Spark：连接2个大型DFs时，大小超过Integer.max_value

Question

各位，

当我试图将spark中的两个大型数据流（每个100GB+)连接到每行一个密钥标识符时，我遇到了这个问题。

我在EMR上使用Spark1.6，下面是我正在做的事情：

val df1 = sqlContext.read.json("hdfs:///df1/")
val df2 = sqlContext.read.json("hdfs:///df2/")

// clean up and filter steps later 

df1.registerTempTable("df1")
df2.registerTempTable("df2")

val df3 = sql("select df1.*, df2.col1 from df1 left join df2 on df1.col3 = df2.col4")

df3.write.json("hdfs:///df3/")

java.lang.IllegalArgumentException: Size exceeds Integer.MAX_VALUE
    at sun.nio.ch.FileChannelImpl.map(FileChannelImpl.java:860)
    at org.apache.spark.storage.DiskStore$$anonfun$getBytes$2.apply(DiskStore.scala:127)
    at org.apache.spark.storage.DiskStore$$anonfun$getBytes$2.apply(DiskStore.scala:115)
    at org.apache.spark.util.Utils$.tryWithSafeFinally(Utils.scala:1250)
    at org.apache.spark.storage.DiskStore.getBytes(DiskStore.scala:129)
    at org.apache.spark.storage.DiskStore.getBytes(DiskStore.scala:136)
    at org.apache.spark.storage.BlockManager.doGetLocal(BlockManager.scala:503)
    at org.apache.spark.storage.BlockManager.getLocal(BlockManager.scala:420)
    at org.apache.spark.storage.BlockManager.get(BlockManager.scala:625)
    at org.apache.spark.CacheManager.getOrCompute(CacheManager.scala:44)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:268)
    at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
    at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
    at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
    at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:73)
    at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:41)
    at org.apache.spark.scheduler.Task.run(Task.scala:89)
    at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:213)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
    at java.lang.Thread.run(Thread.java:745)

null

我也尝试过使用更大的集群，但没有帮助。此链接表示，如果洗牌分区大小超过2GB，将引发此错误。但是我尝试将分区数量增加到一个很高的值，仍然没有运气。

我怀疑这可能与懒惰加载有关。当我在一个DF上做10个操作时，它们只在最后一步执行。我尝试在不同的DFs存储级别上添加.persist()，但没有成功。我还尝试删除临时表，清空所有早期的DFs以进行清理。

然而，如果我将代码分解为两个部分，代码就可以工作了--将最后的临时数据（2个数据帧）写入磁盘，退出。重新启动以只联接两个DF。

Exception in thread "main" java.util.concurrent.TimeoutException: Futures timed out after [300 seconds]
    at scala.concurrent.impl.Promise$DefaultPromise.ready(Promise.scala:219)
    at scala.concurrent.impl.Promise$DefaultPromise.result(Promise.scala:223)
    at scala.concurrent.Await$$anonfun$result$1.apply(package.scala:107)
    at scala.concurrent.BlockContext$DefaultBlockContext$.blockOn(BlockContext.scala:53)
    at scala.concurrent.Await$.result(package.scala:107)
    at org.apache.spark.sql.execution.joins.BroadcastHashOuterJoin.doExecute(BroadcastHashOuterJoin.scala:113)
    at org.apache.spark.sql.execution.SparkPlan$$anonfun$execute$5.apply(SparkPlan.scala:132)
    at org.apache.spark.sql.execution.SparkPlan$$anonfun$execute$5.apply(SparkPlan.scala:130)
    at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:150)
    at org.apache.spark.sql.execution.SparkPlan.execute(SparkPlan.scala:130)
    at org.apache.spark.sql.execution.Project.doExecute(basicOperators.scala:46)
    at org.apache.spark.sql.execution.SparkPlan$$anonfun$execute$5.apply(SparkPlan.scala:132)
    at org.apache.spark.sql.execution.SparkPlan$$anonfun$execute$5.apply(SparkPlan.scala:130)
    at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:150)
    at org.apache.spark.sql.execution.SparkPlan.execute(SparkPlan.scala:130)
    at org.apache.spark.sql.execution.QueryExecution.toRdd$lzycompute(QueryExecution.scala:55)
    at org.apache.spark.sql.execution.QueryExecution.toRdd(QueryExecution.scala:55)
    at org.apache.spark.sql.DataFrame.toJSON(DataFrame.scala:1724)
    at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
    at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:57)
    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
    at java.lang.reflect.Method.invoke(Method.java:606)
    at org.apache.spark.deploy.SparkSubmit$.org$apache$spark$deploy$SparkSubmit$$runMain(SparkSubmit.scala:731)
    at org.apache.spark.deploy.SparkSubmit$.doRunMain$1(SparkSubmit.scala:181)
    at org.apache.spark.deploy.SparkSubmit$.submit(SparkSubmit.scala:206)
    at org.apache.spark.deploy.SparkSubmit$.main(SparkSubmit.scala:121)
    at org.apache.spark.deploy.SparkSubmit.main(SparkSubmit.scala)

Answer 1

在spark u中，洗牌块不能大于2GB。这是因为，Spark存储将块洗牌为ByteBuffer。以下是如何分配它：

ByteBuffer.Allocate(int ；容量）

因为ByteBuffer受integer.max_size(2GB)的限制，shuffle块也是如此！！解决方案是通过在SparkSQL中使用spark.sql.shuffle.partitions，或者通过对RDD使用rdd.partition()或rdd.colease()来增加分区的数量，这样每个分区的大小都<=2GB。