openGauss内核分析(六): 执行计划生成
  lYE0sTgD5uUi 2023年11月02日 97 0

openGauss内核分析(六): 执行计划生成

SQL语句解析完成后被解析成Query结构,在进行优化时是以Query为单位进行的,Query的优化分为基于规则的逻辑优化(查询重写)和基于代价的物理优化(计划生成),主入口函数为subquery_planner。subquery_planner函数接收Query(查询树),返回一个Plan(计划树)。

Plan* subquery_planner(PlannerGlobal* glob, Query* parse, PlannerInfo* parent_root, bool hasRecursion,
double tuple_fraction, PlannerInfo** subroot, int options, ItstDisKey* diskeys, List* subqueryRestrictInfo)
{
PlannerInfo* root = NULL;
Plan* plan = NULL; //返回结果
…
preprocess_const_params(root, (Node*)parse->jointree); // 常数替换等式
…
if (parse->hasSubLinks) {
pull_up_sublinks(root); //提升子链接
DEBUG_QRW("After sublink pullup");
}
/* Reduce orderby clause in subquery for join */
reduce_orderby(parse, false); //减少orderby
DEBUG_QRW("After order by reduce");
if (u_sess->attr.attr_sql.enable_constraint_optimization) {
removeNotNullTest(root); //删除NotNullTest
DEBUG_QRW("After soft constraint removal");
}
…
if ((LAZY_AGG & u_sess->attr.attr_sql.rewrite_rule) && permit_from_rewrite_hint(root, LAZY_AGG)) {
lazyagg_main(parse); // lazyagg重写
DEBUG_QRW("After lazyagg");
}
…
parse->jointree = (FromExpr*)pull_up_subqueries(root, (Node*)parse->jointree); //提升子查询
…
if (parse->setOperations) {
flatten_simple_union_all(root); //UNIONALL优化
DEBUG_QRW("After simple union all flatten");
}
…
expand_inherited_tables(root); //展开继承表
…
parse->targetList = (List*)preprocess_expression(root, (Node*)parse->targetList, EXPRKIND_TARGET); //预处理表达式
…
parse->havingQual = (Node *) newHaving; //处理HAVING子句
…
reduce_outer_joins(root); //外连接消除
…
reduce_inequality_fulljoins(root); //全连接重写
   …
plan = grouping_planner(root, tuple_fraction); //主要的计划过程
…
return plan;
}

subquery_planner函数由函数standard_planner调用,standard_planner函数由exec_simple_query->pg_plan_queries->pg_plan_query->planner函数调用。standard_planner将Query(查询树)生成规划好的语句,可用于执行器实际执行。

PlannedStmt* standard_planner(Query* parse, int cursorOptions, ParamListInfo boundParams)
{
PlannedStmt* result = NULL;  //返回结果
PlannerGlobal* glob = NULL;
double tuple_fraction;
PlannerInfo* root = NULL;
Plan* top_plan = NULL;
…
glob = makeNode(PlannerGlobal);
/* primary planning entry point (may recurse for subqueries) */
top_plan = subquery_planner(glob, parse, NULL, false, tuple_fraction, &root);  //主规划过程入口
…
/* build the PlannedStmt result */
result = makeNode(PlannedStmt); //构造PlannedStmt
result->commandType = parse->commandType;
result->queryId = parse->queryId;
result->uniqueSQLId = parse->uniqueSQLId;
result->hasReturning = (parse->returningList != NIL);
result->hasModifyingCTE = parse->hasModifyingCTE;
result->canSetTag = parse->canSetTag;
result->transientPlan = glob->transientPlan;
result->dependsOnRole = glob->dependsOnRole;
result->planTree = top_plan; //执行计划
result->rtable = glob->finalrtable;
result->resultRelations = glob->resultRelations;
…
return result;
}

仍然以前文的join列子来说明:

SELECT * FROM t1 inner JOIN t2 ON t1.c1 = t2.c1;

在planner函数打断点,用gdb查看standard_planner返回的PlannedStmt:

(gdb) bt
#0 planner (parse=0x7fd93a410288, cursorOptions=0, boundParams=0x0) at planner.cpp:389
#1 0x0000000001936fbd in pg_plan_query (querytree=0x7fd93a410288, cursorOptions=0, boundParams=0x0, underExplain=false) at postgres.cpp:1197
#2 0x0000000001937381 in pg_plan_queries (querytrees=0x7fd939b81090, cursorOptions=0, boundParams=0x0) at postgres.cpp:1315
#3 0x000000000193a6b8 in exec_simple_query (query_string=0x7fd966ad2060 "SELECT * FROM t1 inner JOIN t2 ON t1.c1 = t2.c1;", messageType=QUERY_MESSAGE, msg=0x7fd931056210)
at postgres.cpp:2560
#4 0x0000000001947104 in PostgresMain (argc=1, argv=0x7fd93a2cf1c0, dbname=0x7fd93a2ce1f8 "postgres", username=0x7fd93a2ce1b0 "test") at postgres.cpp:8403
#5 0x0000000001890740 in BackendRun (port=0x7fd931056720) at postmaster.cpp:8053
#6 0x00000000018a00b1 in GaussDbThreadMain<(knl_thread_role)1> (arg=0x7fd97c55c5f0) at postmaster.cpp:12181
#7 0x000000000189c0de in InternalThreadFunc (args=0x7fd97c55c5f0) at postmaster.cpp:12755
#8 0x00000000024bf7d8 in ThreadStarterFunc (arg=0x7fd97c55c5e0) at gs_thread.cpp:382
#9 0x00007fd9a60cfdd5 in start_thread () from /lib64/libpthread.so.0
#10 0x00007fd9a5df8ead in clone () from /lib64/libc.so.6
(gdb) p *result
$14 = {type = T_PlannedStmt, commandType = CMD_SELECT, queryId = 0, hasReturning = false, hasModifyingCTE = false, canSetTag = true, transientPlan = false, dependsOnRole = false,
planTree = 0x7fd93a409d58, rtable = 0x7fd939b81660, …}
(gdb) p *result->planTree->lefttree
$46 = {type = T_SeqScan, plan_node_id = 2, parent_node_id = 1, exec_type = EXEC_ON_DATANODES, startup_cost = 0, total_cost = 1.03, plan_rows = 3, multiple = 1, plan_width = 8,…}

将Query规划后得到PlannedStmt:

openGauss内核分析(六): 执行计划生成_List

openGauss内核分析(六): 执行计划生成_List_02

可以看到,Plannedstmt 与explain执行计划是一致的:

openGauss内核分析(六): 执行计划生成_返回结果_03

【版权声明】本文内容来自摩杜云社区用户原创、第三方投稿、转载,内容版权归原作者所有。本网站的目的在于传递更多信息,不拥有版权,亦不承担相应法律责任。如果您发现本社区中有涉嫌抄袭的内容,欢迎发送邮件进行举报,并提供相关证据,一经查实,本社区将立刻删除涉嫌侵权内容,举报邮箱: cloudbbs@moduyun.com

  1. 分享:
最后一次编辑于 2023年11月08日 0

暂无评论

推荐阅读
  lh6O4DgR0ZQ8   2023年11月02日   61   0   0 javaListjson
  ZXL8ALkrtBPG   2023年11月02日   74   0   0 Listcpp
  X5zJxoD00Cah   2023年11月02日   56   0   0 javaList函数返回
  X5zJxoD00Cah   2023年11月02日   49   0   0 进程名返回结果
  zEbHO6cVnp49   2023年11月02日   43   0   0 SystemListjson
lYE0sTgD5uUi