[英]Closing Spark Streaming Context after first batch (trying to retrieve kafka offsets)
我正在嘗試為我的Spark Batch作業檢索Kafka偏移量。 檢索偏移量后,我想關閉流上下文。
我嘗試將streamlistener添加到流上下文中,並實現onBatchCompleted方法以在作業完成后關閉流,但是我收到異常“無法在偵聽器總線線程內停止StreamingContext” 。
有針對這個的解決方法嗎? 我正在嘗試檢索偏移量以調用KafkaUtils.createRDD(sparkContext,kafkaProperties,OffsetRange [],LocationStrateg)
private OffsetRange[] getOffsets(SparkConf sparkConf) throws InterruptedException {
final AtomicReference<OffsetRange[]> atomicReference = new AtomicReference<>();
JavaStreamingContext sc = new JavaStreamingContext(sparkConf, Duration.apply(50));
JavaInputDStream<ConsumerRecord<String, String>> stream =
KafkaUtils.createDirectStream(sc, LocationStrategies.PreferConsistent(), ConsumerStrategies.<String, String>Subscribe(Arrays.asList("test"), getKafkaParam()));
stream.foreachRDD((VoidFunction<JavaRDD<ConsumerRecord<String, String>>>) rdd -> {
atomicReference.set(((HasOffsetRanges) rdd.rdd()).offsetRanges());
// sc.stop(false); //this would throw exception saying consumer is already closed
}
);
sc.addStreamingListener(new TopicListener(sc)); //Throws exception saying "Cannot stop StreamingContext within listener bus thread."
sc.start();
sc.awaitTermination();
return atomicReference.get();
}
public class TopicListener implements StreamingListener {
private JavaStreamingContext sc;
public TopicListener(JavaStreamingContext sc){
this.sc = sc;
}
@Override
public void onBatchCompleted(StreamingListenerBatchCompleted streamingListenerBatchCompleted) {
sc.stop(false);
}
非常感謝stackoverflow-ers :)我已經嘗試搜索可能的解決方案,但到目前為止尚未成功
編輯 :我用KafkaConsumer來獲取分區信息。 一旦獲得分區信息,就創建一個TopicPartition pojos列表,並調用position和endOffsets方法分別獲取我的groupId的當前位置和結束位置。
final List<PartitionInfo> partitionInfos = kafkaConsumer.partitionsFor("theTopicName");
final List<TopicPartition> topicPartitions = new ArrayList<>();
partitionInfos.forEach(partitionInfo -> topicPartitions.add(new TopicPartition("theTopicName", partitionInfo.partition())));
final List<OffsetRange> offsetRanges = new ArrayList<>();
kafkaConsumer.assign(topicPartitions);
topicPartitions.foreach(topicPartition -> {
long fromOffset = kafkaConsumer.position(topicPartition);
kafkaConsumer.seekToEnd(Collections.singleton(topicPartition));
long untilOffset = kafkaConsumer.position(topicPartition);
offsetRanges.add(new OffsetRange(topicPartition.topic(), topicPartition.partition(), fromOffset, untilOffset));
});
return offsetRanges.toArray(new OffsetRange[offsetRanges.size()]);
如果要控制流,則可以考慮使用輪詢而不是流式API。 這樣一來,您就可以在達到目標后清楚地停止輪詢。
還檢查一下...
聲明:本站的技術帖子網頁,遵循CC BY-SA 4.0協議,如果您需要轉載,請注明本站網址或者原文地址。任何問題請咨詢:yoyou2525@163.com.