Flume拦截器使用-实现分表、解决零点漂移等

1.场景分析

使用flume做数据传输时，可能遇到将一个数据流中的多张表分别保存到各自位置的问题，同时由于采集时间和数据实际发生时间存在差异，因此需要根据数据实际发生时间进行分区保存。
鉴于此，需要设计flume拦截器配置conf文件实现上述功能，废话不多说，直接上代码。

2.配置文件

<dependencies>
        <dependency>
            <groupId>org.apache.flumegroupId>
            <artifactId>flume-ng-coreartifactId>
            <version>1.9.0version>
            <scope>providedscope>
        dependency>

        <dependency>
            <groupId>com.alibabagroupId>
            <artifactId>fastjsonartifactId>
            <version>1.2.83_noneautotypeversion>
        dependency>
    dependencies>

3.主程序

public class test implements Interceptor  {
    @Override
    public void initialize() {

    }

    @Override
    public Event intercept(Event event) {
        //1、获取header和body的数据
        try {
            byte[] body = event.getBody();
            Map<String, String> headers = event.getHeaders();
            String log = new String(body, StandardCharsets.UTF_8);
            //2、判断字符串是否是一个合法的json，是：返回当前event；不是：返回null
            JSONObject jsonObject = JSONObject.parseObject(log);
            //3、header中timestamp时间字段替换成日志生成的时间戳（解决数据漂移问题和历史数据同步）
            Long ts = DateFormatUtil.toTsAddTimeZone(jsonObject.getString("@timestamp"));
            String topicHeader = headers.get("topic");
//            System.out.println("topicHeader主题名称为:"+topicHeader);
            String httpUserAgent = jsonObject.getString("topicHeader");
            //数据筛选
            if("clb-healthcheck".equals(httpUserAgent) || (StringUtils.isNotEmpty(httpUserAgent) && httpUserAgent.startsWith("kube-probe"))){
//                System.out.println("过滤的事件为:"+event);
                return null;
            }else {
                jsonObject.put("timestamp",ts);
                headers.put("timestamp", ts.toString());
                if("xxx".equals(topicHeader)){
                    headers.put("table","table1");
                }else if("xxxx".equals(topicHeader)){
                    headers.put("table","table2");
                }else{
                    headers.put("table","other");
                }

                event.setBody(jsonObject.toString().getBytes(StandardCharsets.UTF_8));
//                System.out.println("传输的事件为:"+event);
                return event;
            }

        }catch (JSONException e){
//            System.out.println("格式有问题的事件为:"+event);
            return null;
        }
    }


    @Override
    public List<Event> intercept(List<Event> list) {
        Iterator<Event> iterator = list.iterator();

        while (iterator.hasNext()){
            Event next = iterator.next();
            if(intercept(next)==null){
                iterator.remove();
            }
        }

        return list;
    }

    @Override
    public void close() {

    }
    public static class Builder implements Interceptor.Builder {
        @Override
        public Interceptor build() {
            return new test();
        }

        @Override
        public void configure(Context context) {
        }
    }
}

4.utils-时间处理程序

/**
 * 日期转换工具类
 * 注意：SimpleDateFormat在对日期进行转换的时候，存在线程安全的问题
 * 建议：使用JDK1.8之后提供的日期包下的相关类完成封装
 */
public class DateFormatUtil {
    private static final DateTimeFormatter dtf = DateTimeFormatter.ofPattern("yyyy-MM-dd");
    private static final DateTimeFormatter dtf1 = DateTimeFormatter.ofPattern("yyyyMMdd");
    private static final DateTimeFormatter dtf2 = DateTimeFormatter.ofPattern("yyyyMMddHH");
    private static final DateTimeFormatter dtfFull = DateTimeFormatter.ofPattern("yyyy-MM-dd HH:mm:ss");
    private static final DateTimeFormatter dtfFull1 = DateTimeFormatter.ofPattern("yyyy/MM/dd HH:mm:ss SSS");
    private static final DateTimeFormatter dtfFull2 = DateTimeFormatter.ofPattern("yyyy-MM-dd'T'HH:mm:ss.SSS'Z'");

    public static Long toTs(String dtStr, boolean isFull) {

        LocalDateTime localDateTime = null;
        if (!isFull) {
            dtStr = dtStr + " 00:00:00";
        }
        localDateTime = LocalDateTime.parse(dtStr, dtfFull);

        return localDateTime.toInstant(ZoneOffset.of("+8")).toEpochMilli();
    }

    public static Long toTsAddTimeZone(String dtStr) {

        LocalDateTime localDateTime = null;
        localDateTime = LocalDateTime.parse(dtStr, dtfFull2);

        return localDateTime.toInstant(ZoneOffset.of("+0")).toEpochMilli();
    }
    public static Long toTs1(String dtStr) {

        LocalDateTime localDateTime = null;

        localDateTime = LocalDateTime.parse(dtStr, dtfFull1);

        return localDateTime.toInstant(ZoneOffset.of("+8")).toEpochMilli();
    }


    public static Long toTs(String dtStr) {
        return toTs(dtStr, false);
    }

    public static String toDate(Long ts) {
        Date dt = new Date(ts);
        LocalDateTime localDateTime = LocalDateTime.ofInstant(dt.toInstant(), ZoneId.systemDefault());
        return dtf.format(localDateTime);
    }

    public static String toYmdHms(Long ts) {
        Date dt = new Date(ts);
        LocalDateTime localDateTime = LocalDateTime.ofInstant(dt.toInstant(), ZoneId.systemDefault());
        return dtfFull.format(localDateTime);
    }

    public static String toYmd(Long ts) {
        Date dt = new Date(ts);
        LocalDateTime localDateTime = LocalDateTime.ofInstant(dt.toInstant(), ZoneId.systemDefault());
        return dtf1.format(localDateTime);
    }
    public static String toYmdH(Long ts) {
        Date dt = new Date(ts);
        LocalDateTime localDateTime = LocalDateTime.ofInstant(dt.toInstant(), ZoneId.systemDefault());
        return dtf2.format(localDateTime);
    }
    public static int toYmdHInt(Long ts) {
        Date dt = new Date(ts);
        LocalDateTime localDateTime = LocalDateTime.ofInstant(dt.toInstant(), ZoneId.systemDefault());
        return Integer.parseInt(dtf2.format(localDateTime));
    }

    public static void main(String[] args) {


        long t1= 1670833135997L;
        String s1 = toYmdH(t1);
        int i1 = toYmdHInt(t1);
        String s2 = toYmdH(1670833136000L);
//        long s3 = toTsAddTimeZone("2024-01-31 05:43:46");
        long s4 = toTsAddTimeZone("2024-01-31T06:11:54.000Z");

//        System.out.println(s3);
        System.out.println(s4);



    }

}

5.打包放入flume的lib目录

mv test.jar /data/flume-1.9.0/lib/

6.编写配置文件运行

#定义组件
a1.sources=r1
a1.channels=c1
a1.sinks=k1 k2

#配置source
a1.sources.r1.type= org.apache.flume.source.kafka.KafkaSource
a1.sources.r1.batchSize = 2000
a1.sources.r1.kafka.consumer.group.id= xxx
a1.sources.r1.batchDurationMillis = 2000
a1.sources.r1.kafka.bootstrap.servers = xxx:9092
a1.sources.r1.kafka.topics = xxx,xxxx
a1.sources.r1.kafka.consumer.auto.offset.reset = latest
a1.sources.r1.interceptors = i1
#此处需要写jar包的详细reference
a1.sources.r1.interceptors.i1.type =test$Builder



#memory channel
a1.channels.c1.type = memory
#channel的event个数
a1.channels.c1.capacity = 20000
#事务event个数
a1.channels.c1.transactionCapacity = 10000
a1.channels.c1.byteCapacityBufferPercentage = 20
a1.channels.c1.byteCapacity = 2147483648

#配置channel
#a1.channels.c1.type = file
#a1.channels.c1.checkpointDir =/data/xxx
#a1.channels.c1.dataDirs = /data/module/xxx
#a1.channels.c1.maxFileSize = 2147483648
#a1.channels.c1.capacity = 2000000
#a1.channels.c1.transactionCapacity=20000
#a1.channels.c1.keep-alive = 6
#a1.chhannels.c1.checkpointInterval=60000
#a1.minimumRequirdSpace=26214400

#配置sink1
a1.sinks.k1.type = hdfs
a1.sinks.k1.hdfs.path =/hadoop/dm_dw/tmp_data/log/%{table}/%Y%m%d/%H
a1.sinks.k1.hdfs.filePrefix = log1
a1.sinks.k1.hdfs.round = false
a1.sinks.k1.hdfs.rollInterval = 360
a1.sinks.k1.hdfs.rollSize = 1174405120
a1.sinks.k1.hdfs.rollCount = 0
a1.sinks.k1.hdfs.batchSize=3000
#控制输出文件类型
a1.sinks.k1.hdfs.fileType = CompressedStream
a1.sinks.k1.hdfs.codeC = gzip


#配置sink2
a1.sinks.k2.type = hdfs
a1.sinks.k2.hdfs.path =/hadoop/dm_dw/tmp_data/log/%{table}/%Y%m%d/%H
a1.sinks.k2.hdfs.filePrefix = log2
a1.sinks.k2.hdfs.round = false
a1.sinks.k2.hdfs.rollInterval = 360
a1.sinks.k2.hdfs.rollSize = 1174405120
a1.sinks.k2.hdfs.rollCount = 0
a1.sinks.k2.hdfs.batchSize=3000
#控制输出文件类型
a1.sinks.k2.hdfs.fileType = CompressedStream
a1.sinks.k2.hdfs.codeC = gzip


#组装 
a1.sources.r1.channels = c1
a1.sinks.k1.channel = c1
a1.sinks.k2.channel = c1

#k1.batchSize+k2.batchSize < c1.capacity

启动命令

 nohup /data/module/flume-1.9.0/bin/flume-ng agent -Xms1024m -Xmx2048m -n a1 -c /data/module/flume-1.9.0/conf -f /data/module/flume-1.9.0/job/test.conf -Dflume.monitoring.type=http -Dflume.monitoring.port=36001  >/dev/null 2>&1 &