问题导读:
1、用什么方式判断空RDD?
2、Spark Streaming与Kafka如何处理空RDD?
在Spark Streaming中,job不断的产生,有时候会产生一些空RDD,而基于这些空RDD生成的job大多数情况下是没必要提交到集群执行的。执行没有结果的job,就是浪费计算资源,数据库连接资源,产生空文件等。
这里介绍两种判断空RDD的方式,第一种是以Receiver接收数据时产生的BlockRDD或WriteAheadLogBackedBlockRDD,所有以Receiver方式接收数据都会产生BlockRDD或WriteAheadLogBackedBlockRDD,第二种是以Direct Kafka方式接收数据产生的KafkaRDD。
第一种情况
以Receiver方式接收数据,计算wordCount为例来说明空RDD如何处理,代码如下
[mw_shl_code=applescript,true]object ReceiverWordCount {
def main(args: Array[String]) {
val conf = new SparkConf().setAppName("ReceiverWordCount").setMaster("local[3]")
conf.set("spark.testing.memory", "2147480000")
val ssc = new StreamingContext(conf, Seconds(10))
val lines = ssc.socketTextStream("10.10.63.106", 8589, StorageLevel.MEMORY_AND_DISK_SER)
val words= lines.flatMap(_.split(""))
val wordCounts= words.map(x => (x,1)).reduceByKey((num1:Int,num2:Int)=>num1+num2,2)
wordCounts.foreachRDD(rdd=>{
if(rdd.dependencies(0).rdd.partitions.isEmpty){
println(">>>RDD:Empty")
}else{
rdd.foreach(x=>println(x._1+"\t"+x._2))
}
})
ssc.start()
ssc.awaitTermination()
}
}[/mw_shl_code]
这里为了方便,在foreachRDD中使用了rdd.foreach(x=>println(x._1+"\t"+x._2))来打印结果,只是简单的效果演示,生产环境一般会输出到外部存储系统中,例如mysql、redis 、hdfs等
这里总结了三种判断空RDD方式的,我们来看一下这三种方式有什么不同:
第一种:if(rdd.count==0)
RDD的count操作会触发一个action,提交一个job,这种方式不是我们想要的
第二种:if(rdd.partitions.isEmpty)
判断rdd的partitions是否为空,那我们需要看一下这里的rdd是怎么得来的,经过上面WordCount中的一系列transformation操作后,最后一个reduceByKey操作产生的ShuffledRDD 。经过reduceByKey操作后,分区数量会受到默认分区数或用户指定的分区数的影响,和最初BlockRDD的分区数不一样,因为ShuffledRDD的分区数不可能为0,所以if(rdd.partitions.isEmpty)无效。if(rdd.partitions.isEmpty)在什么有效呢?只有在当前rdd和BlockRDD在同一个stage时才会有效,因为分区数没有变化
第三种:if(rdd.dependencies(0).rdd.partitions.isEmpty)
根据RDD的依赖关系,从后向前寻找BlockRDD,因为在BlockRDD生成的时候分区数受blockInfos(Receiver接收数据的元数据信息)的影响,代码如下
[mw_shl_code=applescript,true]private[streaming] def createBlockRDD(time: Time, blockInfos: Seq[ReceivedBlockInfo]): RDD[T] = {
if (blockInfos.nonEmpty) {
val blockIds = blockInfos.map { _.blockId.asInstanceOf[BlockId] }.toArray
// Are WAL record handles present with all the blocks
val areWALRecordHandlesPresent = blockInfos.forall { _.walRecordHandleOption.nonEmpty }
if (areWALRecordHandlesPresent) {
// If all the blocks have WAL record handle, then create a WALBackedBlockRDD
val isBlockIdValid = blockInfos.map { _.isBlockIdValid() }.toArray
val walRecordHandles = blockInfos.map { _.walRecordHandleOption.get }.toArray
new WriteAheadLogBackedBlockRDD[T](ssc.sparkContext, blockIds, walRecordHandles, isBlockIdValid)
} else {
// Else, create a BlockRDD. However, if there are some blocks with WAL info but not
// others then that is unexpected and log a warning accordingly.
if (blockInfos.find(_.walRecordHandleOption.nonEmpty).nonEmpty) {
if (WriteAheadLogUtils.enableReceiverLog(ssc.conf)) {
logError("Some blocks do not have Write Ahead Log information; this is unexpected and data may not be recoverable after driver failures")
} else {
logWarning("Some blocks have Write Ahead Log information; this is unexpected")
}
}
val validBlockIds = blockIds.filter { id => ssc.sparkContext.env.blockManager.master.contains(id) }
if (validBlockIds.size != blockIds.size) {
logWarning("Some blocks could not be recovered as they were not found in memory. " +
"To prevent such data loss, enabled Write Ahead Log (see programming guide " +
"for more details.")
}
new BlockRDD[T](ssc.sc, validBlockIds)
}
} else {
// If no block is ready now, creating WriteAheadLogBackedBlockRDD or BlockRDD
// according to the configuration
if (WriteAheadLogUtils.enableReceiverLog(ssc.conf)) {
new WriteAheadLogBackedBlockRDD[T](
ssc.sparkContext, Array.empty, Array.empty, Array.empty)
} else {
new BlockRDD[T](ssc.sc, Array.empty)
}
}
}[/mw_shl_code]
如果blockInfos为空,BlockRDD的分区数也为空,所以要判断BlockRDD的分区数。这里只判断了当前rdd的父RDD分区是否为空,因为父RDD和BlockRDD在同一个stage内,分区数是一致的。RDD的依赖关系可以通过rdd.toDebugString和web页面获得,stage划分也可以通过web页面获得。
第二种情况
以Direct kafka的方式接收数据的方式,计算WordCount为例,代码如下
[mw_shl_code=applescript,true]object DirectKafkaDemo{
def main(args: Array[String]) {
val topics = "DirectKafkaDemo"
val brokers = "*:9092,*:9092"
val sparkConf = new SparkConf().setAppName("DirectKafkaWordCount").setMaster("local[2]")
sparkConf.set("spark.testing.memory", "2147480000")
val ssc = new StreamingContext(sparkConf, Seconds(10))
val topicsSet = topics.split(",").toSet
val kafkaParams = Map[String, String]("metadata.broker.list" -> brokers)
val messages = KafkaUtils.createDirectStream[String, String, StringDecoder, StringDecoder](
ssc, kafkaParams, topicsSet)
val result = messages.map(_._2).flatMap(_.split(" ")).map((_, 1)).reduceByKey(_ + _)
result.foreachRDD(rdd => {
val num= rdd.dependencies(0).rdd.dependencies(0).rdd.dependencies(0).rdd.dependencies(0).rdd.count()
if(num>0) {
rdd.foreachPartition(data => {
val conn = MDBManager.getConnection
conn.setAutoCommit(false)
val sql = "insert into word set key1=?,num=?;"
val preparedStatement = conn.prepareStatement(sql)
data.foreach(recode => {
val key = recode._1;
val num = recode._2;
preparedStatement.setString(1, key)
preparedStatement.setInt(2, num)
preparedStatement.addBatch()
println("key:" + key + "\tnum:" + num)
})
preparedStatement.executeBatch()
conn.commit()
conn.close()
})
}else{
println(">>>>>>>>>>>>>>>>>>>>>>RDD Empty")
}
})
ssc.start()
ssc.awaitTermination()
}
}[/mw_shl_code]
这里使用了KafkaRDD的count操作来判断KafkaRDD是否为空,如果不为空,将计算结果保存到数据库中,减少不必要是数据库操作。获取KafkaRDD的代码如下,不同代码编写RDD的依赖关系是不一样的,要根据代码而定
[mw_shl_code=applescript,true]val num= rdd.dependencies(0).rdd.dependencies(0).rdd.dependencies(0).rdd.dependencies(0).rdd.count()[/mw_shl_code]
看一下KafkaRDD的count()方法,他重写了RDD的count方法,代码如下
[mw_shl_code=applescript,true]override def count(): Long = offsetRanges.map(_.count).sum[/mw_shl_code]
他并没有触发一个runJob操作,而是通过读取kafka分区的offset偏移量来计算RDD记录的个数,这里是利用了kafka的特性。通过依赖关系找到KafkaRDD,然后调用KafkaRDD的count()方法,就知道KafkaRDD是否为空,如果KafkaRDD为空,就没必要runJob了。
那么判断KafkaRDD的分区数是否也可以,看一下KafkaRDD的分区数是怎么得来的,代码如下
[mw_shl_code=applescript,true]override def getPartitions: Array[Partition] = {
offsetRanges.zipWithIndex.map { case (o, i) =>
val (host, port) = leaders(TopicAndPartition(o.topic, o.partition))
new KafkaRDDPartition(i, o.topic, o.partition, o.fromOffset, o.untilOffset, host, port)
}.toArray
}[/mw_shl_code]
和offsetRanges的数量有关,因为offsetRanges是根据kafka的分区数而来,offsetRanges的数量是固定不变的,从而KafkaRDD的分区数是固定的,不管分区有没有数据,因此不能判断KafkaRDD的分区数
总结
不同数据接收方式的RDD,表现数据为空都可能是不一样的,通过RDD的依赖关系正确找到数据源RDD是最关键的。此方法使用一定要结合业务和RDD的具体生成方式,这里说的依赖关系都是之有一个父RDD,如果有多个父RDD要根据情况决定是否可以使用此方法。
来源:海纳百川
|
|