antlr解析hive语句

hive是使用antlr来解析的

parser要做的事情，是从无结构的字符串里面，解码产生有结构的数据结构（a parser is a function accepting strings as input and returning some structure as output），参考 Parser_combinator wiki

parser分成两种，一种是parser combinator，一种是parser generator，区别可以参考 王垠的文章——对 Parser 的误解

 

1.parser combinator是需要手写parser，a parser combinator is a higher-order function that accepts several parsers as input and returns a new parser as its output，比如Thrift的Parser

https://github.com/apache/thrift/blob/master/compiler/cpp/src/thrift/main.cc

 

2.parser generator是需要你用某种指定的描述语言来表示出语法，然后自动把他们转换成parser的代码，比如Antlr里面的g4语法文件，calcite的ftl语法文件，hue使用的jison以及flex和cup等，缺点是由于代码是生成的，排错比较困难

使用了Antlr的parser有Hive，Presto，Spark SQL

美团点评的文章

https://tech.meituan.com/2014/02/12/hive-sql-to-mapreduce.html

以及hive源码的测试用例

https://github.com/apache/hive/blob/branch-1.1/ql/src/test/org/apache/hadoop/hive/ql/parse/TestHiveDecimalParse.java

hive的g4文件如下

老版本的hive

https://github.com/apache/hive/blob/59d8665cba4fe126df026f334d35e5b9885fc42c/parser/src/java/org/apache/hadoop/hive/ql/parse/HiveParser.g

新版本的hive

https://github.com/apache/hive/blob/master/hplsql/src/main/antlr4/org/apache/hive/hplsql/Hplsql.g4

spark的g4文件如下

https://github.com/apache/spark/blob/master/sql/catalyst/src/main/antlr4/org/apache/spark/sql/catalyst/parser/SqlBase.g4

Presto的g4文件如下

https://github.com/prestodb/presto/blob/master/presto-parser/src/main/antlr4/com/facebook/presto/sql/parser/SqlBase.g4

confluent的kSql的g4文件

https://github.com/confluentinc/ksql/blob/master/ksqldb-parser/src/main/antlr4/io/confluent/ksql/parser/SqlBase.g4

　　

使用了Apache Calcite的parser有Apache Flink，Mybatis，Apache Storm等

参考：Apache Calcite 为什么能这么流行

Flink的ftl文件如下

https://github.com/apache/flink/blob/master/flink-table/flink-sql-parser/src/main/codegen/includes/parserImpls.ftl

Mybatis的mapper模板生成

https://github.com/abel533/Mapper/blob/master/generator/src/main/resources/generator/mapper.ftl

Storm的ftl文件如下

https://github.com/apache/storm/blob/master/sql/storm-sql-core/src/codegen/includes/parserImpls.ftl

 

以及使用了flex和cup的impala，如何使用impala的parser来解析query可以参考另一篇文章：使用Impala parser解析SQL

parser的测试用例

https://github.com/cloudera/Impala/blob/master/fe/src/test/java/com/cloudera/impala/analysis/ParserTest.java

源码

https://github.com/apache/impala/blob/master/fe/src/main/jflex/sql-scanner.flex

 和

https://github.com/apache/impala/blob/master/fe/src/main/cup/sql-parser.cup

impala也用了少量的antlr

https://github.com/apache/impala/blob/master/fe/src/main/java/org/apache/impala/analysis/ToSqlUtils.java

 

还有hue使用的jison，jison是JavaScript语言的语法分析器

https://github.com/cloudera/hue/tree/master/desktop/core/src/desktop/js/parse/jison

 

以hive的Hplsql.g4为例，来解析一句sql

antlr4 Hplsql.g4
javac Hplsql*.java

解析select语句

grun Hplsql r -tokens
Warning: TestRig moved to org.antlr.v4.gui.TestRig; calling automatically
select * from db1.tb1;
[@0,0:5='select',<T_SELECT>,1:0]
[@1,7:7='*',<'*'>,1:7]
[@2,9:12='from',<T_FROM>,1:9]
[@3,14:16='db1',<L_ID>,1:14]
[@4,17:17='.',<'.'>,1:17]
[@5,18:20='tb1',<L_ID>,1:18]
[@6,21:21=';',<';'>,1:21]
[@7,23:22='<EOF>',<EOF>,2:0]
No method for rule r or it has arguments

可以看到打印出token流

解析建表语句

grun Hplsql r -tokens
Warning: TestRig moved to org.antlr.v4.gui.TestRig; calling automatically
CREATE TABLE IF NOT EXISTS db1.tb1 (
  `f1` string,
  `f2` bigint,
  `f3` string,
  `f4` string,
  `f5` string)
partitioned by(ds string)
stored as parquet
TBLPROPERTIES ("parquet.compression"="SNAPPY");
[@0,0:5='CREATE',<T_CREATE>,1:0]
[@1,7:11='TABLE',<T_TABLE>,1:7]
[@2,13:14='IF',<T_IF>,1:13]
[@3,16:18='NOT',<T_NOT>,1:16]
[@4,20:25='EXISTS',<T_EXISTS>,1:20]
[@5,27:29='db1',<L_ID>,1:27]
[@6,30:30='.',<'.'>,1:30]
[@7,31:33='tb1',<L_ID>,1:31]
[@8,35:35='(',<'('>,1:35]
[@9,39:42='`f1`',<L_ID>,2:2]
[@10,44:49='string',<T_STRING>,2:7]
[@11,50:50=',',<','>,2:13]
[@12,54:57='`f2`',<L_ID>,3:2]
[@13,59:64='bigint',<T_BIGINT>,3:7]
[@14,65:65=',',<','>,3:13]
[@15,69:72='`f3`',<L_ID>,4:2]
[@16,74:79='string',<T_STRING>,4:7]
[@17,80:80=',',<','>,4:13]
[@18,84:87='`f4`',<L_ID>,5:2]
[@19,89:94='string',<T_STRING>,5:7]
[@20,95:95=',',<','>,5:13]
[@21,99:102='`f5`',<L_ID>,6:2]
[@22,104:109='string',<T_STRING>,6:7]
[@23,110:110=')',<')'>,6:13]
[@24,112:122='partitioned',<L_ID>,7:0]
[@25,124:125='by',<T_BY>,7:12]
[@26,126:126='(',<'('>,7:14]
[@27,127:128='ds',<L_ID>,7:15]
[@28,130:135='string',<T_STRING>,7:18]
[@29,136:136=')',<')'>,7:24]
[@30,138:143='stored',<T_STORED>,8:0]
[@31,145:146='as',<T_AS>,8:7]
[@32,148:154='parquet',<L_ID>,8:10]
[@33,156:168='TBLPROPERTIES',<L_ID>,9:0]
[@34,170:170='(',<'('>,9:14]
[@35,171:191='"parquet.compression"',<L_ID>,9:15]
[@36,192:192='=',<'='>,9:36]
[@37,193:200='"SNAPPY"',<L_ID>,9:37]
[@38,201:201=')',<')'>,9:45]
[@39,202:202=';',<';'>,9:46]
[@40,204:203='<EOF>',<EOF>,10:0]
No method for rule r or it has arguments

 

上面介绍了antlr如果解析hive语句，而在hive中使用的就是由antlr编译出来的java代码来解析hive语句

接下来介绍如何使用java代码解析hive语句，首先引用依赖

<dependency>
    <groupId>org.apache.hive</groupId>
    <artifactId>hive-exec</artifactId>
    <version>1.1.0-cdh5.16.2</version>
</dependency>

代码

import com.google.common.collect.Lists;
import com.google.common.collect.Maps;
import org.apache.hadoop.hive.conf.HiveConf;
import org.apache.hadoop.hive.metastore.api.FieldSchema;
import org.apache.hadoop.hive.ql.Context;
import org.apache.hadoop.hive.ql.lib.*;
import org.apache.hadoop.hive.ql.parse.*;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import java.io.IOException;
import java.util.List;
import java.util.Map;
import java.util.Stack;

public class MyProcessor implements NodeProcessor {

    private static Logger logger = LoggerFactory.getLogger(MyProcessor.class);
    private static Context context = null;
    private final static String HDFS_SESSION_PATH_KEY = "_hive.hdfs.session.path";
    private final static String LOCAL_SESSION_PATH_KEY = "_hive.local.session.path";

    private static String hdfsTemporaryDirectory(HiveConf hiveConf) {
        return hiveConf.get("hadoop.tmp.dir", "/tmp");
    }


    private static String localTemporaryDirectory() {
        return System.getProperty("java.io.tmpdir", "/tmp");
    }

    static {
        HiveConf hiveConf = new HiveConf();
        if (hiveConf.get(HDFS_SESSION_PATH_KEY) == null) {
            hiveConf.set(HDFS_SESSION_PATH_KEY, hdfsTemporaryDirectory(hiveConf));
        }
        if (hiveConf.get(LOCAL_SESSION_PATH_KEY) == null) {
            hiveConf.set(LOCAL_SESSION_PATH_KEY, localTemporaryDirectory());
        }
        try {
            context = new Context(hiveConf);
        } catch (IOException e) {
            logger.error("Init hive context fail, message: " + e);
        }
    }

    String tableName = "";
    List<FieldSchema> fieldSchemas;

    public void parse(String query) throws ParseException, SemanticException {
        ParseDriver pd = new ParseDriver();
        ASTNode tree = pd.parse(query, context);
        while ((tree.getToken() == null) &amp;&amp; (tree.getChildCount() > 0)) {
            tree = (ASTNode) tree.getChild(0);
        }
        logger.info("start to analyze query: {}, ASTNode: {}", query, tree.dump());
        Map<Rule, NodeProcessor> rules = Maps.newLinkedHashMap();
        Dispatcher disp = new DefaultRuleDispatcher(this, rules, null);
        GraphWalker ogw = new DefaultGraphWalker(disp);
        final List<Node> topNodes = Lists.newArrayList(tree);

        // 遍历
        ogw.startWalking(topNodes, null);
        // 打印
        System.out.println(tableName);
        System.out.println(fieldSchemas);
    }

    @Override
    public Object process(Node nd, Stack<Node> stack, NodeProcessorCtx procCtx, Object... nodeOutputs) throws SemanticException {
        ASTNode pt = (ASTNode) nd;
        switch (pt.getToken().getType()) {
            case org.apache.hadoop.hive.ql.parse.HiveParser.TOK_CREATETABLE:
                for (Node node : pt.getChildren()) {
                    ASTNode createTableChild = (ASTNode) node;
                    if (createTableChild.getToken().getType() == HiveParser.TOK_TABNAME) {
                        tableName = BaseSemanticAnalyzer.getUnescapedName(createTableChild);
                    } else if (createTableChild.getToken().getType() == HiveParser.TOK_TABCOLLIST) {
                        fieldSchemas = BaseSemanticAnalyzer.getColumns(createTableChild, true);
                    }
                }
        }
        return null;
    }
}

测试用例，解析了hive的建表语句

import org.junit.Test;

public class MyProcessorTest {

    @Test
    public void parse() throws Exception{
        String query = "create table my_table(id int,name string)row format delimited fields terminated by '\\t'";
        MyProcessor processor = new MyProcessor();
        processor.parse(query);
    }

}

输出

上面例子中是将hive表名和字段解析出来，其他属性也可以使用类似的方法从语法树中取出