jackylk commented on a change in pull request #3538: [CARBONDATA-3637] Optimize
insert into flow
URL: https://github.com/apache/carbondata/pull/3538#discussion_r376192708
##########
File path:
integration/spark-common/src/main/scala/org/apache/carbondata/spark/load/DataLoadProcessBuilderOnSpark.scala
##########
@@ -179,6 +182,93 @@ object DataLoadProcessBuilderOnSpark {
updateLoadStatus(model, partialSuccessAccum)
}
+ def insertDataUsingGlobalSortWithInternalRow(sparkSession: SparkSession,
+ scanResultRDD : RDD[InternalRow],
+ model: CarbonLoadModel,
+ hadoopConf: Configuration): Array[(String, (LoadMetadataDetails,
ExecutionErrors))] = {
+ val originRDD = scanResultRDD
+
+ val sc = sparkSession.sparkContext
+ val modelBroadcast = sc.broadcast(model)
+ val partialSuccessAccum = sc.longAccumulator("Partial Success Accumulator")
+ val sortStepRowCounter = sc.longAccumulator("Sort Processor Accumulator")
+ val writeStepRowCounter = sc.longAccumulator("Write Processor Accumulator")
+
+ hadoopConf
+ .set(CarbonCommonConstants.CARBON_WRITTEN_BY_APPNAME,
sparkSession.sparkContext.appName)
+
+ val conf = SparkSQLUtil.broadCastHadoopConf(sc, hadoopConf)
+ val configuration = DataLoadProcessBuilder.createConfiguration(model)
+
+ // 1. Convert internalRow to object array
+ val fields = Util
+
.convertToSparkSchemaFromColumnSchema(model.getCarbonDataLoadSchema.getCarbonTable,
true)
+ .fields
+ .toSeq
+
+ val dataTypes = fields.map(field => field.dataType)
+ val map: mutable.Map[String, GenericDataType[_]] = mutable.Map[String,
GenericDataType[_]]()
+ CommonUtil.convertComplexDataType(map, configuration)
+ val rdd = originRDD.map { internalRow =>
+
CommonUtil.getObjectArrayFromInternalRowAndConvertComplexTypeForGlobalSort(internalRow,
+ fields,
+ map)
+ }
+
+ // 2. sort
+ var numPartitions = CarbonDataProcessorUtil.getGlobalSortPartitions(
+
configuration.getDataLoadProperty(CarbonCommonConstants.LOAD_GLOBAL_SORT_PARTITIONS))
+ if (numPartitions <= 0) {
+ numPartitions = originRDD.partitions.length
+ }
+ // Because if the number of partitions greater than 1, there will be
action operator
+ // (sample) in
+ // sortBy operator. So here we cache the rdd to avoid do input and convert
again.
+ if (numPartitions > 1) {
+ rdd.persist(StorageLevel.fromString(
+ CarbonProperties.getInstance().getGlobalSortRddStorageLevel()))
+ }
+ val sortColumnsLength =
model.getCarbonDataLoadSchema.getCarbonTable.getSortColumns.size()
+ val sortColumnDataTypes = dataTypes.take(sortColumnsLength)
+ val rowComparator =
GlobalSortHelper.generateRowComparator(sortColumnDataTypes)
+ val sortRDD = rdd.sortBy(x => getKey(x, sortColumnsLength,
sortColumnDataTypes),
Review comment:
move `getKey(x, sortColumnsLength, sortColumnDataTypes)` to next line
use pattern like ```
rdd.sortBy { row =>
...
}
```
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to 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
