Critical Description
After upgrading our Spark from 1.6.2 to 2.1.0, I encounter a confusing NullPointerException when creating table under Spark 2.1.0, but the problem does not exists in Spark 1.6.1.
Environment: Hive 1.2.1, Hadoop 2.6.4
==================== Code ====================
// spark is an instance of HiveContext
// merge is a Hive UDF
val df = spark.sql("SELECT merge(field_a, null) AS new_a, field_b AS new_b FROM tb_1 group by field_a, field_b")
df.createTempView("tb_temp")
spark.sql("create table tb_result stored as parquet as " +
"SELECT new_a" +
"FROM tb_temp" +
"LEFT JOIN `tb_2` ON " +
"if(((`tb_temp`.`new_b`) = '' OR (`tb_temp`.`new_b`) IS NULL), concat('GrLSRwZE_', cast((rand() * 200) AS int)), (`tb_temp`.`new_b`)) = `tb_2`.`fka6862f17`")
==================== Physical Plan ====================
*Project [new_a]
+- *BroadcastHashJoin [if (((new_b = ) || isnull(new_b))) concat(GrLSRwZE_, cast(cast((_nondeterministic * 200.0) as int) as string)) else new_b], [fka6862f17], LeftOuter, BuildRight
:- HashAggregate(keys=[field_a, field_b], functions=[], output=[new_a, new_b, _nondeterministic])
: +- Exchange(coordinator ) hashpartitioning(field_a, field_b, 180), coordinator[target post-shuffle partition size: 1024880]
: +- *HashAggregate(keys=[field_a, field_b], functions=[], output=[field_a, field_b])
: +- *FileScan parquet bdp.tb_1[field_a,field_b] Batched: true, Format: Parquet, Location: InMemoryFileIndex[hdfs://hdcluster/data/tb_1, PartitionFilters: [], PushedFilters: [], ReadSchema: struct
+- BroadcastExchange HashedRelationBroadcastMode(List(input[0, string, true]))
+- *Project [fka6862f17]
+- *FileScan parquet bdp.tb_2[fka6862f17] Batched: true, Format: Parquet, Location: InMemoryFileIndex[hdfs://hdcluster/data/tb_2, PartitionFilters: [], PushedFilters: [], ReadSchema: struct
What does '*' mean before HashAggregate?
==================== Exception ====================
org.apache.spark.SparkException: Task failed while writing rows
...
java.lang.NullPointerException
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply_2$(Unknown Source)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source)
at org.apache.spark.sql.execution.aggregate.AggregationIterator$$anonfun$generateResultProjection$3.apply(AggregationIterator.scala:260)
at org.apache.spark.sql.execution.aggregate.AggregationIterator$$anonfun$generateResultProjection$3.apply(AggregationIterator.scala:259)
at org.apache.spark.sql.execution.aggregate.TungstenAggregationIterator.next(TungstenAggregationIterator.scala:392)
at org.apache.spark.sql.execution.aggregate.TungstenAggregationIterator.next(TungstenAggregationIterator.scala:79)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIterator.processNext(Unknown Source)
at org.apache.spark.sql.execution.BufferedRowIterator.hasNext(BufferedRowIterator.java:43)
at org.apache.spark.sql.execution.WholeStageCodegenExec$$anonfun$8$$anon$1.hasNext(WholeStageCodegenExec.scala:377)
at org.apache.spark.sql.execution.datasources.FileFormatWriter$SingleDirectoryWriteTask.execute(FileFormatWriter.scala:252)
at org.apache.spark.sql.execution.datasources.FileFormatWriter$$anonfun$org$apache$spark$sql$execution$datasources$FileFormatWriter$$executeTask$3.apply(FileFormatWriter.scala:199)
at org.apache.spark.sql.execution.datasources.FileFormatWriter$$anonfun$org$apache$spark$sql$execution$datasources$FileFormatWriter$$executeTask$3.apply(FileFormatWriter.scala:197)
at org.apache.spark.util.Utils$.tryWithSafeFinallyAndFailureCallbacks(Utils.scala:1341)
at org.apache.spark.sql.execution.datasources.FileFormatWriter$.org$apache$spark$sql$execution$datasources$FileFormatWriter$$executeTask(FileFormatWriter.scala:202)
at org.apache.spark.sql.execution.datasources.FileFormatWriter$$anonfun$write$1$$anonfun$4.apply(FileFormatWriter.scala:138)
at org.apache.spark.sql.execution.datasources.FileFormatWriter$$anonfun$write$1$$anonfun$4.apply(FileFormatWriter.scala:137)
at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:87)
at org.apache.spark.scheduler.Task.run(Task.scala:99)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:282)
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)
I also found that when I changed my code as follow:
spark.sql("create table tb_result stored as parquet as " +
"SELECT new_b" +
"FROM tb_temp" +
"LEFT JOIN `tb_2` ON " +
"if(((`tb_temp`.`new_b`) = '' OR (`tb_temp`.`new_b`) IS NULL), concat('GrLSRwZE_', cast((rand() * 200) AS int)), (`tb_temp`.`new_b`)) = `tb_2`.`fka6862f17`")
or
spark.sql("create table tb_result stored as parquet as " +
"SELECT new_a" +
"FROM tb_temp" +
"LEFT JOIN `tb_2` ON " +
"if(((`tb_temp`.`new_b`) = '' OR (`tb_temp`.`new_b`) IS NULL), concat('GrLSRwZE_', cast((200) AS int)), (`tb_temp`.`new_b`)) = `tb_2`.`fka6862f17`")
will not have this problem.
== Physical Plan of select new_b ... ==
*Project [new_b]
+- *BroadcastHashJoin [if (((new_b = ) || isnull(new_b))) concat(GrLSRwZE_, cast(cast((_nondeterministic * 200.0) as int) as string)) else new_b], [fka6862f17], LeftOuter, BuildRight
:- *HashAggregate(keys=[field_a, field_b], functions=[], output=[new_b, _nondeterministic])
: +- Exchange(coordinator ) hashpartitioning(field_a, field_b, 180), coordinator[target post-shuffle partition size: 1024880]
: +- *HashAggregate(keys=[field_a, field_b], functions=[], output=[field_a, field_b])
: +- *FileScan parquet bdp.tb_1[field_a,field_b] Batched: true, Format: Parquet, Location: InMemoryFileIndex[hdfs://hdcluster/data/tb_1, PartitionFilters: [], PushedFilters: [], ReadSchema: struct
+- BroadcastExchange HashedRelationBroadcastMode(List(input[0, string, true]))
+- *Project [fka6862f17]
+- *FileScan parquet bdp.tb_2[fka6862f17] Batched: true, Format: Parquet, Location: InMemoryFileIndex[hdfs://hdcluster/data/tb_2, PartitionFilters: [], PushedFilters: [], ReadSchema: struct
Difference is `HashAggregate(keys=[field_a, field_b], functions=[], output=[new_b, _nondeterministic])` has a '*' char before it.
It looks like something wrong with WholeStageCodegen when combine HiveUDF + rand() + group by + join.
