HyukjinKwon commented on a change in pull request #23559: [SPARK-26630][SQL]
Fix ClassCastException in TableReader while creating HadoopRDD
URL: https://github.com/apache/spark/pull/23559#discussion_r249274336
##########
File path: sql/hive/src/main/scala/org/apache/spark/sql/hive/TableReader.scala
##########
@@ -288,29 +285,105 @@ class HadoopTableReader(
}
}
+ /**
+ * The entry of creating a RDD.
+ */
+ private def getRDD(
+ inputClassName: String,
+ localTableDesc: TableDesc,
+ inputPathStr: String): RDD[Writable] = {
+ if (isCreateNewHadoopRDD(inputClassName)) {
+ createNewHadoopRdd(
+ localTableDesc,
+ inputPathStr,
+ inputClassName)
+ } else {
+ createHadoopRdd(
+ localTableDesc,
+ inputPathStr,
+ inputClassName)
+ }
+ }
+
/**
* Creates a HadoopRDD based on the broadcasted HiveConf and other job
properties that will be
* applied locally on each slave.
*/
private def createHadoopRdd(
tableDesc: TableDesc,
path: String,
- inputFormatClass: Class[InputFormat[Writable, Writable]]): RDD[Writable] =
{
+ inputClassName: String): RDD[Writable] = {
val initializeJobConfFunc =
HadoopTableReader.initializeLocalJobConfFunc(path, tableDesc) _
val rdd = new HadoopRDD(
sparkSession.sparkContext,
_broadcastedHadoopConf.asInstanceOf[Broadcast[SerializableConfiguration]],
Some(initializeJobConfFunc),
- inputFormatClass,
+ getInputFormat(inputClassName),
classOf[Writable],
classOf[Writable],
_minSplitsPerRDD)
// Only take the value (skip the key) because Hive works only with values.
rdd.map(_._2)
}
+
+ /**
+ * Creates a HadoopRDD based on the broadcasted HiveConf and other job
properties that will be
+ * applied locally on each slave.
+ */
+ private def createNewHadoopRdd(
+ tableDesc: TableDesc,
+ path: String,
+ inputClassName: String): RDD[Writable] = {
+
+ val initializeJobConfFunc =
HadoopTableReader.initializeLocalJobConfFunc(path, tableDesc) _
+
+ val newJobConf = new JobConf(hadoopConf)
+ initializeJobConfFunc.apply(newJobConf)
+ val rdd = new NewHadoopRDD(
+ sparkSession.sparkContext,
+ getNewInputFormat(inputClassName),
+ classOf[Writable],
+ classOf[Writable],
+ newJobConf
+ )
+
+ // Only take the value (skip the key) because Hive works only with values.
+ rdd.map(_._2)
+ }
+
+ /**
+ * If `spark.sql.hive.fileInputFormat.enabled` is true, this function will
optimize the input
+ * method while reading Hive tables.
+ * For old input format `org.apache.hadoop.mapred.InputFormat`.
+ */
+ private def getInputFormat(
Review comment:
those three functions below has a couple of lines and they are only called
once. I wouldn't make a separate function. The codes can be moved into the
functions above with some comments.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
