Spark 整合ElasticSearch

Spark 整合ElasticSearch

因为做资料搜索用到了ElasticSearch,最近又了解一下 Spark ML,先来演示一个Spark 读取/写入 ElasticSearch 简单示例。(spark 读取ElasticSearch中数据)

环境:IDEA2016,JDK8,windows10,安装的 ElasticSearch6.3.2 和 spark-2.3.1-bin-hadoop2.7,使用mvn package 将程序打成jar包,采用spark-submit提交给spark执行。

先在ElasticSearch中创建一个索引用来演示。因为是文本数据,因此采用ik分词。可参考:elasticsearch-ik

  • 创建索引:PUT /index_ik_test

  • 设置mapping 及相应的分词器,这里指定 content 字段为 ElasticSearch 的text 类型,并使用ik_max_word 分词模式

    POST index_ik_test/fulltext/_mapping
    {
    "properties": {
    "content":{
    "type": "text",
    "analyzer": "ik_max_word",
    "search_analyzer": "ik_max_word"
    }
    }
    }

  • 存几篇文档到ElasticSearch中

    POST index_ik_test/fulltext/1

  • ik 分词器有两种分词模式:ik_max_wordik_smart。可通过如下方式查看一下这两者的区别:

    GET index_ik_test/_analyze
    {
    "text": ["其中国家投资了500万"],
    "tokenizer": "ik_smart"
    }

    分词结果:其中、国家、投资、了、500万

    GET index_ik_test/_analyze
    {
    "text": ["其中国家投资了500万"],
    "tokenizer": "ik_max_word"
    }

    分词结果:其中、中国、国家、投资、了、500、万

  • 使用GET index_ik_test/_mapping可查看索引的配置信息

    {
    "index_ik_test": {
    "mappings": {
    "fulltext": {
    "properties": {
    "content": {
    "type": "text",
    "analyzer": "ik_max_word"
    }
    }
    }
    }
    }
    }

好,现在ElasticSearch中有数据了,现在看怎么基于Spark读取ElasticSearch中的数据。

IDEA2016中新建一个Maven工程,当然也可以用SpringBoot工程,但是这里的是单纯的Maven Project。

ElasticSearch官方提供了elasticsearch-hadoop来供Spark访问ElasticSearch。具体可参考:官方文档es for spark

官方提供了elasticsearch-hadoopmaven 依赖,这个依赖包括了:ElasticSearch for Hadoop MR、ElasticSearch for Hadoop Hive、ElasticSearch for Hadoop Spark。如果只用到了Spark,也可以只添加ElasticSearch for spark依赖。具体可参考:(这个链接)[https://www.elastic.co/guide/en/elasticsearch/hadoop/current/install.html]

<dependency>
  <groupId>org.elasticsearch</groupId>
  <artifactId>elasticsearch-spark-20_2.10</artifactId>
  <version>6.3.2</version>
</dependency>

创建spark运行上下文时需要spark-sql_2.11依赖,可参考:spark 官方文档quick start

To build the program, we also write a Maven pom.xml file that lists Spark as a dependency. Note that Spark artifacts are tagged with a Scala version.

在本文的示例中,添加了下面3个maven依赖:

<dependency>
  <groupId>org.elasticsearch</groupId>
  <artifactId>elasticsearch-hadoop</artifactId>
  <version>6.3.2</version>
</dependency>
<!-- Spark dependency -->
<dependency>
  <groupId>org.apache.spark</groupId>
  <artifactId>spark-sql_2.11</artifactId>
  <version>2.3.1</version>
</dependency>

<dependency>
  <groupId>com.google.guava</groupId>
  <artifactId>guava</artifactId>
  <version>22.0</version>
</dependency>

下面来直接看示例代码:

向ElasticSearch中写入数据

  • spark配置连接ElasticSearch。可参考:elasticsearch-hadoop-master,我们采用的是:Configure the connector to run in WAN mode

    SparkConf sparkConf = new SparkConf().setAppName("writeEs").setMaster("local[*]").set("es.index.auto.create", "true")
    			.set("es.nodes", "ELASTIC_SEARCH_IP").set("es.port", "9200").set("es.nodes.wan.only", "true");
    
  • 将数据写入到ElasticSearch

    JavaRDD<Map<String, ?>> javaRDD = jsc.parallelize(ImmutableList.of(numbers, airports));
    JavaEsSpark.saveToEs(javaRDD, elasticIndex);
    

从ElasticSearch查询数据

	JavaRDD<Map<String, Object>> searchRdd = esRDD(jsc, "index_ik_test/fulltext", "?q=中国").values();
	for (Map<String, Object> item : searchRdd.collect()) {
	    item.forEach((key, value)->{
		System.out.println("search key:" + key + ", search value:" + value);
	    });
	}

使用?q=中国作为查询条件。整个完整示例代码如下:

import com.google.common.collect.ImmutableList;
import com.google.common.collect.ImmutableMap;
import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaRDD;
import org.apache.spark.api.java.JavaSparkContext;
import org.apache.spark.sql.SparkSession;
import org.elasticsearch.spark.rdd.api.java.JavaEsSpark;

import java.util.Map;

import static org.elasticsearch.spark.rdd.api.java.JavaEsSpark.esRDD;

/**
 * Created by Administrator on 2018/8/28.
 */
public class EsSparkTest {
    public void writeEs() {
	String elasticIndex = "spark/docs";
	//https://www.elastic.co/guide/en/elasticsearch/hadoop/current/spark.html#spark-native
	SparkConf sparkConf = new SparkConf().setAppName("writeEs").setMaster("local[*]").set("es.index.auto.create", "true")
			.set("es.nodes", "ELASTIC_SEARCH_IP").set("es.port", "9200").set("es.nodes.wan.only", "true");
	SparkSession sparkSession = SparkSession.builder().config(sparkConf).getOrCreate();
	JavaSparkContext jsc = new JavaSparkContext(sparkSession.sparkContext());//adapter
	Map<String, ?> numbers = ImmutableMap.of("one", 1, "two", 2);
	Map<String, ?> airports = ImmutableMap.of("OTP", "Otopeni", "SFO", "San Fran");
	JavaRDD<Map<String, ?>> javaRDD = jsc.parallelize(ImmutableList.of(numbers, airports));
	JavaEsSpark.saveToEs(javaRDD, elasticIndex);
    }

    public void readEs() {
	SparkConf sparkConf = new SparkConf().setAppName("writeEs").setMaster("local[*]").set("es.index.auto.create", "true")
			.set("es.nodes", "ELASTIC_SEARCH_IP").set("es.port", "9200").set("es.nodes.wan.only", "true");
	SparkSession sparkSession = SparkSession.builder().config(sparkConf).getOrCreate();
	JavaSparkContext jsc = new JavaSparkContext(sparkSession.sparkContext());//adapter
	JavaRDD<Map<String, Object>> searchRdd = esRDD(jsc, "index_ik_test/fulltext", "?q=中国").values();
	for (Map<String, Object> item : searchRdd.collect()) {
	    item.forEach((key, value)->{
		System.out.println("search key:" + key + ", search value:" + value);
	    });
	}
	sparkSession.stop();
    }
}

DemoApplication.java 入口main类

public class DemoApplication {
    public static void main(String[] args) {
	new EsSparkTest().readEs();
    }
}

IDEA菜单栏:view ---> window tools --->maven projects 打开maven 侧边栏。直接双击package打包。

$rz -bey esdemo-1.0-SNAPSHOT.jar 将打成的jar包上传到部署spark服务器上,使用如下命令提交运行:

~/spark-2.3.1-bin-hadoop2.7/bin/spark-submit --class DemoApplication esdemo-1.0-SNAPSHOT.jar

--class 是类的全路径名。如果执行过程中抛出ClassNotFoundException异常,要看一下pom.xml中指定的依赖是否在Spark安装目录下的 jars/ 目录下(比如事先把Guava jar 和 elasticsearch-hadoop-6.3.2.jar 上传到 jars/目录下)。最终执行readEs()方法查询得到的文档如下:

因为 content 字段采用的是ik_max_word分词模式,因此文本其中国家投资了500万 分词结果中包含了 中国,从而使得这篇document被查询到了。

后期补充:

在使用Spark 查询ElasticSearch中数据时,由于ElasticSearch索引user中定义了一个日期字段,如下:

    "created": {
      "type": "date",
      "format": "yyyy-MM-dd HH:mm:ss"
    }

导致Spark执行下面语句查询

JavaRDD<Map<String, Object>> searchRdd = JavaEsSpark.esRDD(jsc, "user/profile", "?q=test").values();
for (Map<String, Object> item : searchRdd.collect()) {
    item.forEach((key, value)->{
        System.out.println("search key:" + key + ", search value:" + value);
    });
}

反序列化构建日期对象时,报错:

Caused by: org.elasticsearch.hadoop.EsHadoopIllegalArgumentException: Cannot invoke method public org.joda.time.DateTime org.joda.time.format.DateTimeFormatter.parseDateTime(java.lang.String)
at org.elasticsearch.hadoop.util.ReflectionUtils.invoke(ReflectionUtils.java:93)
at org.elasticsearch.hadoop.util.DateUtils$JodaTime.parseDate(DateUtils.java:105)
at org.elasticsearch.hadoop.util.DateUtils.parseDate(DateUtils.java:122)
at org.elasticsearch.hadoop.serialization.builder.JdkValueReader.parseDate(JdkValueReader.java:424)
at org.elasticsearch.hadoop.serialization.builder.JdkValueReader.date(JdkValueReader.java:412)
at org.elasticsearch.hadoop.serialization.builder.JdkValueReader.readValue(JdkValueReader.java:88)
at org.elasticsearch.hadoop.serialization.ScrollReader.parseValue(ScrollReader.java:789)
at org.elasticsearch.hadoop.serialization.ScrollReader.read(ScrollReader.java:739)
... 31 more
Caused by: java.lang.reflect.InvocationTargetException
at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
at java.lang.reflect.Method.invoke(Method.java:498)
at org.elasticsearch.hadoop.util.ReflectionUtils.invoke(ReflectionUtils.java:91)
... 38 more
Caused by: java.lang.IllegalArgumentException: Invalid format: "2018-10-08 19:00:41" is malformed at " 19:00:41"
at org.joda.time.format.DateTimeFormatter.parseDateTime(DateTimeFormatter.java:945)
... 43 more

这应该是我索引中定义的日期格式是yyyy-MM-dd HH:mm:ss,而org.joda.time.format.DateTimeFormatter默认使用的日期格式不同导致的,但是又不知道在哪里指定日期格式进行Format,所以真的是又遇到了个坑……

如下测试,joda 是支持如下格式的日期格式的:

        String pattern = "yyyy-MM-dd HH:mm:ss";
        String aTime = "2018-10-08 19:00:41";
        DateTimeFormatter format = DateTimeFormat.forPattern(pattern);
        DateTime dateTime = format.parseDateTime(aTime);//no error

spark2.3中依赖的:joda的版本如下:

~/spark-2.3.1-bin-hadoop2.7/jars$ ls | grep joda
joda-time-2.9.3.jar

posted @ 2018-08-28 21:42  大熊猫同学  阅读(25234)  评论(1编辑  收藏  举报