Spark fails on big shuffle jobs with java.io.IOException: Filesystem closed

Question

I often find spark fails with large jobs with a rather unhelpful meaningless exception. The worker logs look normal, no errors, but they get state "KILLED". This is extremely common for large shuffles, so operations like .distinct.

The question is, how do I diagnose what's going wrong, and ideally, how do I fix it?

Given that a lot of these operations are monoidal I've been working around the problem by splitting the data into, say 10, chunks, running the app on each chunk, then running the app on all of the resulting outputs. In other words - meta-map-reduce.

14/06/04 12:56:09 ERROR client.AppClient$ClientActor: Master removed our application: FAILED; stopping client
14/06/04 12:56:09 WARN cluster.SparkDeploySchedulerBackend: Disconnected from Spark cluster! Waiting for reconnection...
14/06/04 12:56:09 WARN scheduler.TaskSetManager: Loss was due to java.io.IOException
java.io.IOException: Filesystem closed
    at org.apache.hadoop.hdfs.DFSClient.checkOpen(DFSClient.java:703)
    at org.apache.hadoop.hdfs.DFSInputStream.readWithStrategy(DFSInputStream.java:779)
    at org.apache.hadoop.hdfs.DFSInputStream.read(DFSInputStream.java:840)
    at java.io.DataInputStream.read(DataInputStream.java:149)
    at org.apache.hadoop.io.compress.DecompressorStream.getCompressedData(DecompressorStream.java:159)
    at org.apache.hadoop.io.compress.DecompressorStream.decompress(DecompressorStream.java:143)
    at org.apache.hadoop.io.compress.DecompressorStream.read(DecompressorStream.java:85)
    at java.io.InputStream.read(InputStream.java:101)
    at org.apache.hadoop.util.LineReader.fillBuffer(LineReader.java:180)
    at org.apache.hadoop.util.LineReader.readDefaultLine(LineReader.java:216)
    at org.apache.hadoop.util.LineReader.readLine(LineReader.java:174)
    at org.apache.hadoop.mapred.LineRecordReader.next(LineRecordReader.java:209)
    at org.apache.hadoop.mapred.LineRecordReader.next(LineRecordReader.java:47)
    at org.apache.spark.rdd.HadoopRDD$$anon$1.getNext(HadoopRDD.scala:164)
    at org.apache.spark.rdd.HadoopRDD$$anon$1.getNext(HadoopRDD.scala:149)
    at org.apache.spark.util.NextIterator.hasNext(NextIterator.scala:71)
    at org.apache.spark.InterruptibleIterator.hasNext(InterruptibleIterator.scala:27)
    at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:327)
    at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:327)
    at scala.collection.Iterator$class.foreach(Iterator.scala:727)
    at scala.collection.AbstractIterator.foreach(Iterator.scala:1157)
    at scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:48)
    at scala.collection.mutable.ListBuffer.$plus$plus$eq(ListBuffer.scala:176)
    at scala.collection.mutable.ListBuffer.$plus$plus$eq(ListBuffer.scala:45)
    at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:273)
    at scala.collection.AbstractIterator.to(Iterator.scala:1157)
    at scala.collection.TraversableOnce$class.toList(TraversableOnce.scala:257)
    at scala.collection.AbstractIterator.toList(Iterator.scala:1157)
    at $line5.$read$$iwC$$iwC$$iwC$$iwC$$anonfun$2.apply(<console>:13)
    at $line5.$read$$iwC$$iwC$$iwC$$iwC$$anonfun$2.apply(<console>:13)
    at org.apache.spark.rdd.RDD$$anonfun$1.apply(RDD.scala:450)
    at org.apache.spark.rdd.RDD$$anonfun$1.apply(RDD.scala:450)
    at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:34)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:241)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:232)
    at org.apache.spark.rdd.MappedRDD.compute(MappedRDD.scala:31)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:241)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:232)
    at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:34)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:241)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:232)
    at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:161)
    at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:102)
    at org.apache.spark.scheduler.Task.run(Task.scala:53)
    at org.apache.spark.executor.Executor$TaskRunner$$anonfun$run$1.apply$mcV$sp(Executor.scala:213)
    at org.apache.spark.deploy.SparkHadoopUtil$$anon$1.run(SparkHadoopUtil.scala:42)
    at org.apache.spark.deploy.SparkHadoopUtil$$anon$1.run(SparkHadoopUtil.scala:41)
    at java.security.AccessController.doPrivileged(Native Method)
    at javax.security.auth.Subject.doAs(Subject.java:415)
    at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1548)
    at org.apache.spark.deploy.SparkHadoopUtil.runAsUser(SparkHadoopUtil.scala:41)
    at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:178)
    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:744)

Solution 1

As of September 1st 2014, this is an "open improvement" in Spark. Please see https://issues.apache.org/jira/browse/SPARK-3052. As syrza pointed out in the given link, the shutdown hooks are likely done in incorrect order when an executor failed which results in this message. I understand you will have to little more investigation to figure out the main cause of problem (i.e. why your executor failed). If it is a large shuffle, it might be an out-of-memory error which cause executor failure which then caused the Hadoop Filesystem to be closed in their shutdown hook. So, the RecordReaders in running tasks of that executor throw "java.io.IOException: Filesystem closed" exception. I guess it will be fixed in subsequent release and then you will get more helpful error message :)

Solution 2

Something calls DFSClient.close() or DFSClient.abort(), closing the client. The next file operation then results in the above exception.

I would try to figure out what calls close()/abort(). You could use a breakpoint in your debugger, or modify the Hadoop source code to throw an exception in these methods, so you would get a stack trace.