Spark——Structured Streaming 监控Kafka消费进度

前言

Structured Streaming 消费 Kafka 时并不会将 Offset 提交到 Kafka 集群，本文介绍利用 StreamingQueryListener 间接实现对 Kafka 消费进度的监控。

基于StreamingQueryListener向Kafka提交Offset

监听StreamingQuery各种事件的接口，如下：

public abstract class StreamingQueryListener {

    public abstract void onQueryStarted(StreamingQueryListener.QueryStartedEvent var1);

    public abstract void onQueryProgress(StreamingQueryListener.QueryProgressEvent var1);

    public abstract void onQueryTerminated(StreamingQueryListener.QueryTerminatedEvent var1);

}

在QueryProgressEvent中，我们是可以拿到每个Source消费的Offset的。因此，基于StreamingQueryListener，可以将消费的offset的提交到kafka集群，进而实现对Kafka Lag的监控。

KafkaOffsetCommiter

package com.trace.spark;

import com.alibaba.fastjson.JSONObject;
import org.apache.kafka.clients.consumer.ConsumerConfig;
import org.apache.kafka.clients.consumer.KafkaConsumer;
import org.apache.kafka.clients.consumer.OffsetAndMetadata;
import org.apache.kafka.common.TopicPartition;
import org.apache.spark.sql.streaming.SourceProgress;
import org.apache.spark.sql.streaming.StreamingQueryListener;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import java.util.HashMap;
import java.util.Map;
import java.util.Properties;

public class CustomStreamingQueryListener extends StreamingQueryListener {

    /**
     * logger.
     */
    private static final Logger logger = LoggerFactory.getLogger(CustomStreamingQueryListener.class);

    /**
     * 初始化KafkaConsumer.
     */
    private static KafkaConsumer kafkaConsumer = null;
    Properties properties= new Properties();
    {
        properties.put(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG, "172.16.10.91:9092,172.16.10.92:9092,172.16.10.93:9092");
        properties.put(ConsumerConfig.GROUP_ID_CONFIG, "sparkStructuredStreaming");
        properties.put(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG, "org.apache.kafka.common.serialization.StringDeserializer");
        properties.put(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG, "org.apache.kafka.common.serialization.StringDeserializer");
        kafkaConsumer = new KafkaConsumer(properties);
    }

    @Override
    public void onQueryStarted(QueryStartedEvent event) {
        logger.info("Started query with id : {}, name: {},runId : {}", event.id(), event.name(), event.runId().toString());
    }

    @Override
    public void onQueryProgress(QueryProgressEvent event) {
        logger.info("Streaming query made progress: {}", event.progress().prettyJson());
        for (SourceProgress sourceProgress : event.progress().sources()) {
            Map<String, Map<String, String>> endOffsetMap = JSONObject.parseObject(sourceProgress.endOffset(), Map.class);
            for (String topic : endOffsetMap.keySet()) {
                Map<TopicPartition, OffsetAndMetadata> topicPartitionsOffset = new HashMap<>();
                Map<String, String> partitionMap = endOffsetMap.get(topic);
                for (String partition : partitionMap.keySet()) {
                    TopicPartition topicPartition = new TopicPartition(topic, Integer.parseInt(partition));
                    OffsetAndMetadata offsetAndMetadata = new OffsetAndMetadata(Long.parseLong(String.valueOf(partitionMap.get(partition))));
                    topicPartitionsOffset.put(topicPartition, offsetAndMetadata);
                }
                logger.info("【commitSync offset】topicPartitionsOffset={}", topicPartitionsOffset);
                kafkaConsumer.commitSync(topicPartitionsOffset);
            }
        }
    }

    @Override
    public void onQueryTerminated(QueryTerminatedEvent event) {
        logger.info("Stream exited due to exception : {}, id : {}, runId: {}", event.exception().toString(), event.id(), event.runId());
    }

}

Structured Streaming App

package com.trace.spark;

import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.apache.spark.api.java.function.FlatMapFunction;
import org.apache.spark.sql.Dataset;
import org.apache.spark.sql.Encoders;
import org.apache.spark.sql.Row;
import org.apache.spark.sql.SparkSession;
import org.apache.spark.sql.functions;
import org.apache.spark.sql.streaming.StreamingQuery;
import org.springframework.util.StringUtils;

import com.alibaba.fastjson.JSONObject;
import com.google.common.collect.Lists;
import com.trace.spark.function.CleanDataFunction;
import com.trace.spark.function.HbaseForeachWriter;
import com.trace.spark.function.HbaseHelper;
import com.trace.util.KafkaOpsUtils;
import com.trace.util.TraceUtils;

import scala.Tuple3;

public class SparkKafkaManager {

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

    public static AtomicLong atomic = new AtomicLong(1);

    public static void main(String[] args) throws Exception {
        String brokers = "172.16.10.91:9092,172.16.10.92:9092,172.16.10.93:9092";
        final SparkSession spark = SparkSession.builder().appName("TimeWindow").getOrCreate();
        spark.udf().register("cleanData", new CleanDataFunction());
        spark.streams().addListener(new CustomStreamingQueryListener());
        Dataset<Row> lines = spark
                .readStream()
                .format("kafka")
                .option("subscribe", KafkaOpsUtils.trace_topic_name)
                .option("kafka.bootstrap.servers", brokers)
                .load();

省略后续代码...................

查看Kafka Offset

其它

还有以下两种方式也可以实现监控效果：

• Checkpoint
• StreamingQuery API

引用：

• https://blog.csdn.net/wangpei1949/article/details/105187210
• Structured Streaming Kafka Source Offset Storage
• Spark Streaming 管理 Kafka Offsets 的方式探讨