Hive SQL血缘解析
- Druid可以直接获得所有的列
http://t.csdn.cn/mO4TX - 利用Hive提供的LineageLogger与Execution Hooks机制做血缘
https://blog.csdn.net/qq_44831907/article/details/123033137 - Apache Calcite
- gudusoft 解析方案 商业
https://blog.csdn.net/qq_31557939/article/details/126277212
6.github开源项目：
https://github.com/Shkin1/hathor
https://github.com/sqlparser/sqlflow_public
===========================
一、数仓经常会碰到的几类问题：
1、两个数据报表进行对比，结果差异很大，需要人工核对分析指标的维度信息，比如从头分析数据指标从哪里来，处理条件是什么，最后才能分析出问题原因。
2、基础数据表因某种原因需要修改字段，需要评估其对数仓的影响，费时费力，然后在做方案。
二、问题分析：
数据源长途跋涉，经过大量的处理和组件来传递，呈现在业务用户面前，对数据进行回溯其实很难。元数据回溯在有效决策、策略制定、差异分析等过程中很重要。这两类问题都属于数据血缘分析问题，第一类叫做数据回溯、第二类叫做影响分析，是数据回溯的逆向。
三、解决方法：
自己实现了一套基于hive数仓的数据血缘分析工具，来完成各个数据表、字段之间的关系梳理，进而解决上面两个问题。
工具主要目标：解析计算脚本中的HQL语句，分析得到输入输出表、输入输出字段和相应的处理条件，进行分析展现。
实现思路：对AST深度优先遍历，遇到操作的token则判断当前的操作，遇到子句则压栈当前处理，处理子句。子句处理完，栈弹出。处理字句的过程中，遇到子查询就保存当前子查询的信息，判断与其父查询的关系，最终形成树形结构； 遇到字段或者条件处理则记录当前的字段和条件信息、组成Block，嵌套调用。
关键点解析：
1、遇到TOK_TAB或TOK_TABREF则判断出当前操作的表
2、压栈判断是否是join，判断join条件
3、定义数据结构Block,遇到在where\select\join时获得其下相应的字段和条件，组成Block
4、定义数据结构ColLine,遇到TOK_SUBQUERY保存当前的子查询信息，供父查询使用
5、定义数据结构ColLine,遇到TOK_UNION结束时，合并并截断当前的列信息
6、遇到select 或者未明确指出的字段，查询元数据进行辅助分析
7、解析结果进行相关校验
————————————————
版权声明：本文为CSDN博主「thomas0yang」的原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接及本声明。
原文链接：https://blog.csdn.net/thomas0yang/article/details/49449723
==================================
HiveSqlBloodFigure
如果你喜欢这个项目，那就点击一下右上方的【Star】以及【Fork】，支持一下我，让我有动力持续更新！
项目介绍
在数据仓库建设中，经常会使用到数据血缘追中方面的功能，本项目实现了对hql集合进行静态分析，获取hql对应的血缘图（表血缘 + 字段血缘）
项目升级内容
删除hive-exec与hadoop-common的maven依赖，使得项目更加的轻量级。
重构项目代码，优化解析，修复无字段血缘时，不能获取表血缘的BUG。
规范化接口输入输出，血缘图均为自定义实体，方便进行JSON序列化。
新增接口层，方便Spring的注入，同时也提供静态方式调用。
下个版本升级内容
引入JDBC获取元数据信息，使血缘图更加丰富，同时可以解决Sql中有select * 的问题。
测试用例
请关注test下的HiveBloodEngineTest与HiveSqlBloodFactoryTest。
接口地址
接口类：HiveBloodEngine，实现类：HiveBloodEngineImpl。（spring接入）
工具类：HiveSqlBloodFactory。（静态调用）
使用说明
运行：
结果：
表血缘：
字段血缘：
1 研究背景
随着企业信息化和业务的发展，数据资产日益庞大，数据仓库构建越来越复杂，在数仓构建的过程中，常遇到数据溯源困难，数据模型修改导致业务分析困难等难题，此类问题主要是由于数据血缘分析不足造成的，只有强化血缘关系，才能帮助企业更好的发挥数据价值。
SQL血缘关系是数据仓库模型构建的核心依赖。通过对SQL语句进行梳理与解析，得到各个业务层表之间依赖关系和属性依赖关系，并进行可视化展示，形成数据表和属性血缘层次关系图，充分展示了原始字段数据与数据模型的映射关系。拥有良好的SQL血缘关系系统，不仅有利于数据分析师对业务场景的梳理，还极大帮助对其数仓分层的构建，同时对企业数据质量控制方面起到很好的朔源作用，对构造数据链路图，监控数据变化起到很好的辅助作用。
市场存在一系列血缘关系解析工具，如Druids，但由于只支持对mysql语句的解析，且解析力度不够，不支持复杂的sql逻辑等问题，导致无法在企业中得到广泛使用。同样的hive自身的血缘解析往往在sql执行之后，才可得到解析结果，如果sql执行比较耗时，导致血缘关系无法快速展现，同时造成没有办法提前进行元数据安全和权限认证等问题，在企业真实应用中有一定的限制。
本文结合公司自身业务，研究Hive血缘关系解析源码，并进行优化，首先简化SQL语句剪枝和对包含CTE别名数据表的识别与剔除，降低SQL解析的复杂性，提高血缘解析性能；其次，提供元数据信息服务模块，既保证元信息的完整性，又提供安全的数据表权限认证，维护数据表的操作权限，保证操作的安全性；然后,将postExecuteHook前置，即在SQL执行物理优化前即可获得SQL 的血缘关系，极大提高了获取SQL血缘关系的效率，后续将对这些优化策略进行逐一展开。
————————————————
版权声明：本文为CSDN博主「搜狐技术产品小编2023」的原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接及本声明。
原文链接：https://blog.csdn.net/SOHU_TECH/article/details/110605919
2 相关技术介绍
2.1 SQL血缘关系介绍
在数据仓库构建的过程中，SQL血缘关系体现了各个数据表以及相关属性的依赖关系。SQL血缘关系即是对业务流程涉及到的模型表进行梳理，它包含了集群血缘关系、系统血缘关系、表级血缘关系和字段血缘关系，其指向数据的上游来源，向上游追根溯源。通过简单的SQL语句展现各个表之间的数据关系。
————————————————
版权声明：本文为CSDN博主「搜狐技术产品小编2023」的原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接及本声明。
原文链接：https://blog.csdn.net/SOHU_TECH/article/details/110605919
Apache Atlas
Apache Calcite(可以用)
Trino
ANTLR(可以用)
Antlr4(可以用)
Hive ASTNode(可以用)
Druid
hive hook(需要执行SQL)
https://github.com/webgjc/sql-parser/blob/master/src/main/java/cn/ganjiacheng/hive/HiveSqlFieldLineageParser.java
http://ganjiacheng.cn/article/2020/article_14_%E5%9F%BA%E4%BA%8Eantlr4%E5%AE%9E%E7%8E%B0HQL%E7%9A%84%E8%A7%A3%E6%9E%90-%E8%A1%A8%E8%A1%80%E7%BC%98%E5%92%8C%E5%AD%97%E6%AE%B5%E8%A1%80%E7%BC%98/
https://blog.csdn.net/thomas0yang/article/details/49449723
https://download.csdn.net/download/thomas0yang/9354943
https://download.csdn.net/download/thomas0yang/9369949
http://tech.meituan.com/hive-sql-to-mapreduce.html
http://www.cnblogs.com/drawwindows/p/4595771.html
https://cwiki.apache.org/confluence/display/Hive/LanguageManual
————————————————
版权声明：本文为CSDN博主「thomas0yang」的原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接及本声明。
原文链接：https://blog.csdn.net/thomas0yang/article/details/49449723
https://download.csdn.net/download/xl_1803/75865628?utm_medium=distribute.pc_relevant_download.none-task-download-2defaultBlogCommendFromBaiduRate-1-75865628-download-9369949.257%5Ev11%5Epc_dl_relevant_income_base1&depth_1-utm_source=distribute.pc_relevant_download.none-task-download-2defaultBlogCommendFromBaiduRate-1-75865628-download-9369949.257%5Ev11%5Epc_dl_relevant_income_base1&spm=1003.2020.3001.6616.1
https://download.csdn.net/download/weixin_42131618/16734276?utm_medium=distribute.pc_relevant_download.none-task-download-2defaultBlogCommendFromBaiduRate-12-16734276-download-9369949.257%5Ev11%5Epc_dl_relevant_income_base1&depth_1-utm_source=distribute.pc_relevant_download.none-task-download-2defaultBlogCommendFromBaiduRate-12-16734276-download-9369949.257%5Ev11%5Epc_dl_relevant_income_base1&spm=1003.2020.3001.6616.13
https://github.com/JunNan-X/HiveSqlBloodFigure
https://github.com/lihuigang/dp_dw_lineage
https://blog.csdn.net/qq_44831907/article/details/123033137
https://github.com/webgjc/sql-parser
http://ganjiacheng.cn/article/2020/article_6_%E5%9F%BA%E4%BA%8Eantlr4%E5%AE%9E%E7%8E%B0HQL%E7%9A%84%E8%A7%A3%E6%9E%90-%E5%85%83%E6%95%B0%E6%8D%AE/
http://ganjiacheng.cn/article/2020/article_14_%E5%9F%BA%E4%BA%8Eantlr4%E5%AE%9E%E7%8E%B0HQL%E7%9A%84%E8%A7%A3%E6%9E%90-%E8%A1%A8%E8%A1%80%E7%BC%98%E5%92%8C%E5%AD%97%E6%AE%B5%E8%A1%80%E7%BC%98/
http://ganjiacheng.cn/article/2020/article_12_%E5%9F%BA%E4%BA%8Eantlr4%E5%AE%9E%E7%8E%B0HQL%E7%9A%84%E8%A7%A3%E6%9E%90-%E6%A0%BC%E5%BC%8F%E5%8C%96/
https://gitee.com/Kingkazuma111/sql-parser-lineage?_from=gitee_search
https://gitee.com/hassan1314/flink-sql-lineage?_from=gitee_search
https://blog.csdn.net/SOHU_TECH/article/details/110605919
https://github.com/reata/sqllineage
https://reata.github.io/blog/sqllineage-a-sql-lineage-analysis-tool/
跨分区扫描
package com.atguigu.dga.governance.assess.assessor.calc;
import com.alibaba.fastjson.JSON;
import com.alibaba.fastjson.JSONObject;
import com.atguigu.dga.governance.assess.Assessor;
import com.atguigu.dga.governance.bean.AssessParam;
import com.atguigu.dga.governance.bean.GovernanceAssessDetail;
import com.atguigu.dga.util.SqlParser;
import com.atguigu.dga.meta.bean.TableMetaInfo;
import com.google.common.collect.Sets;
import lombok.*;
import org.apache.hadoop.hive.ql.lib.Dispatcher;
import org.apache.hadoop.hive.ql.lib.Node;
import org.apache.hadoop.hive.ql.parse.ASTNode;
import org.apache.hadoop.hive.ql.parse.BaseSemanticAnalyzer;
import org.apache.hadoop.hive.ql.parse.HiveParser;
import org.apache.hadoop.hive.ql.parse.SemanticException;
import org.springframework.stereotype.Component;
import java.math.BigDecimal;
import java.util.*;
@Component("MULTI_PARTITION")
public class MultiPartitionAssessor extends Assessor {
@Override
protected void checkProblem(GovernanceAssessDetail governanceAssessDetail, AssessParam assessParam) throws Exception {
if (assessParam.getTableMetaInfo().getTableMetaInfoExtra().getDwLevel().equals("ODS") || assessParam.getTDsTaskDefinition() == null) { //ods层没有sql处理
return;
}
//跨分区扫描
//提取sql 进行
if (assessParam.getTableMetaInfo().getTableName().equals("ads_order_to_pay_interval_avg")) {
System.out.println(111);
}
String sql = assessParam.getTDsTaskDefinition().getSql();
governanceAssessDetail.setAssessComment(sql);
Map<String, TableMetaInfo> tableMetaInfoMap = assessParam.getTableMetaInfoMap();
CheckMultiPartitionScanDispatcher dispatcher = new CheckMultiPartitionScanDispatcher();
dispatcher.setTableMetaInfoMap(tableMetaInfoMap);
dispatcher.setDefaultSchemaName(assessParam.getTableMetaInfo().getSchemaName());
SqlParser.parse(dispatcher, sql);
Map<String, Set<String>> tableFilterFieldMap = new HashMap<>();
Map<String, Set<String>> tableRangeFilterFieldMap = new HashMap<>();
//把条件列表 整理为 表-被过滤字段 的结构
List<CheckMultiPartitionScanDispatcher.WhereCondition> whereConditionList = dispatcher.getWhereConditionList();
for (CheckMultiPartitionScanDispatcher.WhereCondition whereCondition : whereConditionList) {
List<CheckMultiPartitionScanDispatcher.OriginTableField> tableFieldList = whereCondition.getTableFieldList();
for (CheckMultiPartitionScanDispatcher.OriginTableField originTableField : tableFieldList) {
Set<String> tableFilterFieldSet = tableFilterFieldMap.get(originTableField.getOriginTable());
if (tableFilterFieldSet == null) {
tableFilterFieldSet = new HashSet<>();
tableFilterFieldMap.put(originTableField.getOriginTable(), tableFilterFieldSet);
}
tableFilterFieldSet.add(originTableField.getField());
if (!whereCondition.operator.equals("=")) {
Set<String> tableFilterFilterFieldSet = tableRangeFilterFieldMap.get(originTableField.getOriginTable());
if (tableFilterFilterFieldSet == null) {
tableFilterFilterFieldSet = new HashSet<>();
tableRangeFilterFieldMap.put(originTableField.getOriginTable(), tableFilterFieldSet);
}
tableFilterFilterFieldSet.add(originTableField.getField());
}
}
}
StringBuilder assessProblem = new StringBuilder();
// 获得所有引用表的清单
Map<String, List<CheckMultiPartitionScanDispatcher.CurTableField>> refTableFieldMap = dispatcher.getRefTableFieldMap();
for (String refTableName : refTableFieldMap.keySet()) {
//获得元数据的分区字段
TableMetaInfo tableMetaInfo = tableMetaInfoMap.get(refTableName);
Set<String> tableFilterFieldSet = tableFilterFieldMap.get(refTableName);
Set<String> tableFilterRangeFieldSet = tableRangeFilterFieldMap.get(refTableName);
//检查每个分区字段 1 是否被过滤 2 是否被范围查询
String partitionColNameJson = tableMetaInfo.getPartitionColNameJson();
List<JSONObject> partitionJsonObjList = JSON.parseArray(partitionColNameJson, JSONObject.class);
for (JSONObject partitionJsonObj : partitionJsonObjList) {
String partitionName = partitionJsonObj.getString("name");
if (tableFilterFieldSet == null || !tableFilterFieldSet.contains(partitionName)) {
assessProblem.append("引用表:" + refTableName + "中的分区字段" + partitionName + "未参与过滤 ;");
}
if (tableFilterRangeFieldSet != null && tableFilterRangeFieldSet.contains(partitionName)) {
assessProblem.append("引用表:" + refTableName + "中的分区字段" + partitionName + "涉及多分区扫描 ;");
}
}
}
if (assessProblem.length() > 0) {
governanceAssessDetail.setAssessScore(BigDecimal.ZERO);
governanceAssessDetail.setAssessProblem(assessProblem.toString());
governanceAssessDetail.setAssessComment(JSON.toJSONString(dispatcher.refTableFieldMap.keySet()) + "||" + sql);
}
}
//节点处理器 会经过sql所有节点处理环节，每经过一个节点执行dispatch方法
public class CheckMultiPartitionScanDispatcher implements Dispatcher {
//检查策略
// 1 获得比较条件的语句
// 2 查看比较字段是否为分区字段
// 3 如果比较符号为 >= <= < > <> in 则是为多分区
// 4 如果比较符号为 = 则获得比较的值 如果同一个分区字段 有多个值 则视为多分区
@Setter
Map<String, TableMetaInfo> tableMetaInfoMap = new HashMap<>();
@Getter
Map<String, List<CurTableField>> subqueryTableFiedlMap = new HashMap<>(); //表<表名,<字段名,Set<原始字段>>
@Getter
Map<String, List<CurTableField>> insertTableFieldMap = new HashMap<>(); //表<表名,<字段名,Set<原始字段>>
@Getter
Map<String, List<CurTableField>> refTableFieldMap = new HashMap<>();
@Getter
List<WhereCondition> whereConditionList = new ArrayList<>();
@Setter
String defaultSchemaName = null;
Set<String> operators = Sets.newHashSet("=", ">", "<", ">=", "<=", "<>", "like"); // in / not in 属于函数计算
@Override
public Object dispatch(Node nd, Stack<Node> stack, Object... nodeOutputs) throws SemanticException {
//检查该节点的处理内容
ASTNode queryNode = (ASTNode) nd;
//分析查询
if (queryNode.getType() == HiveParser.TOK_QUERY) {
//System.out.println("astNode = " + queryNode.getText());
Map<String, List<CurTableField>> curQueryTableFieldMap = new HashMap<>();
Map<String, String> aliasMap = new HashMap<>();
for (Node childNode : queryNode.getChildren()) {
ASTNode childAstNode = (ASTNode) childNode;
if (childAstNode.getType() == HiveParser.TOK_FROM) {
loadTablesFromNodeRec(childAstNode, curQueryTableFieldMap, aliasMap);
} else if (childAstNode.getType() == HiveParser.TOK_INSERT) {
for (Node insertChildNode : childAstNode.getChildren()) {
ASTNode insertChildAstNode = (ASTNode) insertChildNode;
if (insertChildAstNode.getType() == HiveParser.TOK_WHERE) {
loadConditionFromNodeRec(insertChildAstNode, whereConditionList, curQueryTableFieldMap, aliasMap);
} else if (insertChildAstNode.getType() == HiveParser.TOK_SELECT || insertChildAstNode.getType() == HiveParser.TOK_SELECTDI) {
// 如果有子查询 //把查询字段写入缓存 // 没有子查询作为最终输出字段
List<CurTableField> tableFieldOutputList = getTableFieldOutput(insertChildAstNode, curQueryTableFieldMap, aliasMap);
//向上追溯子查询的别名
ASTNode subqueryNode = (ASTNode) queryNode.getAncestor(HiveParser.TOK_SUBQUERY);
//保存到子查询
if (subqueryNode != null) {
cacheSubqueryTableFieldMap(subqueryNode, tableFieldOutputList);
}
ASTNode insertTableNode = (ASTNode) childAstNode.getFirstChildWithType(HiveParser.TOK_DESTINATION).getChild(0);
if (insertTableNode.getType() == HiveParser.TOK_TAB) { //需要做sql最终输出
cacheInsertTableField(insertTableNode, tableFieldOutputList);
}
}
}
}
}
}
return null;
}
private void loadTableFiledByTableName(String tableName, Map<String, List<CurTableField>> curTableFieldMap, Map<String, String> aliasMap) {
String tableWithSchema = tableName;
if (tableName.indexOf(".") < 0) {
tableWithSchema = defaultSchemaName + "." + tableName;
}
TableMetaInfo tableMetaInfo = tableMetaInfoMap.get(tableWithSchema);
if (tableMetaInfo != null) { //是 真实表
List<CurTableField> curFieldsList = new ArrayList<>();
//加载普通字段
String colNameJson = tableMetaInfo.getColNameJson();
List<JSONObject> colJsonObjectList = JSON.parseArray(colNameJson, JSONObject.class);
for (JSONObject colJsonObject : colJsonObjectList) {
CurTableField curTableField = new CurTableField();
OriginTableField originTableField = new OriginTableField();
originTableField.setField(colJsonObject.getString("name"));
originTableField.setPartition(false);
originTableField.setOriginTable(tableMetaInfo.getSchemaName() + "." + tableMetaInfo.getTableName());
if (colJsonObject.getString("type").indexOf("struct") > 0) {
Set<String> structFieldSet = getStructFieldSet(colJsonObject.getString("type"));
originTableField.setSubFieldSet(structFieldSet);
}
curTableField.getOriginTableFieldList().add(originTableField);
curTableField.setCurFieldName(colJsonObject.getString("name"));
curFieldsList.add(curTableField);
}
//加载分区字段
String partitionColNameJson = tableMetaInfo.getPartitionColNameJson();
List<JSONObject> partitionJsonObjectList = JSON.parseArray(partitionColNameJson, JSONObject.class);
for (JSONObject partitionColJsonObject : partitionJsonObjectList) {
CurTableField curTableField = new CurTableField();
OriginTableField originTableField = new OriginTableField();
originTableField.setField(partitionColJsonObject.getString("name"));
originTableField.setPartition(true);
originTableField.setOriginTable(tableMetaInfo.getSchemaName() + "." + tableMetaInfo.getTableName());
curTableField.getOriginTableFieldList().add(originTableField);
curTableField.setCurFieldName(partitionColJsonObject.getString("name"));
curFieldsList.add(curTableField);
}
curTableFieldMap.put(tableName, curFieldsList);
refTableFieldMap.put(tableName, curFieldsList);
String tableWithoutSchema = tableWithSchema.substring(tableWithSchema.indexOf(".") + 1);
aliasMap.put(tableWithoutSchema, tableWithSchema); //把不带库名的表名 作为别名的一种
} else { //不是真实表 从缓存中提取
List<CurTableField> subqueryFieldsList = subqueryTableFiedlMap.get(tableName);
if (subqueryFieldsList == null) {
throw new RuntimeException("未识别对应表: " + tableName);
}
curTableFieldMap.put(tableName, subqueryFieldsList);
}
}
// 递归查找某个节点下的引用的表，并进行加载
public void loadTablesFromNodeRec(ASTNode astNode, Map<String, List<CurTableField>> tableFieldMap, Map<String, String> aliasMap) {
if (astNode.getType() == HiveParser.TOK_TABREF) {
ASTNode tabTree = (ASTNode) astNode.getChild(0);
String tableName = null;
if (tabTree.getChildCount() == 1) {
tableName = BaseSemanticAnalyzer.getUnescapedName((ASTNode) tabTree.getChild(0));
} else {
tableName = BaseSemanticAnalyzer.getUnescapedName((ASTNode) tabTree.getChild(0)) + "." + tabTree.getChild(1); //自动拼接表名
}
//根据表名和补充元数据
loadTableFiledByTableName(tableName, tableFieldMap, aliasMap);
//涉及别名
if (astNode.getChildren().size() == 2) {
ASTNode aliasNode = (ASTNode) astNode.getChild(1);
aliasMap.put(aliasNode.getText(), tableName);
}
} else if (astNode.getType() == HiveParser.TOK_SUBQUERY) {
String aliasName = astNode.getFirstChildWithType(HiveParser.Identifier).getText();
loadTableFiledByTableName(aliasName, tableFieldMap, aliasMap);
} else if (astNode.getChildren() != null && astNode.getChildren().size() > 0) {
for (Node childNode : astNode.getChildren()) {
ASTNode childAstNode = (ASTNode) childNode;
loadTablesFromNodeRec(childAstNode, tableFieldMap, aliasMap);
}
}
}
//递归检查并收集条件表达式
public void loadConditionFromNodeRec(ASTNode node, List<WhereCondition> whereConditionList, Map<String, List<CurTableField>> queryTableFieldMap, Map<String, String> aliasMap) {
if (operators.contains(node.getText())
|| (node.getType() == HiveParser.TOK_FUNCTION && node.getChild(0).getText().equals("in"))) {
WhereCondition whereCondition = new WhereCondition();
if (node.getType() == HiveParser.TOK_FUNCTION && node.getChild(0).getText().equals("in")) {
if (node.getParent().getText().equals("not")) {
whereCondition.setOperator("nin");
} else {
whereCondition.setOperator("in");
}
} else {
whereCondition.setOperator(node.getText());
}
ArrayList<Node> children = node.getChildren();
for (Node child : children) {
ASTNode operatorChildNode = (ASTNode) child;
if (operatorChildNode.getType() == HiveParser.DOT) { //带表名的字段名
ASTNode prefixNode = (ASTNode) operatorChildNode.getChild(0).getChild(0);
ASTNode fieldNode = (ASTNode) operatorChildNode.getChild(1);
getWhereField(whereCondition, prefixNode.getText(), fieldNode.getText(), queryTableFieldMap, aliasMap);
whereConditionList.add(whereCondition);
} else if (operatorChildNode.getType() == HiveParser.TOK_TABLE_OR_COL) { //不带表名的字段名
ASTNode fieldNode = (ASTNode) operatorChildNode.getChild(0);
getWhereField(whereCondition, null, fieldNode.getText(), queryTableFieldMap, aliasMap);
whereConditionList.add(whereCondition);
}
}
} else {
if (node.getChildren() != null) {
for (Node nd : node.getChildren()) {
ASTNode nodeChild = (ASTNode) nd;
loadConditionFromNodeRec(nodeChild, whereConditionList, queryTableFieldMap, aliasMap);
}
}
}
}
private String getInputTableName(Stack<Node> stack) {
ASTNode globalQueryNode = (ASTNode) stack.firstElement();
ASTNode insertNode = (ASTNode) globalQueryNode.getFirstChildWithType(HiveParser.TOK_INSERT);
ASTNode tableNode = (ASTNode) insertNode.getChild(0).getChild(0); //TOK_DESINATION->TOK_TAB
if (tableNode.getChildCount() == 1) {
return defaultSchemaName + "." + BaseSemanticAnalyzer.getUnescapedName((ASTNode) tableNode.getChild(0)); //不带库名 补库名
} else {
return BaseSemanticAnalyzer.getUnescapedName((ASTNode) tableNode.getChild(0)) + "." + tableNode.getChild(1); //带库名
}
}
public List<CurTableField> getTableFieldOutput(ASTNode selectNode, Map<String, List<CurTableField>> curQueryTableFieldMap, Map<String, String> aliasMap) {
List<CurTableField> outputTableFieldList = new ArrayList<>();
for (Node selectXPRNode : selectNode.getChildren()) {
ASTNode selectXPRAstNode = (ASTNode) selectXPRNode;
// 如果是隐性 或者select * //返回子查询
if (selectXPRAstNode.getChild(0).getType() == HiveParser.TOK_SETCOLREF || selectXPRAstNode.getChild(0).getType() == HiveParser.TOK_ALLCOLREF) {
for (List<CurTableField> curTableFieldList : curQueryTableFieldMap.values()) {
outputTableFieldList.addAll(curTableFieldList);
}
return outputTableFieldList;
} else {
//逐个取节点下的select 的字段
CurTableField curTableField = new CurTableField();
loadCurTableFieldFromNodeRec((ASTNode) selectXPRAstNode, curTableField, curQueryTableFieldMap, aliasMap);
if (selectXPRNode.getChildren().size() == 2) { //说明为字段起了别名
ASTNode aliasNode = (ASTNode) ((ASTNode) selectXPRNode).getChild(1);
curTableField.setCurFieldName(aliasNode.getText());
}
outputTableFieldList.add(curTableField);
}
}
return outputTableFieldList;
}
//把对象保存到子查询缓存中
private void cacheSubqueryTableFieldMap(ASTNode subqueryNode, List<CurTableField> curTableFieldList) {
ASTNode subqueryAliasNode = (ASTNode) subqueryNode.getFirstChildWithType(HiveParser.Identifier);
String aliasName = subqueryAliasNode.getText();
List<CurTableField> existsTableFieldList = subqueryTableFiedlMap.get(aliasName);
if (existsTableFieldList != null) { //说明已经有查询声明为改别名了 ，主要原因是因为union造成的， 这种情况要按照顺序把每个字段的原始字段信息追加
for (int i = 0; i < existsTableFieldList.size(); i++) {
CurTableField existsTableField = existsTableFieldList.get(i);
CurTableField curTableField = curTableFieldList.get(i);
existsTableField.getOriginTableFieldList().addAll(curTableField.getOriginTableFieldList());
}
} else { //把子查询加入缓存
subqueryTableFiedlMap.put(aliasName, curTableFieldList);
}
}
private void cacheInsertTableField(ASTNode outputTableNode, List<CurTableField> curTableFieldList) {
String outputTableName = null;
if (outputTableNode.getChildCount() == 2) {
outputTableName = outputTableNode.getChild(0).getChild(0).getText() + "." + outputTableNode.getChild(1).getText();
} else {
outputTableName = defaultSchemaName + "." + outputTableNode.getChild(0).getChild(0).getText();
}
insertTableFieldMap.put(outputTableName, curTableFieldList);
}
// 利用递归获得当前节点下的字段信息
public void loadCurTableFieldFromNodeRec(ASTNode recNode, CurTableField curTableField, Map<String, List<CurTableField>> curQueryTableFieldMap, Map<String, String> aliasMap) {
if (recNode.getChildren() != null) {
for (Node subNode : recNode.getChildren()) {
ASTNode subAstNode = (ASTNode) subNode;
if (subAstNode.getType() == HiveParser.DOT) { //带表的字段
ASTNode prefixNode = (ASTNode) subAstNode.getChild(0).getChild(0);
ASTNode fieldNode = (ASTNode) subAstNode.getChild(1);
String prefix = prefixNode.getText();
List<OriginTableField> originTableFieldList = getOriginFieldByFieldName(prefix, fieldNode.getText(), curQueryTableFieldMap, aliasMap);
curTableField.getOriginTableFieldList().addAll(originTableFieldList);
curTableField.setCurFieldName(fieldNode.getText());
} else if (subAstNode.getType() == HiveParser.TOK_TABLE_OR_COL) {
ASTNode fieldNode = (ASTNode) subAstNode.getChild(0);
//不带表的字段要从
List<OriginTableField> originTableFieldList = getOriginTableFieldList(curQueryTableFieldMap, fieldNode.getText());
curTableField.getOriginTableFieldList().addAll(originTableFieldList);
curTableField.setCurFieldName(fieldNode.getText());
} else {
loadCurTableFieldFromNodeRec(subAstNode, curTableField, curQueryTableFieldMap, aliasMap);
}
}
}
}
//前缀
public void getWhereField(WhereCondition whereCondition, String prefix, String fieldName, Map<String, List<CurTableField>> queryTableFieldMap, Map<String, String> aliasMap) {
List<OriginTableField> originTableFieldList = null;
if (prefix == null) {
originTableFieldList = getOriginTableFieldList(queryTableFieldMap, fieldName);
} else { //有前缀
originTableFieldList = getOriginFieldByFieldName(prefix, fieldName, queryTableFieldMap, aliasMap); //把前缀作为表查询
}
if (originTableFieldList == null) {
throw new RuntimeException("无法识别的字段名：" + fieldName);
}
whereCondition.setTableFieldList(originTableFieldList);
}
// private List getOriginTableFieldList(List curTableField, String fieldName) {
//
// }
// private List getCurTableFieldListByPrefix(String prefix, Map> queryTableFieldMap, Map aliasMap) {
// List curFieldList = queryTableFieldMap.get(prefix); //把前缀作为表查询
// if (curFieldList == null) {//未查询出 尝试换为字段查询
// String tableName = aliasMap.get(prefix);
// if (tableName != null) {
// curFieldList = queryTableFieldMap.get(tableName);
// }
// }
//
// return curFieldList;
// }
//根据前缀和字段名 获得从表结构中获得 原始字段列表
private List<OriginTableField> getOriginFieldByFieldName(String prefix, String fieldName, Map<String, List<CurTableField>> queryTableFieldMap, Map<String, String> aliasMap) {
List<CurTableField> curFieldList = queryTableFieldMap.get(prefix); //把前缀作为表查询
if (curFieldList == null) {//未查询出 尝试换为字段查询
String tableName = aliasMap.get(prefix);
if (tableName != null) {
curFieldList = queryTableFieldMap.get(tableName);
}
}
if (curFieldList == null) {
return getOriginTableFieldList(queryTableFieldMap, prefix); //前缀有可能是结构体字段名
}
if (curFieldList == null) {
throw new RuntimeException("不明确的表前缀：" + prefix);
}
return getOriginTableFieldList(fieldName, curFieldList);
}
private List<OriginTableField> getOriginTableFieldList(String fieldName, List<CurTableField> curFieldList) {
for (CurTableField tableField : curFieldList) {
if (tableField.getCurFieldName().equals(fieldName)) {
return tableField.getOriginTableFieldList();
}
}
return new ArrayList<>(); // 一般是常量字段 比 lateral产生的常量字段 不是从表中计算而来
}
private List<OriginTableField> getOriginTableFieldList(Map<String, List<CurTableField>> queryTableFieldMap, String fieldName) {
List<OriginTableField> originTableFieldList = null;
for (Map.Entry entry : queryTableFieldMap.entrySet()) {
List<CurTableField> curTableFieldList = (List<CurTableField>) entry.getValue();
List<OriginTableField> matchedOriginTableFieldList = getOriginTableFieldList(fieldName, curTableFieldList);
if (originTableFieldList != null && originTableFieldList.size() > 0 && matchedOriginTableFieldList.size() > 0) {
throw new RuntimeException("归属不明确的字段：" + fieldName);
} else {
originTableFieldList = matchedOriginTableFieldList;
}
}
return originTableFieldList;
}
//拆分子字段
//structHive 解析语法树
package com.atguigu.dga.util;
import org.apache.hadoop.hive.ql.lib.DefaultGraphWalker;
import org.apache.hadoop.hive.ql.lib.Dispatcher;
import org.apache.hadoop.hive.ql.lib.GraphWalker;
import org.apache.hadoop.hive.ql.lib.Node;
import org.apache.hadoop.hive.ql.parse.*;
import java.util.Collections;
import java.util.Stack;
public class SqlParser {
//1 把sql转换为语法树 有工具 完成 在hive依赖中就已经提供了
//
//2 提供了一个 遍历器 后序遍历
//
//3 自定义一个节点处理器
//4 把处理器放到遍历器中
//5 让遍历器遍历语法树
public static void parse(Dispatcher dispatcher,String sql ) throws Exception {
//1 把sql转换为语法树 有工具 完成 在hive依赖中就已经提供了
ParseDriver parseDriver = new ParseDriver(); //用于把sql转为语法树
ASTNode astNode = parseDriver.parse(sql);
//2 提供了一个 遍历器 后序遍历
while(astNode.getType()!= HiveParser.TOK_QUERY){ //循环遍历直到找到第一个query节点 ，循环退出 ，用query节点作为根节点。
astNode=(ASTNode)astNode.getChild(0);
}
//3 自定义一个节点处理器 //根据不同的需求在方法外部定义 ，定义好后传递
//4 把处理器放到遍历器中
GraphWalker graphWalker=new DefaultGraphWalker(dispatcher);
//5 让遍历器遍历语法树
graphWalker.startWalking(Collections.singletonList(astNode),null);
}
public static void main(String[] args) throws Exception {
String sql = " select a,b,c from gmall.user_info u where u.id='123' and dt='123123' ";
// 自定义一个节点处理器
TestDispatcher testDispatcher = new TestDispatcher();
SqlParser.parse( testDispatcher,sql);
}
static class TestDispatcher implements Dispatcher {
//每到达一个节点要处理的事项
@Override
public Object dispatch(Node nd, Stack<Node> stack, Object... nodeOutputs) throws SemanticException {
ASTNode astNode = (ASTNode) nd;
System.out.println("type"+astNode.getType() +" || token:"+astNode.getToken().getText());
return null;
}
}
}
druid 解析SQL语法树
import com.alibaba.druid.DbType;
import com.alibaba.druid.sql.ast.*;
import com.alibaba.druid.sql.ast.expr.SQLAggregateExpr;
import com.alibaba.druid.sql.ast.expr.SQLMethodInvokeExpr;
import com.alibaba.druid.sql.ast.statement.SQLSelectItem;
import com.alibaba.druid.sql.ast.statement.SQLSelectStatement;
import com.alibaba.druid.sql.dialect.hive.ast.HiveInsertStatement;
import com.alibaba.druid.sql.dialect.hive.parser.HiveStatementParser;
import com.alibaba.druid.sql.parser.SQLParserUtils;
import com.alibaba.druid.sql.parser.SQLStatementParser;
import java.util.List;
import java.util.Map;
import java.util.Set;
import com.alibaba.druid.sql.SQLUtils;
import com.alibaba.druid.sql.ast.SQLExpr;
import com.alibaba.druid.sql.ast.SQLName;
import com.alibaba.druid.sql.ast.SQLStatement;
import com.alibaba.druid.sql.ast.expr.SQLBinaryOpExpr;
import com.alibaba.druid.sql.parser.SQLStatementParser;
import com.alibaba.druid.sql.parser.ParserException;
import com.alibaba.druid.sql.parser.SQLParserFeature;
import com.alibaba.druid.sql.repository.SchemaRepository;
import com.alibaba.druid.sql.visitor.SchemaStatVisitor;
import com.alibaba.druid.stat.TableStat;
import com.alibaba.fastjson.JSON;
import com.alibaba.druid.sql.ast.SQLExpr;
import com.alibaba.druid.sql.ast.SQLStatement;
import com.alibaba.druid.sql.ast.statement.SQLSelect;
import com.alibaba.druid.sql.ast.statement.SQLSelectItem;
import com.alibaba.druid.sql.SQLUtils;
import com.alibaba.druid.sql.ast.SQLStatement;
import com.alibaba.druid.sql.dialect.mysql.visitor.MySqlSchemaStatVisitor;
import com.alibaba.druid.sql.dialect.hive.visitor.*;
/**
* Created by 黄凯 on 2023/6/7 0007 14:11
*
* @author 黄凯
* 永远相信美好的事情总会发生.
*/
public class DruidDemo {
/*public static void main(String[] args) {
String sql = " select a,b,c from gmall.user_info u where u.id='123' and dt='123123' ";
// String sql = "with t1 as (select aa(a), b, c, dt as dd\n" +
// " from tt1,\n" +
// " tt2\n" +
// " where tt1.a = tt2.b\n" +
// " and dt = '2023-05-11')\n" +
// "insert\n" +
// "overwrite\n" +
// "table\n" +
// "tt9\n" +
// "select a, b, c\n" +
// "from t1\n" +
// "where dt = date_add('2023-06-08', -4)\n" +
// "union\n" +
// "select a, b, c\n" +
// "from t2\n" +
// "where dt = date_add('2023-06-08', -7)";
// 提取表名
// SQLSelectItem sqlSelectItem = SQLUtils.toSelectItem(sql, DbType.mysql);
// System.out.println("sqlSelectItem = " + sqlSelectItem);
//
// SQLExpr expr = sqlSelectItem.getExpr();
//
// System.out.println(expr.toString());
// String string1 = JSON.toJSONString(expr);
//
//
//
// System.out.println(expr.getAttributesDirect());
//
// SQLCommentHint hint = expr.getHint();
// Map attributes1 = expr.getAttributes();
// List children1 = expr.getChildren();
// Map attributesDirect = expr.getAttributesDirect();
// Object subQuery = expr.getAttribute("subQuery");
// List afterCommentsDirect = expr.getAfterCommentsDirect();
// SQLObject parent = expr.getParent();
//
//
// Map attributes2 = children1.get(0).getAttributes();
// 创建Parser对象
SQLStatementParser parser = SQLParserUtils.createSQLStatementParser(sql, DbType.hive);
SQLStatement statement = parser.parseStatement();
List children2 = statement.getChildren();
Object select = statement.getAttribute("select");
Object query = statement.getAttribute("query");
Object from = statement.getAttribute("from");
List headHintsDirect = statement.getHeadHintsDirect();
Map attributesDirect1 = statement.getAttributesDirect();
Map attributes3 = statement.getAttributes();
List afterCommentsDirect1 = statement.getAfterCommentsDirect();
List beforeCommentsDirect = statement.getBeforeCommentsDirect();
String string = JSON.toJSONString(children2);
for (SQLObject child : children2) {
if (child instanceof SQLSelectItem) {
SQLSelectItem selectItem = (SQLSelectItem) child;
SQLExpr expr2 = selectItem.getExpr();
SQLObject parent = expr2.getParent();
System.out.println("parent = " + parent);
// 处理selectList中的表达式
System.out.println(expr2.toString());
}
}
List children = statement.getChildren();
for (SQLObject child : children) {
System.out.println("child = " + child);
Map attributes = child.getAttributes();
System.out.println("attributes = " + attributes);
}
SQLLimit limit = SQLUtils.getLimit(statement, DbType.mysql);
System.out.println("limit = " + limit);
//强转
SQLSelectStatement statement1 = (SQLSelectStatement) statement;
// HiveInsertStatement statement1 = (HiveInsertStatement) statement;
//对的
SQLSelect select1 = statement1.getSelect();
// SQLSelect select1 = null;
SQLSelect sqlSelect = select1;
List selectItems = sqlSelect.getQueryBlock().getSelectList();
for (SQLSelectItem selectItem : selectItems) {
SQLExpr expr3 = selectItem.getExpr();
// 处理selectList中的表达式
System.out.println(expr3.toString());
}
SQLExpr where = sqlSelect.getQueryBlock().getWhere();
System.out.println("where = " + where);
//对的
/
// 假设sql是要分析的SQL查询语句
// String sql = "SELECT column1, column2 FROM table1";
// 解析SQL语句
List statements = SQLUtils.parseStatements(sql, DbType.hive);
// 创建血缘分析器
HiveSchemaStatVisitor visitor = new HiveSchemaStatVisitor();
for (SQLStatement statement3 : statements) {
statement3.accept(visitor);
}
System.out.println(visitor);
SchemaStatVisitor schemaStatVisitor = new SchemaStatVisitor();
HiveStatementParser parser3 = new HiveStatementParser(sql);
// 使用Parser解析生成AST，这里SQLStatement就是AST
SQLStatement sqlStatement = parser3.parseStatement();
HiveSchemaStatVisitor visitor3 = new HiveSchemaStatVisitor();
sqlStatement.accept(visitor3);
Map tables = visitor3.getTables();
System.out.println(tables.keySet());
System.out.println("使用visitor数据表：" + visitor3.getTables());
System.out.println("使用visitor字段：" + visitor3.getColumns());
System.out.println("使用visitor条件：" + visitor3.getConditions());
System.out.println("使用visitor分组：" + visitor3.getGroupByColumns());
System.out.println("使用visitor排序：" + visitor3.getOrderByColumns());
// 获取血缘关系结果
// Map> tableLineage = visitor.getColumnsLineage();
// 获取血缘关系结果
// Map> tableLineage = visitor.getco
// 创建血缘分析器
// MySqlSchemaStatVisitor visitor = new MySqlSchemaStatVisitor();
// // 获取血缘关系结果
// Map> tableLineage = visitor.getColumnsLineage();
// 假设statement是SQLStatement对象
// if (select1 instanceof SQLSelect) {
// SQLSelect sqlSelect = (SQLSelect) statement;
// List selectItems = sqlSelect.getQueryBlock().getSelectList();
// for (SQLSelectItem selectItem : selectItems) {
// SQLExpr expr3 = selectItem.getExpr();
// // 处理selectList中的表达式
// System.out.println(expr3.toString());
// }
// }
// // 获取表名
// String tableName = statement.getTableSource().toString();
//
// // 获取列名
// List selectItems = statement.getSelect().getQueryBlock().getSelectList();
// for (SQLSelectItem item : selectItems) {
// String columnName = item.getExpr().toString();
// // 处理列名
// }
//
// // 获取条件
// SQLExpr where = statement.getSelect().getQueryBlock().getWhere();
// if (where != null) {
// // 处理条件
// }
}*/
/**
* 测试Druid
*
* @param args
*/
public static void main(String[] args) {
String sql = "with t1 as (select aa(a), b, c, dt as dd\n" +
" from tt1,\n" +
" tt2\n" +
" where tt1.a = tt2.b\n" +
" and dt = '2023-05-11')\n" +
"insert\n" +
"overwrite\n" +
"table\n" +
"tt9\n" +
"select a, b, c\n" +
"from t1\n" +
"where dt = date_add('2023-06-08', -4)\n" +
"union\n" +
"select a, b, c\n" +
"from t2\n" +
"where dt = date_add('2023-06-08', -7)";
// 解析SQL语句
List<SQLStatement> statements = SQLUtils.parseStatements(sql, DbType.hive);
String string = SQLUtils.toSQLString(statements, DbType.hive);
System.out.println("string = " + string);
// 创建血缘分析器
HiveSchemaStatVisitor visitor = new HiveSchemaStatVisitor();
for (SQLStatement statement3 : statements) {
statement3.accept(visitor);
}
System.out.println(visitor);
SchemaStatVisitor schemaStatVisitor = new SchemaStatVisitor();
HiveStatementParser parser3 = new HiveStatementParser(sql);
// 使用Parser解析生成AST，这里SQLStatement就是AST
SQLStatement sqlStatement = parser3.parseStatement();
HiveSchemaStatVisitor visitor3 = new HiveSchemaStatVisitor();
sqlStatement.accept(visitor3);
Map<TableStat.Name, TableStat> tables = visitor3.getTables();
System.out.println(tables.keySet());
System.out.println("使用visitor数据表：" + visitor3.getTables());
System.out.println("使用visitor字段：" + visitor3.getColumns());
System.out.println("使用visitor条件：" + visitor3.getConditions());
System.out.println("使用visitor分组：" + visitor3.getGroupByColumns());
System.out.println("使用visitor排序：" + visitor3.getOrderByColumns());
List<SQLAggregateExpr> aggregateFunctions = visitor3.getAggregateFunctions();
List<SQLMethodInvokeExpr> functions = visitor3.getFunctions();
List<SQLName> originalTables = visitor3.getOriginalTables();
List<Object> parameters = visitor3.getParameters();
Set<TableStat.Relationship> relationships = visitor3.getRelationships();
SchemaRepository repository = visitor3.getRepository();
}
}
表级沿袭
列级沿袭
