Github user rdblue commented on a diff in the pull request:
https://github.com/apache/spark/pull/11242#discussion_r56878058
--- 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 --
I'm reluctant to make `rdds.zipWithIndex` parallel by default with no way
to disable it. It's a common case for RDDs in the union to wrap a Hadoop
InputFormat. I don't think it's normally an expectation that getSplits could be
called concurrently and custom InputFormats are fairly common. I know
Parquet/Hive had a bug where Parquet's InputFormat was caching results, so I
think users should be able to disable this to avoid similar issues.
How about:
1. Separate the flag from the partition eval method so it is clear that the
flag indicates whether partitions are calculated in parallel (for testing)
2. Use the default thread pool instead of custom parallelism
Would that work?
---
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]
