Java 在Spark Executors上向Kafka提交偏移量
我从卡夫卡那里获取事件,在Spark上丰富/过滤/转换它们,然后将它们存储在ES中。我要把补偿交还给卡夫卡 我有两个问题: (1)我目前的Spark工作非常缓慢 我有50个主题分区和20个执行器。每个执行器都有2个内核和4g内存。我的司机有8g内存。我每秒消耗1000个事件/分区,批处理间隔为10秒。这意味着,我在10秒内消耗了500000个事件 我的ES群集如下所示: 20分片/索引 3个主实例c5.xlarge.elasticsearch 12个实例m4.xlarge.elasticsearch 磁盘/节点=1024 GB,总共12 TB 我得到了巨大的日程安排和处理延迟 (2)如何在执行器上提交偏移量? 目前,我在executors上丰富/转换/过滤我的事件,然后使用BulkRequest将所有内容发送到ES。这是一个同步过程。如果我得到肯定的反馈,我会将偏移列表发送给驱动程序。如果没有,我会发回一个空列表。在驱动程序上,我向卡夫卡提交偏移量。我相信,应该有一种方法,我可以在执行者身上提交补偿,但我不知道如何将卡夫卡流传递给执行者:Java 在Spark Executors上向Kafka提交偏移量,java,apache-spark,
elasticsearch,apache-kafka,Java,Apache Spark,
elasticsearch,Apache Kafka,我从卡夫卡那里获取事件,在Spark上丰富/过滤/转换它们,然后将它们存储在ES中。我要把补偿交还给卡夫卡 我有两个问题: (1)我目前的Spark工作非常缓慢 我有50个主题分区和20个执行器。每个执行器都有2个内核和4g内存。我的司机有8g内存。我每秒消耗1000个事件/分区,批处理间隔为10秒。这意味着,我在10秒内消耗了500000个事件 我的ES群集如下所示: 20分片/索引 3个主实例c5.xlarge.elasticsearch 12个实例m4.xlarge.elasticsear
((CanCommitOffsets) kafkaStream.inputDStream()).commitAsync(offsetRanges, this::onComplete);
这是向Kafka提交偏移量的代码,它需要Kafka流
以下是我的总体代码:
kafkaStream.foreachRDD( // kafka topic
rdd -> { // runs on driver
rdd.cache();
String batchIdentifier =
Long.toHexString(Double.doubleToLongBits(Math.random()));
LOGGER.info("@@ [" + batchIdentifier + "] Starting batch ...");
Instant batchStart = Instant.now();
List<OffsetRange> offsetsToCommit =
rdd.mapPartitionsWithIndex( // kafka partition
(index, eventsIterator) -> { // runs on worker
OffsetRange[] offsetRanges = ((HasOffsetRanges) rdd.rdd()).offsetRanges();
LOGGER.info(
"@@ Consuming " + offsetRanges[index].count() + " events" + " partition: " + index
);
if (!eventsIterator.hasNext()) {
return Collections.emptyIterator();
}
// get single ES documents
List<SingleEventBaseDocument> eventList = getSingleEventBaseDocuments(eventsIterator);
// build request wrappers
List<InsertRequestWrapper> requestWrapperList = getRequestsToInsert(eventList, offsetRanges[index]);
LOGGER.info(
"@@ Processed " + offsetRanges[index].count() + " events" + " partition: " + index + " list size: " + eventList.size()
);
BulkResponse bulkItemResponses = elasticSearchRepository.addElasticSearchDocumentsSync(requestWrapperList);
if (!bulkItemResponses.hasFailures()) {
return Arrays.asList(offsetRanges).iterator();
}
elasticSearchRepository.close();
return Collections.emptyIterator();
},
true
).collect();
LOGGER.info(
"@@ [" + batchIdentifier + "] Collected all offsets in " + (Instant.now().toEpochMilli() - batchStart.toEpochMilli()) + "ms"
);
OffsetRange[] offsets = new OffsetRange[offsetsToCommit.size()];
for (int i = 0; i < offsets.length ; i++) {
offsets[i] = offsetsToCommit.get(i);
}
try {
offsetManagementMapper.commit(offsets);
} catch (Exception e) {
// ignore
}
LOGGER.info(
"@@ [" + batchIdentifier + "] Finished batch of " + offsetsToCommit.size() + " messages " +
"in " + (Instant.now().toEpochMilli() - batchStart.toEpochMilli()) + "ms"
);
rdd.unpersist();
});
kafkaStream.foreachRDD(//kafka topic
rdd->{//在驱动程序上运行
缓存();
字符串批处理标识符=
Long.toHexString(Double.Double-tolongbits(Math.random());
LOGGER.info(“@[”+batchIdentifier+“]开始批处理…”);
Instant batchStart=Instant.now();
列出offsetsToCommit=
rdd.mapPartitionsWithIndex(//kafka分区
(索引,eventsIterator)->{//在辅助服务器上运行
OffsetRange[]offsetRanges=((HasOffsetRanges)rdd.rdd()).offsetRanges();
LOGGER.info(
“@@consing”+offsetRanges[index]。count()+“events”+“partition:”+index
);
如果(!eventsIterator.hasNext()){
返回集合。emptyIterator();
}
//获取单个ES文档
List eventList=getSingleEventBaseDocuments(eventsIterator);
//构建请求包装器
List requestWrapperList=getRequestsToInsert(eventList,offsetRanges[index]);
LOGGER.info(
“@@Processed”+offsetRanges[index].count()+“events”+“partition:“+index+”列表大小:“+eventList.size()
);
BulkResponse bulkItemResponses=elasticSearchRepository.addElasticSearchDocumentsSync(requestWrapperList);
如果(!bulkItemResponses.hasFailures()){
返回Arrays.asList(offsetRanges.iterator();
}
elasticSearchRepository.close();
返回集合。emptyIterator();
},
真的
).收集();
LOGGER.info(
“@@[“+batchIdentifier+”]收集了“+(Instant.now().toEpochMilli()-batchStart.toEpochMilli())+”ms中的所有偏移量”
);
OffsetRange[]offsets=新的OffsetRange[offsetsToCommit.size()];
对于(int i=0;i
您可以将偏移逻辑移到rdd循环上方。。。我使用下面的模板来更好地处理偏移和性能
JavaInputDStream<ConsumerRecord<String, String>> kafkaStream = KafkaUtils.createDirectStream(jssc,
LocationStrategies.PreferConsistent(),
ConsumerStrategies.<String, String>Subscribe(topics, kafkaParams));
kafkaStream.foreachRDD( kafkaStreamRDD -> {
//fetch kafka offsets for manually commiting it later
OffsetRange[] offsetRanges = ((HasOffsetRanges) kafkaStreamRDD.rdd()).offsetRanges();
//filter unwanted data
kafkaStreamRDD.filter(
new Function<ConsumerRecord<String, String>, Boolean>() {
@Override
public Boolean call(ConsumerRecord<String, String> kafkaRecord) throws Exception {
if(kafkaRecord!=null) {
if(!StringUtils.isAnyBlank(kafkaRecord.key() , kafkaRecord.value())) {
return Boolean.TRUE;
}
}
return Boolean.FALSE;
}
}).foreachPartition( kafkaRecords -> {
// init connections here
while(kafkaRecords.hasNext()) {
ConsumerRecord<String, String> kafkaConsumerRecord = kafkaRecords.next();
// work here
}
});
//commit offsets
((CanCommitOffsets) kafkaStream.inputDStream()).commitAsync(offsetRanges);
});
JavaInputDStream kafkaStream=KafkaUtils.createDirectStream(jssc,
LocationStrategies.PreferConsistent(),
订阅(主题,卡夫卡帕兰));
kafkaStream.foreachRDD(kafkaStreamRDD->{
//获取卡夫卡偏移量,以便稍后手动提交
OffsetRange[]offsetRanges=((HasOffsetRanges)kafkaStreamRDD.rdd()).offsetRanges();
//过滤不需要的数据
kafkaStreamRDD.filter(
新函数(){
@凌驾
公共布尔调用(ConsumerRecord kafkaRecord)引发异常{
if(kafkaRecord!=null){
如果(!StringUtils.isAnyBlank(kafkaRecord.key(),kafkaRecord.value())){
返回Boolean.TRUE;
}
}
返回Boolean.FALSE;
}
}).foreachPartition(卡夫卡雷德记录->{
//在这里初始化连接
while(kafkaRecords.hasNext()