近期接到一个任务,需要改造现有从mysql往Elasticsearch导入数据MTE(mysqlToEs)小工具,由于之前采用单线程导入,千亿数据需要两周左右的时间才能导入完成,导入效率非常低。所以楼主花了3天的时间,利用java线程池框架Executors中的FixedThreadPool线程池重写了MTE导入工具,单台服务器导入效率提高十几倍(合理调整线程数据,效率更高)。
干货分享:利用java多线程技术往Elasticsearch导入千亿级数据
打开今日头条,查看更多图片
如果近期有往es导入数据的同学,可以从github上下载。
传送门:
https://github.com/dunzung/mte
关键技术栈
Elasticsearch
jdbc
ExecutorService\Thread
sql
工具说明
maven依赖
<dependency>
<groupId>mysql</groupId>
<artifactId>mysql-connector-java</artifactId>
<version>${mysql.version}</version>
</dependency>
<dependency>
<groupId>org.elasticsearch</groupId>
<artifactId>elasticsearch</artifactId>
<version>${elasticsearch.version}</version>
</dependency>
<dependency>
<groupId>org.elasticsearch.client</groupId>
<artifactId>transport</artifactId>
<version>${elasticsearch.version}</version>
</dependency>
<dependency>
<groupId>org.projectlombok</groupId>
<artifactId>lombok</artifactId>
<version>${lombok.version}</version>
</dependency>
<dependency>
<groupId>com.alibaba</groupId>
<artifactId>fastjson</artifactId>
<version>${fastjson.version}</version>
</dependency>
java线程池设置
默认线程池大小为21个,可调整。其中POR为处理流程已办数据线程池,ROR为处理流程已阅数据线程池。
干货分享:利用java多线程技术往Elasticsearch导入千亿级数据
private static int THREADS = 21;
public static ExecutorService POR = Executors.newFixedThreadPool(THREADS);
public static ExecutorService ROR = Executors.newFixedThreadPool(THREADS);
定义已办生产者线程/已阅生产者线程:ZlPendProducer/ZlReadProducer
public class ZlPendProducer implements Runnable {
...
@Override
public void run() {
System.out.println(threadName + "::启动...");
for (int j = 0; j < Const.TBL.TBL_PEND_COUNT; j++)
try {
....
int size = 1000;
for (int i = 0; i < count; i += size) {
if (i + size > count) {
//作用为size最后没有100条数据则剩余几条newList中就装几条
size = count - i;
}
String sql = "select * from " + tableName + " limit " + i + ", " + size;
System.out.println(tableName + "::sql::" + sql);
rs = statement.executeQuery(sql);
List<HistPendingEntity> lst = new ArrayList<>();
while (rs.next()) {
HistPendingEntity p = PendUtils.getHistPendingEntity(rs);
lst.add(p);
}
MteExecutor.POR.submit(new ZlPendConsumer(lst));
Thread.sleep(2000);
}
....
} catch (Exception e) {
e.printStackTrace();
}
}
}
public class ZlReadProducer implements Runnable {
...已阅生产者处理逻辑同已办生产者
}
定义已办消费者线程/已阅生产者线程:ZlPendConsumer/ZlReadConsumer
public class ZlPendConsumer implements Runnable {
private String threadName;
private List<HistPendingEntity> lst;
public ZlPendConsumer(List<HistPendingEntity> lst) {
this.lst = lst;
}
@Override
public void run() {
...
lst.forEach(v -> {
try {
String json = new Gson().toJson(v);
EsClient.addDataInJSON(json, Const.ES.HistPendDB_Index, Const.ES.HistPendDB_type, v.getPendingId(), null);
Const.COUNTER.LD_P.incrementAndGet();
} catch (Exception e) {
e.printStackTrace();
System.out.println("err::PendingId::" + v.getPendingId());
}
});
...
}
}
public class ZlReadConsumer implements Runnable {
//已阅消费者处理逻辑同已办消费者
}
定义导入Elasticsearch数据监控线程:Monitor
监控线程-Monitor为了计算每分钟导入Elasticsearch的数据总条数,利用监控线程,可以调整线程池的线程数的大小,以便利用多线程更快速的导入数据。
public void monitorToES() {
new Thread(() -> {
while (true) {
StringBuilder sb = new StringBuilder();
sb.append("已办表数::").append(Const.TBL.TBL_PEND_COUNT)
.append("::已办总数::").append(Const.COUNTER.LD_P_TOTAL)
.append("::已办入库总数::").append(Const.COUNTER.LD_P);
sb.append("~~~~已阅表数::").append(Const.TBL.TBL_READ_COUNT);
sb.append("::已阅总数::").append(Const.COUNTER.LD_R_TOTAL)
.append("::已阅入库总数::").append(Const.COUNTER.LD_R);
if (ldPrevPendCount == 0 && ldPrevReadCount == 0) {
ldPrevPendCount = Const.COUNTER.LD_P.get();
ldPrevReadCount = Const.COUNTER.LD_R.get();
start = System.currentTimeMillis();
} else {
long end = System.currentTimeMillis();
if ((end - start) / 1000 >= 60) {
start = end;
sb.append("\n#########################################\n");
sb.append("已办每分钟TPS::" + (Const.COUNTER.LD_P.get() - ldPrevPendCount) + "条");
sb.append("::已阅每分钟TPS::" + (Const.COUNTER.LD_R.get() - ldPrevReadCount) + "条");
ldPrevPendCount = Const.COUNTER.LD_P.get();
ldPrevReadCount = Const.COUNTER.LD_R.get();
}
}
System.out.println(sb.toString());
try {
Thread.sleep(3000);
} catch (InterruptedException e) {
e.printStackTrace();
}
}
}).start();
}
初始化Elasticsearch:EsClient
String cName = meta.get("cName");//es集群名字
String esNodes = meta.get("esNodes");//es集群ip节点
Settings esSetting = Settings.builder()
.put("cluster.name", cName)
.put("client.transport.sniff", true)//增加嗅探机制,找到ES集群
.put("thread_pool.search.size", 5)//增加线程池个数,暂时设为5
.build();
String[] nodes = esNodes.split(",");
client = new PreBuiltTransportClient(esSetting);
for (String node : nodes) {
if (node.length() > 0) {
String[] hostPort = node.split(":");
client.addTransportAddress(new TransportAddress(InetAddress.getByName(hostPort[0]), Integer.parseInt(hostPort[1])));
}
}
初始化数据库连接
conn = DriverManager.getConnection(url, user, password);
启动参数
nohup java -jar mte.jar ES-Cluster2019 192.168.1.10:9300,192.168.1.11:9300,192.168.1.12:9300 root 123456! jdbc:mysql://192.168.1.13
:3306/mte 130 130 >> ./mte.log 2>&1 &
参数说明
ES-Cluster2019 为Elasticsearch集群名字
192.168.1.10:9300,192.168.1.11:9300,192.168.1.12:9300为es的节点IP
130 130为已办已阅分表的数据
程序入口:MteMain
干货分享:利用java多线程技术往Elasticsearch导入千亿级数据
// 监控线程
Monitor monitorService = new Monitor();
monitorService.monitorToES();
// 已办生产者线程
Thread pendProducerThread = new Thread(new ZlPendProducer(conn, "ZlPendProducer"));
pendProducerThread.start();
// 已阅生产者线程
Thread readProducerThread = new Thread(new ZlReadProducer(conn, "ZlReadProducer"));
readProducerThread.start();
小试牛刀
干货分享:利用java多线程技术往Elasticsearch导入千亿级数据
干货分享:利用java多线程技术往Elasticsearch导入千亿级数据