Skip to content

[SPARK-22712][SQL] Use buildReaderWithPartitionValues in native OrcFileFormat #19907

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
dongjoon-hyun wants to merge 2 commits into from
Closed

[SPARK-22712][SQL] Use buildReaderWithPartitionValues in native OrcFileFormat #19907

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
199c835
[SPARK-22712][SQL] Use `buildReaderWithPartitionValues` in native Orc…
dongjoon-hyun Dec 6, 2017
f69fc4e
Use joinedRow.
dongjoon-hyun Dec 6, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 13 additions & 5 deletions sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/orc/OrcDeserializer.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,13 +32,21 @@ import org.apache.spark.unsafe.types.UTF8String
*/
class OrcDeserializer(
dataSchema: StructType,
requiredSchema: StructType,
requestedColIds: Array[Int]) {

private val resultRow = new SpecificInternalRow(requiredSchema.map(_.dataType))
resultSchema: StructType,
requestedColIds: Array[Int],
partitionValues: InternalRow) {

// Make a resultRow and initialize the partition column values once.
private val resultRow = new SpecificInternalRow(resultSchema.map(_.dataType))
private var i = 0
private val start = resultSchema.length - partitionValues.numFields
while (i < partitionValues.numFields) {
resultRow.update(start + i, partitionValues.get(i, resultSchema(start + i).dataType))
i += 1
}

private val fieldWriters: Array[WritableComparable[_] => Unit] = {
requiredSchema.zipWithIndex
resultSchema.zipWithIndex
// The value of missing columns are always null, do not need writers.
.filterNot { case (_, index) => requestedColIds(index) == -1 }
.map { case (f, index) =>
Expand Down
10 changes: 7 additions & 3 deletions sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/orc/OrcFileFormat.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -124,7 +124,7 @@ class OrcFileFormat
true
}

override def buildReader(
override def buildReaderWithPartitionValues(

@dongjoon-hyun dongjoon-hyun Dec 6, 2017

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hi, @cloud-fan . During the previous ORC PR, we left this behind.

sparkSession: SparkSession,
dataSchema: StructType,
partitionSchema: StructType,
Expand All @@ -138,6 +138,8 @@ class OrcFileFormat
}
}

val resultSchema = StructType(requiredSchema.fields ++ partitionSchema.fields)

val broadcastedConf =
sparkSession.sparkContext.broadcast(new SerializableConfiguration(hadoopConf))
val isCaseSensitive = sparkSession.sessionState.conf.caseSensitiveAnalysis
Expand Down Expand Up @@ -167,8 +169,10 @@ class OrcFileFormat
val iter = new RecordReaderIterator[OrcStruct](orcRecordReader)
Option(TaskContext.get()).foreach(_.addTaskCompletionListener(_ => iter.close()))

val unsafeProjection = UnsafeProjection.create(requiredSchema)
val deserializer = new OrcDeserializer(dataSchema, requiredSchema, requestedColIds)
val colIds = requestedColIds ++ List.fill(partitionSchema.length)(-1).toArray[Int]
val unsafeProjection = UnsafeProjection.create(resultSchema)

@cloud-fan cloud-fan Dec 6, 2017

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can we follow parquet and just join the data row and partition row, and do a final unsafe projection? It's much easier and there is no performance difference.

@dongjoon-hyun dongjoon-hyun Dec 6, 2017

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Parquet Vectorization work like the following.

      // UnsafeRowParquetRecordReader appends the columns internally to avoid another copy.
      if (parquetReader.isInstanceOf[VectorizedParquetRecordReader] &&
          enableVectorizedReader) {
        iter.asInstanceOf[Iterator[InternalRow]]

@dongjoon-hyun dongjoon-hyun Dec 6, 2017

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Oh, I see. you meant non-vectorized path. Sorry, I was confused since I focused too much on vectorized path. I'll do.

val deserializer =
new OrcDeserializer(dataSchema, resultSchema, colIds, file.partitionValues)
iter.map(value => unsafeProjection(deserializer.deserialize(value)))
}
}
Expand Down