Github user rdblue commented on a diff in the pull request:
https://github.com/apache/spark/pull/11242#discussion_r56918938
--- Diff: core/src/main/scala/org/apache/spark/rdd/UnionRDD.scala ---
@@ -62,7 +64,23 @@ class UnionRDD[T: ClassTag](
var rdds: Seq[RDD[T]])
extends RDD[T](sc, Nil) { // Nil since we implement getDependencies
+ // Evaluate partitions in parallel. Partitions of each rdd will be
cached by the `partitions`
+ // val in `RDD`.
+ private[spark] lazy val parallelPartitionEval: Boolean = {
--- End diff --
If I remember correctly, no one recommends reusing task VMs for MR. Too
many things assumed that they weren't reused, which is a major advantage for
Spark. Also, mapper VMs aren't normally calling getSplits in MR. Split planning
happens on the driver/job client, which may stick around submit multiple jobs
but usually not in parallel. Caching data used for split planning would
typically make a difference for iterative use cases.
This commit does change the default to parallel split planning, but I'm
arguing that it's a good idea to have the safety valve, not that this is a
significant enough risk that we should make parallel split planning opt-in.
---
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]
