EMR 5.28 无法在 s3 上加载镶木地板文件

Question

在 EMR 集群 5.28.0 上，从 s3 读取镶木地板文件失败，出现以下异常，而在 EMR 5.18.0 上，同样工作正常。 以下是 EMR 5.28.0 上的堆栈跟踪。

我什至从spark-shell尝试：

sqlContext.read.load(("s3://s3_file_path/*")
df.take(5) 

但失败并出现相同的异常：

Job aborted due to stage failure: Task 3 in stage 1.0 failed 4 times, most recent failure: Lost task 3.3 in stage 1.0 (TID 17, ip-x.x.x.x.ec2.internal, executor 1): **org.apache.spark.sql.execution.datasources.FileDownloadException: Failed to download file path: s3://somedir/somesubdir/434560/1658_1564419581.parquet, range: 0-7928, partition values: [empty row], isDataPresent: false**
at org.apache.spark.sql.execution.datasources.AsyncFileDownloader.next(AsyncFileDownloader.scala:142)
at org.apache.spark.sql.execution.datasources.FileScanRDD$$anon$1.getNextFile(FileScanRDD.scala:241)
at org.apache.spark.sql.execution.datasources.FileScanRDD$$anon$1.nextIterator(FileScanRDD.scala:171)
at org.apache.spark.sql.execution.datasources.FileScanRDD$$anon$1.hasNext(FileScanRDD.scala:130)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage1.scan_nextBatch_0$(Unknown Source)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage1.processNext(Unknown Source)
at org.apache.spark.sql.execution.BufferedRowIterator.hasNext(BufferedRowIterator.java:43)
at org.apache.spark.sql.execution.WholeStageCodegenExec$$anonfun$13$$anon$1.hasNext(WholeStageCodegenExec.scala:636)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:409)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:409)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:409)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:409)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:409)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage1.agg_doAggregateWithKeys_0$(Unknown Source)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage1.processNext(Unknown Source)
at org.apache.spark.sql.execution.BufferedRowIterator.hasNext(BufferedRowIterator.java:43)
at org.apache.spark.sql.execution.WholeStageCodegenExec$$anonfun$13$$anon$1.hasNext(WholeStageCodegenExec.scala:636)
at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:409)
at org.apache.spark.shuffle.sort.BypassMergeSortShuffleWriter.write(BypassMergeSortShuffleWriter.java:125)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:99)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:55)
at org.apache.spark.scheduler.Task.run(Task.scala:123)
at org.apache.spark.executor.Executor$TaskRunner$$anonfun$10.apply(Executor.scala:408)
at org.apache.spark.util.Utils$.tryWithSafeFinally(Utils.scala:1360)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:414)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
at java.lang.Thread.run(Thread.java:748)
**Caused by: java.lang.NullPointerException
at org.apache.spark.sql.execution.datasources.parquet.ParquetFileFormat.org$apache$spark$sql$execution$datasources$parquet$ParquetFileFormat$$isCreatedByParquetMr(ParquetFileFormat.scala:352)
at** org.apache.spark.sql.execution.datasources.parquet.ParquetFileFormat$$anonfun$buildPrefetcherWithPartitionValues$1.apply(ParquetFileFormat.scala:676)
at org.apache.spark.sql.execution.datasources.parquet.ParquetFileFormat$$anonfun$buildPrefetcherWithPartitionValues$1.apply(ParquetFileFormat.scala:579)
at org.apache.spark.sql.execution.datasources.AsyncFileDownloader.org$apache$spark$sql$execution$datasources$AsyncFileDownloader$$downloadFile(AsyncFileDownloader.scala:93)
at org.apache.spark.sql.execution.datasources.AsyncFileDownloader$$anonfun$initiateFilesDownload$2$$anon$1.call(AsyncFileDownloader.scala:73)
at org.apache.spark.sql.execution.datasources.AsyncFileDownloader$$anonfun$initiateFilesDownload$2$$anon$1.call(AsyncFileDownloader.scala:72)
at java.util.concurrent.FutureTask.run(FutureTask.java:266)
... 3 more

我找不到记录在案的。有没有人在 EMR 5.28.0 上遇到过这个问题并能够解决这个问题？

在 5.28 上，我能够读取通过 EMR 写入 s3 的文件，但读取由镶木地板写入的现有镶木地板文件会抛出异常，而在 EMR 5.18 上工作正常

更新： 在检查镶木地板文件时，仅适用于 5.18 的旧文件缺少统计数据

creator:            null 
file schema:        parquet-go-root 
timestringhr:        BINARY SNAPPY DO:0 FPO:21015 SZ:1949/25676/13.17 VC:1092 ENC:RLE,BIT_PACKED,PLAIN ST:[no stats for this column]
timeseconds:         INT64 SNAPPY DO:0 FPO:22964 SZ:1397/9064/6.49 VC:1092 ENC:RLE,BIT_PACKED,PLAIN ST:[min: 1564419460, max: 1564419581, num_nulls not defined]

而那些同时适用于 EMR 5.18 和 5.28 的就像

creator:            parquet-mr version 1.10.0 (build 031a6654009e3b82020012a18434c582bd74c73a) 
extra:              org.apache.spark.sql.parquet.row.metadata = {<schema_here>}    
timestringhr:        BINARY SNAPPY DO:0 FPO:3988 SZ:156/152/0.97 VC:1092 ENC:PLAIN_DICTIONARY,RLE,BIT_PACKED ST:[min: 2019-07-29 16:00:00, max: 2019-07-29 16:00:00, num_nulls: 0]
timeseconds:         INT64 SNAPPY DO:0 FPO:4144 SZ:954/1424/1.49 VC:1092 ENC:PLAIN_DICTIONARY,RLE,BIT_PACKED ST:[min: 1564419460, max: 1564419581, num_nulls: 0]

这可能会导致空指针异常。发现与镶木地板先生 https://issues.apache.org/jira/browse/PARQUET-1217 相关的问题。我可以尝试在类路径中包含 parquet 的更新版本或在 EMR 6 beta 上进行测试，看看是否可以解决问题。

Answer 1

尝试向页脚添加created_by值。我将一个 NPE 追溯到 Spark 中的页脚/created_by检查。如果您使用的是xitongsys/parquet-go，请考虑这一点：

var writer_version = "parquet-go version 1.0"
...
...
pw, err := writer.NewJSONWriter(schemaStr, fw, 4)
pw.Footer.CreatedBy = &writer_version

Answer 2

确保镶木地板文件不包含具有零行的行组。加载文件时，您可能必须使用读取器对其进行调试。我们在 AWS Glue 中遇到了这个问题，获得"0 行的非法行组"。

修复： 我们使用 Parquet.net nuget 包，并限制它编写不包含数据的行组。

Answer 3

这很可能是由于 EMR 代入角色缺少访问文件所在的 S3 位置的 IAM 权限造成的。