Github user yingjieMiao commented on a diff in the pull request:
https://github.com/apache/spark/pull/2648#discussion_r18539122
--- Diff: core/src/main/scala/org/apache/spark/rdd/RDD.scala ---
@@ -1084,10 +1084,10 @@ abstract class RDD[T: ClassTag](
if (buf.size == 0) {
numPartsToTry = partsScanned * 4
} else {
- numPartsToTry = (1.5 * num * partsScanned / buf.size).toInt
+ // the left side of max is >=1 whenever partsScanned >= 2
+ numPartsToTry = ((1.5 * num * partsScanned / buf.size).toInt -
partsScanned) max 1
--- End diff --
If we assume some 'uniformness' in the distribution of items over
partitions, then in your example, we probably have to try 1000 partitions.
(uniformness --> linear prediction). However, since I am really new to Spark,
I have little idea about such distribution.
That being said, having a upper bound also makes sense. I am wondering how
other people think about it.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
