Skip to main content

org.apache.spark.SparkException: Kryo serialization failed: Buffer overflow

 

We were running an application which was leading to below error - 

Job aborted due to stage failure: Task 137 in stage 5.0 failed 4 times, most recent failure: Lost task 137.3 in stage 5.0 (TID 2090, ncABC.hadoop.com, executor 1): org.apache.spark.SparkException: Kryo serialization failed: Buffer overflow. Available: 0, required: 59606960. To avoid this, increase spark.kryoserializer.buffer.max value.
	at org.apache.spark.serializer.KryoSerializerInstance.serialize(KryoSerializer.scala:330)
	at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:456)
	at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
	at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
	at java.lang.Thread.run(Thread.java:748)
Caused by: com.esotericsoftware.kryo.KryoException: Buffer overflow. Available: 0, required: 59606960
	at com.esotericsoftware.kryo.io.Output.require(Output.java:167)
	at com.esotericsoftware.kryo.io.Output.writeBytes(Output.java:251)
	at com.esotericsoftware.kryo.io.Output.writeBytes(Output.java:237)
	at com.esotericsoftware.kryo.serializers.DefaultArraySerializers$ByteArraySerializer.write(DefaultArraySerializers.java:49)
	at com.esotericsoftware.kryo.serializers.DefaultArraySerializers$ByteArraySerializer.write(DefaultArraySerializers.java:38)
	at com.esotericsoftware.kryo.Kryo.writeClassAndObject(Kryo.java:651)
	at com.twitter.chill.Tuple2Serializer.write(TupleSerializers.scala:37)
	at com.twitter.chill.Tuple2Serializer.write(TupleSerializers.scala:33)
	at com.esotericsoftware.kryo.Kryo.writeClassAndObject(Kryo.java:651)


Solution - 

On further analysis, we found that this error was originating from BroadcastNestedLoopJoin

at org.apache.spark.sql.execution.SparkPlan$$anonfun$executeQuery$1.apply(SparkPlan.scala:165)
at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
at org.apache.spark.sql.execution.SparkPlan.executeQuery(SparkPlan.scala:162)
at org.apache.spark.sql.execution.SparkPlan.executeBroadcast(SparkPlan.scala:150)
at org.apache.spark.sql.execution.joins.BroadcastNestedLoopJoinExec.doExecute(BroadcastNestedLoopJoinExec.scala:343)


So, we removed BroadcastNestedLoopJoin by updating SQL where clause having NOT IN  to NOT EXISTS. Refer details here - http://querydb.blogspot.com/2021/06/spark-disable-broadcast-join-not.html


This solved our problem.


Alternative Solution - 

We read this solution at multiple places but we didn't try to set below properties - 

--conf  spark.kryoserializer.buffer.max=1024m  spark.kryoserializer.buffer=512m 

And, don't recommend to set it anything other than default values because,

There is a note that there will be one buffer per core on each worker. This buffer will grow up to spark.kryoserializer.buffer.max if needed. 


That said, if you have worker with 4 cores then 4*512m ~ 2GB is taken up for Kryo Buffer, and that seems a good chunk of memory. 


Comments

Popular posts

Spark MongoDB Connector Not leading to correct count or data while reading

  We are using Scala 2.11 , Spark 2.4 and Spark MongoDB Connector 2.4.4 Use Case 1 - We wanted to read a Shareded Mongo Collection and copy its data to another Mongo Collection. We noticed that after Spark Job successful completion. Output MongoDB did not had many records. Use Case 2 -  We read a MongoDB collection and doing count on dataframe lead to different count on each execution. Analysis,  We realized that MongoDB Spark Connector is missing data on bulk read as a dataframe. We tried various partitioner, listed on page -  https://www.mongodb.com/docs/spark-connector/v2.4/configuration/  But, none of them worked for us. Finally, we tried  MongoShardedPartitioner  this lead to constant count on each execution. But, it was greater than the actual count of records on the collection. This seems to be limitation with MongoDB Spark Connector. But,  MongoShardedPartitioner  seemed closest possible solution to this kind of situation. But, it per...




Scala Spark building Jar leads java.lang.StackOverflowError

  Exception -  [Thread-3] ERROR scala_maven.ScalaCompileMojo - error: java.lang.StackOverflowError [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.generic.TraversableForwarder$class.isEmpty(TraversableForwarder.scala:36) [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.mutable.ListBuffer.isEmpty(ListBuffer.scala:45) [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.mutable.ListBuffer.toList(ListBuffer.scala:306) [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.mutable.ListBuffer.result(ListBuffer.scala:300) [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.mutable.Stack$StackBuilder.result(Stack.scala:31) [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.mutable.Stack$StackBuilder.result(Stack.scala:27) [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.generic.GenericCompanion.apply(GenericCompanion.scala:50) [Thread-3] INFO scala_maven.ScalaCompile...




MongoDB Chunk size many times bigger than configure chunksize (128 MB)

  Shard Shard_0 at Shard_0/xyz.com:27018 { data: '202.04GiB', docs: 117037098, chunks: 5, 'estimated data per chunk': '40.4GiB', 'estimated docs per chunk': 23407419 } --- Shard Shard_1 at Shard_1/abc.com:27018 { data: '201.86GiB', docs: 116913342, chunks: 4, 'estimated data per chunk': '50.46GiB', 'estimated docs per chunk': 29228335 } Per MongoDB-  Starting in 6.0.3, we balance by data size instead of the number of chunks. So the 128MB is now only the size of data we migrate at-a-time. So large data size per chunk is good now, as long as the data size per shard is even for the collection. refer -  https://www.mongodb.com/community/forums/t/chunk-size-many-times-bigger-than-configure-chunksize-128-mb/212616 https://www.mongodb.com/docs/v6.0/release-notes/6.0/#std-label-release-notes-6.0-balancing-policy-changes




AWS EMR Spark – Much Larger Executors are Created than Requested

  Starting EMR 5.32 and EMR 6.2 you can notice that Spark can launch much larger executors that you request in your job settings. For example - We started a Spark Job with  spark.executor.cores  =   4 But, one can see that the executors with 20 cores (instead of 4 as defined by spark.executor.cores) were launched. The reason for allocating larger executors is that there is a AWS specific Spark option spark.yarn.heterogeneousExecutors.enabled (exists in EMR only, does not exist in Open Source Spark) that is set to true by default that combines multiple executor creation requests on the same node into a larger executor container. So as the result you have fewer executor containers than you expected, each of them has more memory and cores that you specified. If you disable this option (--conf "spark.yarn.heterogeneousExecutors.enabled=false"), EMR will create containers with the specified spark.executor.memory and spark.executor.cores settings and will not co...




Hive Count Query not working

Hive with Tez execution engine -  count(*) not working , returning 0 results.  Solution -  set hive.compute.query.using.stats=false Refer -  https://cwiki.apache.org/confluence/display/Hive/Configuration+Properties hive.compute.query.using.stats Default Value:  false Added In: Hive 0.13.0 with  HIVE-5483 When set to true Hive will answer a few queries like min, max, and count(1) purely using statistics stored in the metastore. For basic statistics collection, set the configuration property  hive.stats.autogather   to true. For more advanced statistics collection, run ANALYZE TABLE queries.