JoshRosen commented on a change in pull request #27246:
[SPARK-30536][CORE][SQL] Sort-merge join operator spilling performance
improvements
URL: https://github.com/apache/spark/pull/27246#discussion_r369918361
##
File path:
sql/core/src/main/scala/org/apache/spark/sql/execution/ExternalAppendOnlyUnsafeRowArray.scala
##
@@ -204,22 +195,44 @@ private[sql] class ExternalAppendOnlyUnsafeRowArray(
}
}
- private[this] class SpillableArrayIterator(
+ private[this] class MergerIterator(
iterator: UnsafeSorterIterator,
- numFieldPerRow: Int)
+ numFieldPerRow: Int,
+ startIndex: Int)
extends ExternalAppendOnlyUnsafeRowArrayIterator {
-private val currentRow = new UnsafeRow(numFieldPerRow)
+private var currentIndex = startIndex
-override def hasNext(): Boolean = !isModified() && iterator.hasNext
+private val currentRow = {
+ if (startIndex < numRows) {
+inMemoryBuffer(currentIndex)
+ } else {
+new UnsafeRow(numFieldPerRow)
+ }
+}
+
+override def hasNext(): Boolean = {
+ if (currentIndex < numRows) {
+!isModified()
+ } else {
+!isModified() && iterator.hasNext
+ }
+}
override def next(): UnsafeRow = {
throwExceptionIfModified()
- iterator.loadNext()
- currentRow.pointTo(iterator.getBaseObject, iterator.getBaseOffset,
iterator.getRecordLength)
- currentRow
+ if (currentIndex < numRows) {
+val result = inMemoryBuffer(currentIndex)
+currentIndex += 1
+result
+ } else {
+iterator.loadNext()
+currentRow.pointTo(iterator.getBaseObject, iterator.getBaseOffset,
iterator.getRecordLength)
Review comment:
If `startIndex < numRows` then `currentRow` will be an `UnsafeRow` in
`inMemoryBuffer`, so when we then roll past the `numRows` boundary then we'd
start mutating that `UnsafeRow`'s state (which would corrupt the data in case
we iterated over this multiple times).
As a result, I think it's safer to leave `currentRow` as it was defined
before, but rename it to, say, `currentSorterRow` to make it clear that it's
only used for pointing at spilled records.
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:
us...@infra.apache.org
With regards,
Apache Git Services
-
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
