本文介绍了一种高效地将MySQL数据库中的大量数据批量导入到Elasticsearch的方法。通过使用Elasticsearch提供的BulkProcessor API，并结合合理的配置参数，实现了高达2.5万条记录每秒的导入速度。
- 注：笔者环境 ES6.6.2、linux centos6.9、mysql8.0、三个节点、节点内存64G、八核CPU
场景：
- 目前Mysql 数据库数据量约10亿，有几张大表1亿左右，直接在Mysql查询出现各种效率问题，因此想着将数据导一份到ES，从而实现大数据快速检索的功能。
- 通过Logstash插件批量导数据，个人感觉出现各种奇怪的问题，例如ES 内存暴满，mysql 所在服务器内存暴，最主要的是在一次导数时不能导数据量不能太大。经过一次次试探Logstash与优化Logstash导数后，终于还是动手直接运用ES提供的api进行导数了。
- 目前直接模拟测试批量导数据，无论是通过Logstash还是ES 提供的Api峰值均能达到10万每秒左右。下面上代码，主要是通过官网提供的api （RestHighLevelClient、BulkProcessor）整理而来。目前由于Mysql 查询出来的数据需要进行一些处理，基本可达到3.5万+每秒。这个速度还有不小的优化空间，比如笔者通过稍微修改下述代码，启动几个线程同时执行bulk多张表，从kibana界面看出速度达到了成倍的提升，因为速度已基本达到笔者所想要的，便不怎么进行优化代码了。
一、Maven 配置如下
<project xmlns="http://maven.apache.org/POM/4.0.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
<modelVersion>4.0.0</modelVersion>
<groupId>ElasticSearchDemo</groupId>
<artifactId>ElasticSearchDemo</artifactId>
<version>0.0.1-SNAPSHOT</version>
<packaging>jar</packaging>
<name>ElasticSearchDemo</name>
<url>http://maven.apache.org</url>
<properties>
<project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
</properties>
<dependencies>
<dependency>
<groupId>junit</groupId>
<artifactId>junit</artifactId>
<version>3.8.1</version>
<scope>test</scope>
</dependency>
<dependency>
<groupId>org.elasticsearch.client</groupId>
<artifactId>elasticsearch-rest-high-level-client</artifactId>
<version>6.6.2</version>
</dependency>
<dependency>
<groupId>mysql</groupId>
<artifactId>mysql-connector-java</artifactId>
<version>8.0.11</version>
</dependency>
<dependency>
<groupId>org.apache.logging.log4j</groupId>
<artifactId>log4j-core</artifactId>
<version>2.11.1</version>
</dependency>
</dependencies>
</project>
二、代码如下
package service;
import java.io.IOException;
import java.sql.Connection;
import java.sql.PreparedStatement;
import java.sql.ResultSet;
import java.sql.ResultSetMetaData;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.concurrent.TimeUnit;
import java.util.function.BiConsumer;
import org.apache.http.HttpHost;
import org.apache.logging.log4j.LogManager;
import org.apache.logging.log4j.Logger;
import org.elasticsearch.action.ActionListener;
import org.elasticsearch.action.admin.indices.create.CreateIndexRequest;
import org.elasticsearch.action.admin.indices.create.CreateIndexResponse;
import org.elasticsearch.action.bulk.BackoffPolicy;
import org.elasticsearch.action.bulk.BulkProcessor;
import org.elasticsearch.action.bulk.BulkRequest;
import org.elasticsearch.action.bulk.BulkResponse;
import org.elasticsearch.action.index.IndexRequest;
import org.elasticsearch.client.RequestOptions;
import org.elasticsearch.client.RestClient;
import org.elasticsearch.client.RestHighLevelClient;
import org.elasticsearch.common.settings.Settings;
import org.elasticsearch.common.unit.ByteSizeUnit;
import org.elasticsearch.common.unit.ByteSizeValue;
import org.elasticsearch.common.unit.TimeValue;
import utils.DBHelper;
/**
* @author Ye
* @time 2019年3月29日
*
* 类说明：通过BulkProcess批量将Mysql数据导入ElasticSearch中
*/
public class BulkProcessDemo {
private static final Logger logger = LogManager.getLogger(BulkProcessDemo.class);
public static void main(String[] args) {
try {
long startTime = System.currentTimeMillis();
String tableName = "testTable";
createIndex(tableName);
writeMysqlDataToES(tableName);
logger.info(" use time: " + (System.currentTimeMillis() - startTime) / 1000 + "s");
} catch (Exception e) {
logger.error(e.getMessage());
e.printStackTrace();
}
}
/**
* 创建索引
* @param indexName
* @throws IOException
*/
public static void createIndex(String indexName) throws IOException {
RestHighLevelClient client = new RestHighLevelClient(RestClient.builder(new HttpHost("es01", 9200, "http")));
// ES 索引默认需要小写，故笔者将其转为小写
CreateIndexRequest requestIndex = new CreateIndexRequest(indexName.toLowerCase());
// 注： 设置副本数为0，索引刷新时间为-1对大批量索引数据效率的提升有不小的帮助
requestIndex.settings(Settings.builder().put("index.number_of_shards", 5)
.put("index.number_of_replicas", 0)
.put("index.refresh_interval", "-1"));
// CreateIndexResponse createIndexResponse = client.indices().create(requestIndex, RequestOptions.DEFAULT);
client.close();
}
/**
* 将mysql 数据查出组装成es需要的map格式，通过批量写入es中
*
* @param tableName
*/
private static void writeMysqlDataToES(String tableName) {
RestHighLevelClient client = new RestHighLevelClient(RestClient.builder(new HttpHost("eshost", 9200, "http")));// 初始化
BulkProcessor bulkProcessor = getBulkProcessor(client);
Connection conn = null;
PreparedStatement ps = null;
ResultSet rs = null;
try {
conn = DBHelper.getConn();
logger.info("Start handle data :" + tableName);
String sql = "SELECT * from " + tableName;
ps = conn.prepareStatement(sql, ResultSet.TYPE_FORWARD_ONLY, ResultSet.CONCUR_READ_ONLY);
ps.setFetchSize(Integer.MIN_VALUE);
rs = ps.executeQuery();
ResultSetMetaData colData = rs.getMetaData();
ArrayList<HashMap<String, String>> dataList = new ArrayList<HashMap<String, String>>();
// bulkProcessor 添加的数据支持的方式并不多，查看其api发现其支持map键值对的方式，故笔者在此将查出来的数据转换成hashMap方式
HashMap<String, String> map = null;
int count = 0;
String c = null;
String v = null;
while (rs.next()) {
count++;
map = new HashMap<String, String>(128);
for (int i = 1; i <= colData.getColumnCount(); i++) {
c = colData.getColumnName(i);
v = rs.getString(c);
map.put(c, v);
}
dataList.add(map);
// 每10万条写一次，不足的批次的最后再一并提交
if (count % 100000 == 0) {
logger.info("Mysql handle data number : " + count);
// 将数据添加到 bulkProcessor 中
for (HashMap<String, String> hashMap2 : dataList) {
bulkProcessor.add(new IndexRequest(tableName.toLowerCase(), "gzdc", hashMap2.get("S_GUID"))
.source(hashMap2));
}
// 每提交一次便将map与list清空
map.clear();
dataList.clear();
}
}
// count % 100000 处理未提交的数据
for (HashMap<String, String> hashMap2 : dataList) {
bulkProcessor.add(
new IndexRequest(tableName.toLowerCase(), "gzdc", hashMap2.get("S_GUID")).source(hashMap2));
}
logger.info("-------------------------- Finally insert number total : " + count);
// 将数据刷新到es, 注意这一步执行后并不会立即生效，取决于bulkProcessor设置的刷新时间
bulkProcessor.flush();
} catch (Exception e) {
logger.error(e.getMessage());
} finally {
try {
rs.close();
ps.close();
conn.close();
boolean terminatedFlag = bulkProcessor.awaitClose(150L, TimeUnit.SECONDS);
client.close();
logger.info(terminatedFlag);
} catch (Exception e) {
logger.error(e.getMessage());
}
}
}
/**
* 创建bulkProcessor并初始化
* @param client
* @return
*/
private static BulkProcessor getBulkProcessor(RestHighLevelClient client) {
BulkProcessor bulkProcessor = null;
try {
BulkProcessor.Listener listener = new BulkProcessor.Listener() {
@Override
public void beforeBulk(long executionId, BulkRequest request) {
logger.info("Try to insert data number : " + request.numberOfActions());
}
@Override
public void afterBulk(long executionId, BulkRequest request, BulkResponse response) {
logger.info("************** Success insert data number : " + request.numberOfActions() + " , id: "
+ executionId);
}
@Override
public void afterBulk(long executionId, BulkRequest request, Throwable failure) {
logger.error("Bulk is unsuccess : " + failure + ", executionId: " + executionId);
}
};
BiConsumer<BulkRequest, ActionListener<BulkResponse>> bulkConsumer = (request, bulkListener) -> client
.bulkAsync(request, RequestOptions.DEFAULT, bulkListener);
// bulkProcessor = BulkProcessor.builder(bulkConsumer, listener).build();
BulkProcessor.Builder builder = BulkProcessor.builder(bulkConsumer, listener);
builder.setBulkActions(5000);
builder.setBulkSize(new ByteSizeValue(100L, ByteSizeUnit.MB));
builder.setConcurrentRequests(10);
builder.setFlushInterval(TimeValue.timeValueSeconds(100L));
builder.setBackoffPolicy(BackoffPolicy.constantBackoff(TimeValue.timeValueSeconds(1L), 3));
// 注意点：在这里感觉有点坑，官网样例并没有这一步，而笔者因一时粗心也没注意，在调试时注意看才发现，上面对builder设置的属性没有生效
bulkProcessor = builder.build();
} catch (Exception e) {
e.printStackTrace();
try {
bulkProcessor.awaitClose(100L, TimeUnit.SECONDS);
client.close();
} catch (Exception e1) {
logger.error(e1.getMessage());
}
}
return bulkProcessor;
}
｝
数据库连接类
package utils;
import java.sql.Connection;
import java.sql.DriverManager;
public class DBHelper {
public static final String url = "jdbc:mysql://xx.xx.xx.xx:3306/xxdemo?useSSL=true";
public static final String name = "com.mysql.cj.jdbc.Driver";
public static final String user = "xxx";
public static final String password = "xxxx";
public static Connection conn = null;
public static Connection getConn() {
try {
Class.forName(name);
conn = DriverManager.getConnection(url, user, password);//获取连接
} catch (Exception e) {
e.printStackTrace();
}
return conn;
}
}
日志文件配置文件：log4j2.properties
property.filePath=logs
property.filePattern=logs/%d{yyyy}/%d{MM}
#\u8F93\u51FA\u683C\u5F0F
property.layoutPattern=%-d{yyyy-MM-dd HH:mm:ss SSS} [ %p ] [ %c ] %m%n
rootLogger.level = info
appender.console.type = Console
appender.console.name = STDOUT
appender.console.target = SYSTEM_OUT
appender.console.layout.type = PatternLayout
appender.console.layout.pattern = ${layoutPattern}
rootLogger.appenderRef.stdout.ref = STDOUT
appender.I.type = RollingFile
appender.I.name = InfoRollingFile
appender.I.fileName = ${filePath}/es-info.log
appender.I.filePattern = ${filePattern}/es_info.log
appender.I.layout.type = PatternLayout
appender.I.layout.pattern = ${layoutPattern}
appender.I.policies.type = Policies
appender.I.policies.time.type = TimeBasedTriggeringPolicy
appender.I.policies.time.interval = 1
appender.I.policies.time.modulate = true
appender.I.policies.size.type = SizeBasedTriggeringPolicy
appender.I.policies.size.size=20M
appender.I.strategy.type = DefaultRolloverStrategy
appender.I.strategy.max = 100
#\u8FC7\u6EE4INFO\u4EE5\u4E0A\u4FE1\u606F
appender.I.filter.threshold.type = ThresholdFilter
appender.I.filter.threshold.level = WARN
appender.I.filter.threshold.onMatch = DENY
appender.I.filter.threshold.onMisMatch=NEUTRAL
rootLogger.appenderRef.I.ref = InfoRollingFile
rootLogger.appenderRef.I.level=INFO
appender.E.type = RollingFile
appender.E.name = ErrorRollingFile
appender.E.fileName = ${filePath}/es-error.log
appender.E.filePattern = ${filePattern}/es_error.log
appender.E.layout.type = PatternLayout
appender.E.layout.pattern = ${layoutPattern}
appender.E.policies.type = Policies
appender.E.policies.time.type = TimeBasedTriggeringPolicy
appender.E.policies.time.interval = 1
appender.E.policies.time.modulate = true
appender.E.policies.size.type = SizeBasedTriggeringPolicy
appender.E.policies.size.size=20M
appender.E.strategy.type = DefaultRolloverStrategy
appender.E.strategy.max = 100
#\u8FC7\u6EE4ERROR\u4EE5\u4E0A\u4FE1\u606F
appender.E.filter.threshold.type = ThresholdFilter
appender.E.filter.threshold.level = FATAL
appender.E.filter.threshold.onMatch = DENY
appender.E.filter.threshold.onMisMatch=NEUTRAL
rootLogger.appenderRef.E.ref = ErrorRollingFile
rootLogger.appenderRef.E.level=ERROR
核心代码解说：
三、验证上述代码结果
1、起初笔者直接在本地Eclipse中运行，由于电脑配置所限制，写入es 速度只能达到5000/s左右的样子。
2、直接将代码打包成可运行的jar，通过java -jar xxx.jar 方式在linux中运行，基本可以达到2.5万/s，结果如下图所示：
- 注：笔者下图是将一张具有5400万的数据，一次性导入ES，该表60多个字段
运行的部分日志如下：
2019-03-29 17:31:34 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2679
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] Mysql handle data number : 13500000
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2681
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2683
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2682
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:36 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2685
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2686
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2687
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2684
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2688
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2689
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2691
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2693
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2692
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2690
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] Try to insert data number : 5000
2019-03-29 17:31:37 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2696
2019-03-29 17:31:38 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2694
2019-03-29 17:31:38 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2695
2019-03-29 17:31:38 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2697
2019-03-29 17:31:38 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2698
2019-03-29 17:31:38 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2699
2019-03-29 17:31:38 [ INFO ] [ service.BulkProcessDemo ] ************** Success insert data number : 5000 , id: 2700
简单修改下相应参数进行bulk，查看效率
- 修改下每个批次上传 10000条数据，setBulkSize 设置成300M，数据处理批次修改成20万（if (count % 200000 == 0) ），其它配置一样，处理另一张大表，效果如下：
-
该表数据量：该表27个字段
SELECT count(1) from xxxxxxxxx; – 171769567 -
全量bulk进es用时：5113s
2019-04-01 11:39:04.453 INFO 28080 — [nio-8080-exec-1] s.ElasticServiceImpl : use time: 5113s -
es 监控如下图
四、小结
- 笔者个人不深入的学习感觉 Logstash 对于这种需要一次性大批量将数据导入ES的需求适应性可能不太好。
- 大批量的数据导入ES 个人推荐BulkProcessor, 不论是什么数据，只需要将其转换成Map键值对的格式便可运用bulkProcessor实现流式导入。
- 对于导入效率需要看集群的环境以及导入批次的设置，还有ES的相关优化配置。
五、遇到的小坑
- 问题一
由于表数据量太大，一开始本想以时间段（原数据库中有时间相关的字段）分批查出数据再将其导入ES（在用Logstash插件导入时就是这么处理的，但由于数据在不同时间的分布情况很不一样，在运用Logstash插件导入时经常因为某一时间段数据量太大而导致死机问题）。
解决：JDBC resltset中fetchSize的设置。
详见：[正确使用MySQL JDBC setFetchSize()方法解决JDBC处理大结果集内在溢出]
- 问题二
通过官网提供的代码设置了bulder 相关的属性，比如每个批次提交多少数据，但程序运行起来发现这些设置并没有生效。
官网代码如下：
BulkProcessor bulkProcessor =
BulkProcessor.builder(bulkConsumer, listener).build();
BiConsumer<BulkRequest, ActionListener<BulkResponse>> bulkConsumer =
(request, bulkListener) ->
client.bulkAsync(request, RequestOptions.DEFAULT, bulkListener);
BulkProcessor.Builder builder =
BulkProcessor.builder(bulkConsumer, listener);
builder.setBulkActions(500);
builder.setBulkSize(new ByteSizeValue(1L, ByteSizeUnit.MB));
builder.setConcurrentRequests(0);
builder.setFlushInterval(TimeValue.timeValueSeconds(10L));
builder.setBackoffPolicy(BackoffPolicy
.constantBackoff(TimeValue.timeValueSeconds(1L), 3));
解决：其实该坑的出现主要是笔者个人在此太粗心了，还有就是太相信官网了。
在这个问题上稍微留意下上述代码就可以看出 builder 虽然设置了，但并没有将其设置后的值转给那个类调用了，因此bulkProcessor还是使用了默认的配置。
BulkProcessor bulkProcessor =
BulkProcessor.builder(bulkConsumer, listener).build();
其实这一步可以换成如下（该bulkProcessor 的创建应在builder 属性设置完成之后）
BulkProcessor bulkProcessor = builder.build();
# 完整代码如下：
BulkProcessor.Builder builder = BulkProcessor.builder(bulkConsumer, listener);
builder.setBulkActions(5000);
builder.setBulkSize(new ByteSizeValue(100L, ByteSizeUnit.MB));
builder.setConcurrentRequests(10);
builder.setFlushInterval(TimeValue.timeValueSeconds(100L));
builder.setBackoffPolicy(BackoffPolicy.constantBackoff(TimeValue.timeValueSeconds(1L), 3));
bulkProcessor = builder.build();
在这里可能会感觉如何将bulkProcessor 转给监听者（listener），查看builder.build(); 瞄了下底层会发现其实该方法最终是调用如下方法的：
/**
* Builds a new bulk processor.
*/
public BulkProcessor build() {
return new BulkProcessor(consumer, backoffPolicy, listener, concurrentRequests, bulkActions,
bulkSize, flushInterval, scheduler, onClose, createBulkRequestWithGlobalDefaults());
}
完整代码见：https://github.com/yechunbo/BigdataSearchPro.git
参考文档：
- https://www.elastic.co/guide/en/elasticsearch/client/java-rest/6.6/java-rest-high-document-bulk.html
- https://www.elastic.co/guide/en/elasticsearch/client/java-api/6.6/_log4j_2_logger.html
- https://docs.oracle.com/cd/E11882_01/java.112/e16548/resltset.htm#JJDBC28622
- 正确使用MySQL JDBC setFetchSize()方法解决JDBC处理大结果集内在溢出
- JDBC读取数据优化-fetch size
扫一扫
7498
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
