我正在尝试使用 Spark 在 S3 事件上创建一个简单的 sql 查询。我正在加载 ~30GB 的 JSON 文件,如下所示:
val d2 = spark.read.json("s3n://myData/2017/02/01/1234");
d2.persist(org.apache.spark.storage.StorageLevel.MEMORY_AND_DISK);
d2.registerTempTable("d2");
然后我试图写入文件我的查询结果:
val users_count = sql("select count(distinct data.user_id) from d2");
users_count.write.format("com.databricks.spark.csv").option("header", "true").save("s3n://myfolder/UsersCount.csv");
但 Spark 抛出以下异常:
java.lang.IllegalArgumentException: Size exceeds Integer.MAX_VALUE
at sun.nio.ch.FileChannelImpl.map(FileChannelImpl.java:869)
at org.apache.spark.storage.DiskStore$$anonfun$getBytes$2.apply(DiskStore.scala:103)
at org.apache.spark.storage.DiskStore$$anonfun$getBytes$2.apply(DiskStore.scala:91)
at org.apache.spark.util.Utils$.tryWithSafeFinally(Utils.scala:1287)
at org.apache.spark.storage.DiskStore.getBytes(DiskStore.scala:105)
at org.apache.spark.storage.BlockManager.getLocalValues(BlockManager.scala:439)
at org.apache.spark.storage.BlockManager.getOrElseUpdate(BlockManager.scala:672)
at org.apache.spark.rdd.RDD.getOrCompute(RDD.scala:330)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:281)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:319)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:283)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:319)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:283)
at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:319)
at org.apache.spark.rdd.RDD.iterator(RDD.scala:283)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:79)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:47)
at org.apache.spark.scheduler.Task.run(Task.scala:85)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:274)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
at java.lang.Thread.run(Thread.java:745)
请注意,相同的查询适用于更少量的数据。这里有什么问题?
最佳答案
没有 Spark shuffle 块可以大于 2GB(Integer.MAX_VALUE 字节),因此您需要更多/更小的分区。
您应该调整 spark.default.parallelism 和 spark.sql.shuffle.partitions(默认为 200),以便分区数量可以容纳您的数据而不会达到 2GB 的限制(您可以尝试瞄准 256MB/分区,因此对于 200GB,您将获得 800分区)。数千个分区很常见,所以不要害怕按照建议重新分区到 1000 个。
仅供引用,您可以使用 rdd.getNumPartitions(即 d2.rdd.getNumPartitions)之类的内容检查 RDD 的分区数
有一个故事来跟踪解决各种 2GB 限制(现已开放一段时间)的努力:https://issues.apache.org/jira/browse/SPARK-6235
见 http://www.slideshare.net/cloudera/top-5-mistakes-to-avoid-when-writing-apache-spark-applications/25有关此错误的更多信息。
关于Spark/scala 中的 SQL 查询大小超过 Integer.MAX_VALUE,我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/42247630/