Dooyoung-Hwang commented on a change in pull request #22347: [SPARK-25353][SQL]
executeTake in SparkPlan is modified to avoid unnecessary decoding.
URL: https://github.com/apache/spark/pull/22347#discussion_r299316137
##########
File path:
sql/core/src/main/scala/org/apache/spark/sql/execution/SparkPlan.scala
##########
@@ -423,31 +424,37 @@ abstract class SparkPlan extends QueryPlan[SparkPlan]
with Logging with Serializ
// Otherwise, interpolate the number of partitions we need to try, but
overestimate
// it by 50%. We also cap the estimation in the end.
val limitScaleUpFactor = Math.max(sqlContext.conf.limitScaleUpFactor,
2)
- if (buf.isEmpty) {
+ if (scannedRowCount == 0) {
numPartsToTry = partsScanned * limitScaleUpFactor
} else {
- val left = n - buf.size
+ val left = n - scannedRowCount
// As left > 0, numPartsToTry is always >= 1
- numPartsToTry = Math.ceil(1.5 * left * partsScanned / buf.size).toInt
+ numPartsToTry = Math.ceil(1.5 * left * partsScanned /
scannedRowCount).toInt
numPartsToTry = Math.min(numPartsToTry, partsScanned *
limitScaleUpFactor)
}
}
val p = partsScanned.until(math.min(partsScanned + numPartsToTry,
totalParts).toInt)
val sc = sqlContext.sparkContext
- val res = sc.runJob(childRDD,
- (it: Iterator[Array[Byte]]) => if (it.hasNext) it.next() else
Array.empty[Byte], p)
-
- buf ++= res.flatMap(decodeUnsafeRows)
-
+ val res = sc.runJob(childRDD, (it: Iterator[(Long, Array[Byte])]) =>
+ if (it.hasNext) it.next() else (0L, Array.empty[Byte]), p)
+
+ var i = 0
+ while (scannedRowCount < n && i < res.length) {
+ val rows = decodeUnsafeRows(res(i)._2)
Review comment:
Yes, we can remove scannedRowCount.
----------------------------------------------------------------
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
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
