scala - 连接 2 个大型 DF 时 Spark : Size exceeds Integer. MAX_VALUE

Question

各位，

当我尝试在 Spark 中每行一个键标识符加入 2 个大型数据帧(每个数据帧 100GB +)时，我遇到了这个问题。

我在 EMR 上使用 Spark 1.6，这就是我正在做的事情:

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/")

这基本上是我正在做的事情的要点，以及最终加入 df1 和 df2 之间的其他清理和过滤步骤。

我看到的错误是:

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)

配置和引用:
我使用 13 个节点，每个集群 60GB，执行器和驱动程序内 stub 据开销进行相应设置。我尝试调整的事情:

• spark.sql.broadcastTimeout
• spark.sql.shuffle.partitions

我也尝试过使用更大的集群，但没有帮助。 This link表示如果 Shuffle 分区大小超过 2GB，则会引发此错误。但我尝试将分区数量增加到很高的值，但仍然没有成功。

我怀疑这可能与延迟加载有关。当我对 DF 执行 10 次操作时，它们仅在最后一步执行。我尝试在 DF 的各种存储级别上添加 .persist() ，但仍然没有成功。我还尝试删除临时表，清空所有早期的 DF 进行清理。

但是，如果我将代码分为两部分，则代码可以工作 - 将最终的临时数据(2 个数据帧)写入磁盘，然后退出。重新启动以仅加入两个 DF。

我之前收到此错误:

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)

但是当我调整 spark.sql.broadcastTimeout 时，我开始收到第一个错误。

在这种情况下，我们将不胜感激。如果需要，我可以添加更多信息。

Answer 1

在 Spark 中，shuffle block 不能大于 2GB。这是因为，Spark 将随机 block 存储为 ByteBuffer 。以下是分配方式:

ByteBuffer.allocate(int capacity)

由于 ByteBuffer 受到 Integer.MAX_SIZE (2GB) 的限制，shuffle block 也是如此!解决方案是使用 spark.sql.shuffle.partitions 增加分区数量在 SparkSQL 中或通过 rdd.partition() or rdd.colease()对于 rdd，每个分区大小 <= 2GB。

您提到您尝试增加分区数量但仍然失败。您可以检查一下分区大小是否> 2GB。只需确保指定的分区数量足以使每个 block 大小 < 2GB