flume自定义sink开发——flume clickhouse sink

flume优点之一就是支持插件扩展功能，现在clickhouse流行，数据想直接写入clickhouse,flume官网看不了一样，没有clickhouse sink，没有咱们就自已写呗。

网上开发自定义拦截器的文章很多，开发sink的反而找不到，特记录一下，供大家参考。

一、开发流程

1. 搭建flume开发环境。
2. 新建一个类，实现Configurable接口，继承AbstractSink类。
3. 重写configure、start、stop、process方法。
4. 编译打jar包，放到flume的lib目录下。
5. flume.conf文件增加clickhouse的配置。

二、搭建flume开发环境。

新建maven工程，在pom.xml添加如下依赖。

<dependencies>
    <dependency>
        <groupId>org.apache.flumegroupId>
        <artifactId>flume-ng-coreartifactId>
    dependency>

    <dependency>
        <groupId>com.google.guavagroupId>
        <artifactId>guavaartifactId>
        <version>19.0version>
    dependency>

    <dependency>
        <groupId>ru.yandex.clickhousegroupId>
        <artifactId>clickhouse-jdbcartifactId>
        <version>0.2version>
    dependency>
dependencies>

三、构建自定义ClickHouseSink类代码

package org.apache.flume.sink.clickhouse;

import org.apache.flume.conf.Configurable;
import org.apache.flume.sink.AbstractSink;

public class ClickHouseSink extends AbstractSink implements Configurable {
    @Override
    public void configure(Context context) {
    }

    @Override
    public void start() {
    }

    @Override
    public void stop() {
    }

    @Override
    public Status process() throws EventDeliveryException {
    }
}

四、实现说明：

configure函数是解析处理配置参数，它接收Context对象，可以用context.getString(“xxx”)获取配置信息，例如context.getString(“host”)可以获得ClickHouseSink配置的host参数。

	 @Override
    public void configure(Context context) {
        Preconditions.checkArgument(context.getString(HOST) != null && context.getString(HOST).length() > 0, "ClickHouse host must be specified!");
        this.host = context.getString(HOST);
        if (!this.host.startsWith("jdbc:clickhouse://")) {
            this.host = "jdbc:clickhouse://" + this.host;
        }
    }

start函数是启动是的初始化，用于建立clickhouse连接对象，启动flume内置计数器等。

@Override
    public void start() {
        String jdbcUrl = String.format("%s:%s/%s", this.host, this.port, this.database);
        ClickHouseProperties properties = new ClickHouseProperties().withCredentials(this.user, this.password);
        this.dataSource = new BalancedClickhouseDataSource(jdbcUrl, properties);
        sinkCounter.start();
        super.start();
    }

stop函数是负责退出前的回收清理工作。

@Override
    public void stop() {
        logger.info("ClickHouse sink {} stopping", getName());
        sinkCounter.incrementConnectionClosedCount();
        sinkCounter.stop();
        super.stop();
    }

process函数是就核心的处理函数了。要注意是的flume是数据传输是事务的，可以保证数据不丟失，所以我们开发的sink在消费channel的数据时，也是要用事务。

@Override
    public Status process() throws EventDeliveryException {
        Status status = null;
        Channel ch = getChannel();
        Transaction txn = ch.getTransaction();
        txn.begin();
        try {
            txn.commit();
        } catch (Throwable t) {
            txn.rollback();
        } finally {
            txn.close();
        }
        return status;
    }

clickhouse的数据写入方式有很多种，具体大家可以参考官方文档，这里我们用的是json方式写入。

ClickHouseStatement sth = conn.createStatement();
    sth.write().table(String.format(" %s.%s", database, table)).data(new ByteArrayInputStream(batch.toString().getBytes()), ClickHouseFormat.JSONEachRow).addDbParam(ClickHouseQueryParam.MAX_PARALLEL_REPLICAS, "2").send();

每次操作还需要修改flume的内置计数器，例如：

sinkCounter.addToEventDrainAttemptCount(count);//准备处理的event的个数
sinkCounter.incrementEventDrainSuccessCount();//处理成功，将准备处理的event数量累加到成功处理上。

五、编译打jar包

用mvn package编译，生成flume-ng-clickhouse-sink-1.0.jar包，放到flume的lib目录下。

六、flume clickhouse sink 配置

standard_storage.sinks.sink2ch.type = org.apache.flume.sink.clickhouse.ClickHouseSink
standard_storage.sinks.sink2ch.channel = channel2ch
standard_storage.sinks.sink2ch.host = xxxx.xxxxx.com
standard_storage.sinks.sink2ch.port = 8123
standard_storage.sinks.sink2ch.database = xxxxx
standard_storage.sinks.sink2ch.table = xxxxx
standard_storage.sinks.sink2ch.batchSize = 10000
standard_storage.sinks.sink2ch.user = xxxxxx
standard_storage.sinks.sink2ch.password = xxxxxxxxxxxx

上这个例子，配置clickhouse的host、port、database、table、batchSize、user、password即可。type是要配ClickHouseSink的main类名

七、完整代码

ClickHouseSink.java

package org.apache.flume.sink.clickhouse;

import com.google.common.base.Preconditions;
import org.apache.flume.*;
import org.apache.flume.conf.Configurable;
import org.apache.flume.instrumentation.SinkCounter;
import org.apache.flume.sink.AbstractSink;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import ru.yandex.clickhouse.BalancedClickhouseDataSource;
import ru.yandex.clickhouse.ClickHouseConnectionImpl;
import ru.yandex.clickhouse.ClickHouseStatement;
import ru.yandex.clickhouse.domain.ClickHouseFormat;
import ru.yandex.clickhouse.settings.ClickHouseProperties;
import ru.yandex.clickhouse.settings.ClickHouseQueryParam;

import java.io.ByteArrayInputStream;

import static org.apache.flume.sink.clickhouse.ClickHouseSinkConstants.*;

public class ClickHouseSink extends AbstractSink implements Configurable {

    private static final Logger logger = LoggerFactory.getLogger(ClickHouseSink.class);

    private BalancedClickhouseDataSource dataSource = null;
    private SinkCounter sinkCounter = null;
    private String host = null;
    private String port = null;
    private String user = null;
    private String password = null;
    private String database = null;
    private String table = null;

    private int batchSize;


    @Override
    public void configure(Context context) {

        if (sinkCounter == null) {
            sinkCounter = new SinkCounter(getName());
        }

        Preconditions.checkArgument(context.getString(HOST) != null && context.getString(HOST).length() > 0, "ClickHouse host must be specified!");
        this.host = context.getString(HOST);
        if (!this.host.startsWith("jdbc:clickhouse://")) {
            this.host = "jdbc:clickhouse://" + this.host;
        }

        Preconditions.checkArgument(context.getString(DATABASE) != null && context.getString(DATABASE).length() > 0, "ClickHouse database must be specified!");
        this.database = context.getString(DATABASE);
        Preconditions.checkArgument(context.getString(TABLE) != null && context.getString(TABLE).length() > 0, "ClickHouse table must be specified!");
        this.table = context.getString(TABLE);
        this.port = context.getString(PORT, DEFAULT_PORT);
        this.user = context.getString(USER, DEFAULT_USER);
        this.password = context.getString(PASSWORD, DEFAULT_PASSWORD);
        this.batchSize = context.getInteger(BATCH_SIZE, DEFAULT_BATCH_SIZE);
    }

    @Override
    public void start() {
        String jdbcUrl = String.format("%s:%s/%s", this.host, this.port, this.database);
        ClickHouseProperties properties = new ClickHouseProperties().withCredentials(this.user, this.password);
        //properties.setUseServerTimeZone(false);
        this.dataSource = new BalancedClickhouseDataSource(jdbcUrl, properties);
        sinkCounter.start();
        super.start();
    }


    @Override
    public void stop() {
        logger.info("ClickHouse sink {} stopping", getName());
        sinkCounter.incrementConnectionClosedCount();
        sinkCounter.stop();
        super.stop();
    }


    @Override
    public Status process() throws EventDeliveryException {
        Status status = null;

        // Start transaction
        Channel ch = getChannel();
        Transaction txn = ch.getTransaction();
        txn.begin();
        try {
            ClickHouseConnectionImpl conn = (ClickHouseConnectionImpl) dataSource.getConnection();
            int count;
            StringBuilder batch = new StringBuilder();
            for (count = 0; count < batchSize; ++count) {
                Event event = ch.take();
                if (event == null) {
                    break;
                }
                batch.append(new String(event.getBody(), "UTF-8")).append("\n");
            }
            if (count <= 0) {
                sinkCounter.incrementBatchEmptyCount();
                txn.commit();
                return Status.BACKOFF;
            } else if (count < batchSize) {
                sinkCounter.incrementBatchUnderflowCount();
            } else {
                sinkCounter.incrementBatchCompleteCount();
            }
            sinkCounter.addToEventDrainAttemptCount(count);
            ClickHouseStatement sth = conn.createStatement();
            sth.write().table(String.format(" %s.%s", database, table)).data(new ByteArrayInputStream(batch.toString().getBytes()), ClickHouseFormat.JSONEachRow).addDbParam(ClickHouseQueryParam.MAX_PARALLEL_REPLICAS, "2").send();
            sinkCounter.incrementEventDrainSuccessCount();
            status = Status.READY;
            txn.commit();

        } catch (Throwable t) {
            txn.rollback();
            logger.error(t.getMessage(), t);
            status = Status.BACKOFF;
            // re-throw all Errors
            if (t instanceof Error) {
                throw (Error) t;
            }
        } finally {
            txn.close();
        }
        return status;
    }
}

ClickHouseSinkConstants.java

package org.apache.flume.sink.clickhouse;

public class ClickHouseSinkConstants {
    public static final String HOST = "host";
    public static final String PORT = "port";
    public static final String BATCH_SIZE = "batchSize";
    public static final String USER = "user";
    public static final String PASSWORD = "password";
    public static final String DATABASE = "database";
    public static final String TABLE = "table";
    public static final String DEFAULT_PORT = "8123";
    public static final int DEFAULT_BATCH_SIZE = 10000;
    public static final String DEFAULT_USER = "";
    public static final String DEFAULT_PASSWORD = "";
}