[
https://issues.apache.org/jira/browse/SPARK-18737?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15741466#comment-15741466
]
Dr. Michael Menzel commented on SPARK-18737:
--------------------------------------------
We actually want to use Kryo, but want to deactivate it for debugging purposes.
We see random Kryo exceptions that occur even before any of our business logic
comes into play. It seems some initial cluster communication is having issues
from time to time (not constantly).
We would like to investigate the actual issue, but the Kryo serializer turns
the actual issue into a serialization exception. We hoped with turning off Kryo
serialization, we can get to the root cause of the issue.
Alternatively, enabling Kryo logging may help. But we tried that already by
setting the minlog level to debug. Rarely any output at all.
After going through the Kryo code we figured there are hardly any log
statements made :(
Any idea what could provide us any means to debug the issue?
We are on Spark 2.0.1 and 2.0.2, using Spark Streaming to consume messages from
SQS. All works fine and stable on Spark 1.6 and below.
In Spark 2.x the cluster runs unstable and errors in task communication with
workes increase over time. Kryo exceptions are thrown with every communication
error.
> Serialization setting "spark.serializer" ignored in Spark 2.x
> -------------------------------------------------------------
>
> Key: SPARK-18737
> URL: https://issues.apache.org/jira/browse/SPARK-18737
> Project: Spark
> Issue Type: Bug
> Affects Versions: 2.0.0, 2.0.1
> Reporter: Dr. Michael Menzel
>
> The following exception occurs although the JavaSerializer has been activated:
> 16/11/22 10:49:24 INFO TaskSetManager: Starting task 0.0 in stage 9.0 (TID
> 77, ip-10-121-14-147.eu-central-1.compute.internal, partition 1, RACK_LOCAL,
> 5621 bytes)
> 16/11/22 10:49:24 INFO YarnSchedulerBackend$YarnDriverEndpoint: Launching
> task 77 on executor id: 2 hostname:
> ip-10-121-14-147.eu-central-1.compute.internal.
> 16/11/22 10:49:24 INFO BlockManagerInfo: Added broadcast_11_piece0 in memory
> on ip-10-121-14-147.eu-central-1.compute.internal:45059 (size: 879.0 B, free:
> 410.4 MB)
> 16/11/22 10:49:24 WARN TaskSetManager: Lost task 0.0 in stage 9.0 (TID 77,
> ip-10-121-14-147.eu-central-1.compute.internal):
> com.esotericsoftware.kryo.KryoException: Encountered unregistered class ID:
> 13994
> at
> com.esotericsoftware.kryo.util.DefaultClassResolver.readClass(DefaultClassResolver.java:137)
> at com.esotericsoftware.kryo.Kryo.readClass(Kryo.java:670)
> at com.esotericsoftware.kryo.Kryo.readClassAndObject(Kryo.java:781)
> at
> org.apache.spark.serializer.KryoDeserializationStream.readObject(KryoSerializer.scala:229)
> at
> org.apache.spark.serializer.DeserializationStream$$anon$1.getNext(Serializer.scala:169)
> at org.apache.spark.util.NextIterator.hasNext(NextIterator.scala:73)
> at scala.collection.Iterator$class.foreach(Iterator.scala:893)
> at org.apache.spark.util.NextIterator.foreach(NextIterator.scala:21)
> at
> scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:59)
> at
> scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:104)
> at
> scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:48)
> at
> scala.collection.TraversableOnce$class.to(TraversableOnce.scala:310)
> at org.apache.spark.util.NextIterator.to(NextIterator.scala:21)
> at
> scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:302)
> at org.apache.spark.util.NextIterator.toBuffer(NextIterator.scala:21)
> at
> scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:289)
> at org.apache.spark.util.NextIterator.toArray(NextIterator.scala:21)
> at
> org.apache.spark.rdd.RDD$$anonfun$toLocalIterator$1$$anonfun$org$apache$spark$rdd$RDD$$anonfun$$collectPartition$1$1.apply(RDD.scala:927)
> at
> org.apache.spark.rdd.RDD$$anonfun$toLocalIterator$1$$anonfun$org$apache$spark$rdd$RDD$$anonfun$$collectPartition$1$1.apply(RDD.scala:927)
> at
> org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1916)
> at
> org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1916)
> at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:70)
> at org.apache.spark.scheduler.Task.run(Task.scala:86)
> at
> org.apache.spark.executor.Executor$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)
> The code runs perfectly with Spark 1.6.0. Since we moved to 2.0.0 and now
> 2.0.1, we see the Kyro deserialization exception and over time the Spark
> streaming job stops processing since too many tasks failed.
> Our action was to use conf.set("spark.serializer",
> "org.apache.spark.serializer.JavaSerializer") and to disable Kryo class
> registration with conf.set("spark.kryo.registrationRequired", false). We hope
> to identify the root cause of the exception.
> However, setting the serializer to JavaSerializer is oviously ignored by the
> Spark-internals. Despite the setting we still see the exception printed in
> the log and tasks fail. The occurence seems to be non-deterministic, but to
> become more frequent over time.
> Several questions we could not answer during our troubleshooting:
> 1. How can the debug log for Kryo be enabled? -- We tried following the
> minilog documentation, but no output can be found.
> 2. Is the serializer setting effective for Spark internal serializations? How
> can the JavaSerialize be forced on internal serializations for worker to
> driver communication?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
