1

我一直在尝试将来自 kafka 的大量 json 消息(每个大约 2KB)推送到 cassandra 以触发流式传输。

模拟器---->Kafka---->SparkStreaming--->Cassandra。

它们中的每一个都在单独的 ec2 实例上运行,具有 30GB 的 Ram 和 8 核处理器作为独立的单节点设置。

当我试图从模拟器推送大约 500 万条消息时,在大约 100k 条消息之后,cassandra 停止插入消息,并且 spark 流式作业只是继续创建批处理(如 spark 流式 Web ui 中所示)。我什至检查了日志,但没有发现任何问题。

另外,我不确定我在代码中使用 spark 连接器写入 cassandra 的方式。

请看下面的代码,

/**
 * Spark Streaming to cassandra code
 */

package org.sparkexample;



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

import org.apache.spark.SparkConf;
import org.apache.spark.api.java.function.Function;
import org.apache.spark.streaming.Duration;
import org.apache.spark.streaming.api.java.JavaDStream;
import org.apache.spark.streaming.api.java.JavaPairReceiverInputDStream;
import org.apache.spark.streaming.api.java.JavaStreamingContext;
import org.apache.spark.streaming.kafka.KafkaUtils;

import com.datastax.spark.connector.japi.CassandraJavaUtil;
import com.datastax.spark.connector.japi.CassandraStreamingJavaUtil;

import scala.Tuple2;

public class SparkStreamingKafkaTest {


private SparkStreamingKafkaTest() {
}

public static void main(String[] args) {
    if (args.length < 6) {
        System.err.println("Usage: SparkStreamingKafka <zkQuorum> <group> <topics> <numThreads> <conc write> <cassandra ip>");
        System.exit(1);
    }

    SparkConf sparkConf = new SparkConf().setAppName("SparkStreamingKafka");




    //specific to cassandra

    sparkConf.set("spark.cassandra.output.concurrent.writes", args[4]);
    sparkConf.set("spark.cassandra.connection.host",args[5]);

    // Create the context with a 2 second batch size
    JavaStreamingContext jssc = new JavaStreamingContext(sparkConf, new Duration(2000));
    int numThreads = Integer.parseInt(args[3]);
    Map<String, Integer> topicMap = new HashMap<String, Integer>();
    String[] topics = args[2].split(",");
    for (String topic : topics) {
        topicMap.put(topic, numThreads);
    }

    JavaPairReceiverInputDStream<String, String> messages = KafkaUtils.createStream(jssc, args[0], args[1],
            topicMap);

    JavaDStream<WordCount> wc = messages.map(new Function<Tuple2<String, String>, WordCount>() {

        @Override
        public WordCount call(Tuple2<String, String> tuple2) {
            String key = System.currentTimeMillis()+ "_"+ Math.random();
            return new WordCount(key, tuple2._2());
        }
    });

    Map <String, String> map =  new HashMap<String, String>();
    map.put("word", "word");
    map.put("count", "count");

    CassandraStreamingJavaUtil.javaFunctions(wc).writerBuilder("mykeyspace", "wordcount",CassandraJavaUtil.mapToRow(WordCount.class, map)).saveToCassandra(); 

    jssc.start();
    jssc.awaitTermination();
  }


  }

WordCount.java

  package org.sparkexample;

  import java.io.Serializable;

  public class WordCount implements Serializable{

  private String word;
  private String count;

  public WordCount(){

  }

  public String getWord() {
    return word;
  }
  public void setWord(String word) {
    this.word = word;
  }
  public String getCount() {
    return count;
  }
  public void setCount(String count) {
    this.count = count;
  }

  public WordCount(String key, String count) {
    this.word = key;
    this.count = count;
  }
  }

我一直在使用具有以下主要依赖项的默认 cassandra.yml,

  • 火花-cassandra-connector_2.10 - 1.4.0-M3
  • spark-cassandra-connector-java_2.10 - 1.4.0-M3
  • cassandra 驱动程序核心 - 2.1.7.1
  • 火花流-kafka_2.10 - 1.4.1
  • 火花流_2.10 - 1.4.1
  • 火花核心_2.10 - 1.4.1

请提出可能是什么问题。

nodetool info 和 nodetool tpstats 的输出如下。

节点工具信息

节点工具 tpstat

4

0 回答 0