generate_bitmap_or_paths函数分析这一问题有了更深刻的体会，具体使用情况还需要大家实践验证。这里是编程网，小编将为大家推送更多相关知识点的文章，欢迎关注！