openGauss内核：SQL解析过程分析

摘要：在传统数据库中SQL引擎一般指对用户输入的SQL语句进行解析、优化的软件模块。SQL的解析过程主要分为：词法、语法和语义分析。

本文分享自华为云社区《 openGauss内核分析（三)：SQL解析》，作者：Gauss松鼠会。

在传统数据库中SQL引擎一般指对用户输入的SQL语句进行解析、优化的软件模块。

SQL的解析过程主要分为：

• 词法分析：将用户输入的SQL语句拆解成单词(Token)序列，并识别出关键字、标识、常量等。

• 语法分析：分析器对词法分析器解析出来的单词(Token)序列在语法上是否满足SQL语法规则。

• 语义分析：语义分析是SQL解析过程的一个逻辑阶段，主要任务是在语法正确的基础上进行上下文有关性质的审查，在SQL解析过程中该阶段完成表名、操作符、类型等元素的合法性判断，同时检测语义上的二义性。

openGauss在pg_parse_query中调用raw_parser函数对用户输入的SQL命令进行词法分析和语法分析，生成语法树添加到链表parsetree_list中。完成语法分析后，对于parsetree_list中的每一颗语法树parsetree，会调用parse_**yze函数进行语义分析，根据SQL命令的不同，执行对应的入口函数，最终生成查询树。

词法分析

openGauss使用flex工具进行词法分析。flex工具通过对已经定义好的词法文件进行编译，生成词法分析的代码。词法文件是scan.l，它根据SQL语言标准对SQL语言中的关键字、标识符、操作符、常量、终结符进行了定义和识别。在kwlist.h中定义了大量的关键字，按照字母的顺序排列，方便在查找关键字时通过二分法进行查找。在scan.l中处理“标识符”时，会到关键字列表中进行匹配，如果一个标识符匹配到关键字，则认为是关键字，否则才是标识符，即关键字优先. 以“select a, b from item”为例说明词法分析结果。

语法分析

openGauss中定义了bison工具能够识别的语法文件gram.y，根据SQL语言的不同定义了一系列表达Statement的结构体（这些结构体通常以Stmt作为命名后缀），用来保存语法分析结果。以SELECT查询为例，它对应的Statement结构体如下。

typedef struct SelectStmt

{

    NodeTag        type;

    List       *distinctClause; /* NULL, list of DISTINCT ON exprs, or

                                 * lcons(NIL,NIL) for all (SELECT DISTINCT) */

    IntoClause *intoClause;        /* target for SELECT INTO */

    List       *targetList;        /* the target list (of ResTarget) */

    List       *fromClause;        /* the FROM clause */

    Node       *whereClause;    /* WHERE qualification */

    List       *groupClause;    /* GROUP BY clauses */

    Node       *havingClause;    /* HAVING conditional-expression */

    List       *windowClause;    /* WINDOW window_name AS (...), ... */

    WithClause *withClause;        /* WITH clause */

    List       *valuesLists;    /* untransformed list of expression lists */

    List       *sortClause;        /* sort clause (a list of SortBy's) */

    Node       *limitOffset;    /* # of result tuples to skip */

    Node       *limitCount;        /* # of result tuples to return */

    ……

} SelectStmt;

这个结构体可以看作一个多叉树，每个叶子节点都表达了SELECT查询语句中的一个语法结构，对应到gram.y中，它会有一个SelectStmt。代码如下：

从simple_select语法分析结构可以看出，一条简单的查询语句由以下子句组成：去除行重复的distinctClause、目标属性targetList、SELECT INTO子句intoClause、FROM子句fromClause、WHERE子句whereClause、GROUP BY子句groupClause、HAVING子句havingClause、窗口子句windowClause和plan_hint子句。在成功匹配simple_select语法结构后，将会创建一个Statement结构体，将各个子句进行相应的赋值。对simple_select而言，目标属性、FROM子句、WHERE子句是最重要的组成部分。SelectStmt与其他结构体的关系如下：

下面以“select a, b from item”为例说明简单select语句的解析过程，函数exec_simple_query调用pg_parse_query执行解析，解析树中只有一个元素。

(gdb) p *parsetree_list

$47 = {type = T_List, length = 1, head = 0x7f5ff986c8f0, tail = 0x7f5ff986c8f0}

List中的节点类型为T_SelectStmt。

(gdb) p *(Node *)(parsetree_list->head.data->ptr_value)

$45 = {type = T_SelectStmt}

查看SelectStmt结构体，targetList 和fromClause非空。

(gdb) set $stmt = (SelectStmt *)(parsetree_list->head.data->ptr_value)

(gdb) p *$stmt

$50 = {type = T_SelectStmt, distinctClause = 0x0, intoClause = 0x0, targetList = 0x7f5ffa43d588, fromClause = 0x7f5ff986c888, startWithClause = 0x0, whereClause = 0x0, groupClause = 0x0,

  havingClause = 0x0, windowClause = 0x0, withClause = 0x0, valuesLists = 0x0, sortClause = 0x0, limitOffset = 0x0, limitCount = 0x0, lockingClause = 0x0, hintState = 0x0, op = SETOP_NONE, all = false,

  larg = 0x0, rarg = 0x0, hasPlus = false}

查看SelectStmt的targetlist，有两个ResTarget。

(gdb) p *($stmt->targetList)

$55 = {type = T_List, length = 2, head = 0x7f5ffa43d540, tail = 0x7f5ffa43d800}

(gdb) p *(Node *)($stmt->targetList->head.data->ptr_value)

$57 = {type = T_ResTarget}

(gdb) set $restarget1=(ResTarget *)($stmt->targetList->head.data->ptr_value)

(gdb) p *$restarget1

$60 = {type = T_ResTarget, name = 0x0, indirection = 0x0, val = 0x7f5ffa43d378, location = 7}

(gdb) p *$restarget1->val

$63 = {type = T_ColumnRef}

(gdb) p *(ColumnRef *)$restarget1->val

$64 = {type = T_ColumnRef, fields = 0x7f5ffa43d470, prior = false, indnum = 0, location = 7}

(gdb) p *((ColumnRef *)$restarget1->val)->fields

$66 = {type = T_List, length = 1, head = 0x7f5ffa43d428, tail = 0x7f5ffa43d428}

(gdb) p *(Node *)(((ColumnRef *)$restarget1->val)->fields)->head.data->ptr_value

$67 = {type = T_String}

(gdb) p *(Value *)(((ColumnRef *)$restarget1->val)->fields)->head.data->ptr_value

$77 = {type = T_String, val = {ival = 140050197369648, str = 0x7f5ffa43d330 "a"}}

(gdb) set $restarget2=(ResTarget *)($stmt->targetList->tail.data->ptr_value)

(gdb) p *$restarget2

$89 = {type = T_ResTarget, name = 0x0, indirection = 0x0, val = 0x7f5ffa43d638, location = 10}

(gdb) p *$restarget2->val

$90 = {type = T_ColumnRef}

(gdb) p *(ColumnRef *)$restarget2->val

$91 = {type = T_ColumnRef, fields = 0x7f5ffa43d730, prior = false, indnum = 0, location = 10}

(gdb) p *((ColumnRef *)$restarget2->val)->fields

$92 = {type = T_List, length = 1, head = 0x7f5ffa43d6e8, tail = 0x7f5ffa43d6e8}

(gdb) p *(Node *)(((ColumnRef *)$restarget2->val)->fields)->head.data->ptr_value

$93 = {type = T_String}

(gdb) p *(Value *)(((ColumnRef *)$restarget2->val)->fields)->head.data->ptr_value

$94 = {type = T_String, val = {ival = 140050197370352, str = 0x7f5ffa43d5f0 "b"}}

查看SelectStmt的fromClause，有一个RangeVar。

(gdb) p *$stmt->fromClause

$102 = {type = T_List, length = 1, head = 0x7f5ffa43dfe0, tail = 0x7f5ffa43dfe0}

(gdb) set $fromclause=(RangeVar*)($stmt->fromClause->head.data->ptr_value)

(gdb) p *$fromclause

$103 = {type = T_RangeVar, catalogname = 0x0, schemaname = 0x0, relname = 0x7f5ffa43d848 "item", partitionname = 0x0, subpartitionname = 0x0, inhOpt = INH_DEFAULT, relpersistence = 112 'p', alias = 0x0,

  location = 17, ispartition = false, issubpartition = false, partitionKeyValuesList = 0x0, isbucket = false, buckets = 0x0, length = 0, foreignOid = 0, withVerExpr = false}

综合以上分析可以得到语法树结构。

语义分析

在完成词法分析和语法分析后，parse_Ana lyze函数会根据语法树的类型，调用transformSelectStmt将parseTree改写为查询树。

(gdb) p *result

$3 = {type = T_Query, commandType = CMD_SELECT, querySource = QSRC_ORIGINAL, queryId = 0, canSetTag = false, utilityStmt = 0x0, resultRelation = 0, hasAggs = false, hasWindowFuncs = false,

  hasSubLinks = false, hasDistinctOn = false, hasRecursive = false, hasModifyingCTE = false, hasForUpdate = false, hasRowSecurity = false, hasSynonyms = false, cteList = 0x0, rtable = 0x7f5ff5eb8c88,

  jointree = 0x7f5ff5eb9310, targetList = 0x7f5ff5eb9110,…}

(gdb) p *result->targetList

$13 = {type = T_List, length = 2, head = 0x7f5ff5eb90c8, tail = 0x7f5ff5eb92c8}

(gdb) p *(Node *)(result->targetList->head.data->ptr_value)

$8 = {type = T_TargetEntry}

(gdb) p *(TargetEntry*)(result->targetList->head.data->ptr_value)

$9 = {xpr = {type = T_TargetEntry, selec = 0}, expr = 0x7f5ff636ff48, resno = 1, resname = 0x7f5ff5caf330 "a", ressortgroupref = 0, resorigtbl = 24576, resorigcol = 1, resjunk = false}

(gdb) p *(TargetEntry*)(result->targetList->tail.data->ptr_value)

$10 = {xpr = {type = T_TargetEntry, selec = 0}, expr = 0x7f5ff5eb9178, resno = 2, resname = 0x7f5ff5caf5f0 "b", ressortgroupref = 0, resorigtbl = 24576, resorigcol = 2, resjunk = false}

(gdb)

(gdb) p *result->rtable

$14 = {type = T_List, length = 1, head = 0x7f5ff5eb8c40, tail = 0x7f5ff5eb8c40}

(gdb)  p *(Node *)(result->rtable->head.data->ptr_value)

$15 = {type = T_RangeTblEntry}

(gdb) p *(RangeTblEntry*)(result->rtable->head.data->ptr_value)

$16 = {type = T_RangeTblEntry, rtekind = RTE_RELATION, relname = 0x7f5ff636efb0 "item", partAttrNum = 0x0, relid = 24576, partitionOid = 0, isContainPartition = false, subpartitionOid = 0……}

得到的查询树结构如下：

完成词法、语法和语义分析后，SQL解析过程完成，SQL引擎开始执行查询优化，在下一期中再具体分析。

点击关注，第一时间了解华为云新鲜技术~