Spark Swift Integration Parquet

Question

I have using Spark 1.3.0 and using it to write Parquet files to Openstack Swift Object store for a while now. I am using around 12 paritions for the parquet files which writes the parquet file in several parts on Swift. There's no problem in writing the files. But when I try to read it via Spark I get this error:

     ERROR Executor: Exception in task 9.0 in stage 2.0 (TID 22)
java.io.EOFException
    at java.io.DataInputStream.readFully(DataInputStream.java:197)
    at java.io.DataInputStream.readFully(DataInputStream.java:169)
    at parquet.hadoop.ParquetFileReader$ConsecutiveChunkList.readAll(ParquetFileReader.java:730)
    at parquet.hadoop.ParquetFileReader.readNextRowGroup(ParquetFileReader.java:490)
    at parquet.hadoop.InternalParquetRecordReader.checkRead(InternalParquetRecordReader.java:116)
    at parquet.hadoop.InternalParquetRecordReader.nextKeyValue(InternalParquetRecordReader.java:193)
    at parquet.hadoop.ParquetRecordReader.nextKeyValue(ParquetRecordReader.java:204)
    at org.apache.spark.rdd.NewHadoopRDD$$anon$1.hasNext(NewHadoopRDD.scala:143)
    at org.apache.spark.InterruptibleIterator.hasNext(InterruptibleIterator.scala:39)
    at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:327)
    at scala.collection.Iterator$$anon$10.hasNext(Iterator.scala:308)
    at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:327)
    at org.apache.spark.util.collection.ExternalSorter.insertAll(ExternalSorter.scala:210)
    at org.apache.spark.shuffle.sort.SortShuffleWriter.write(SortShuffleWriter.scala:63)
    at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:68)
    at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:41)
    at org.apache.spark.scheduler.Task.run(Task.scala:64)
    at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:203)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)

I am using Hadoop-Swift patch available via https://github.com/openstack/sahara-extra/tree/master/hadoop-swiftfs which enables the Hadoop to recognize Swift as a filesystem.

Note 1: If download the file from Swift to local filesystem, Spark can read the file perfectly then.

Note 2: I have also noticed if I don't partition the parquet file, the reading is perfect.

Anyone who is using Spark, Openstack Swift ?

Answer 1

I have been using Spark 1.3.0 with Openstack Swift as my storage backend as well and encountered the exact same problem.

Using the cache functionality of Spark provided a valid workaround waiting for an actual fix in the swift support of hadoop-cloud.

val df = spark.read.parquet("your_swift_url").persist()