Github user habren commented on a diff in the pull request:
https://github.com/apache/spark/pull/22018#discussion_r208784609
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/InMemoryFileIndex.scala
---
@@ -297,7 +297,7 @@ object InMemoryFileIndex extends Logging {
val missingFiles = mutable.ArrayBuffer.empty[String]
val filteredLeafStatuses = allLeafStatuses.filterNot(
status => shouldFilterOut(status.getPath.getName))
- val resolvedLeafStatuses = filteredLeafStatuses.flatMap {
+ val resolvedLeafStatuses = filteredLeafStatuses.par.flatMap {
--- End diff --
Thanks @viirya for feedback. Yes, this method can be called on executors as
below. Do you think it's not thread-safe ?
Each partitions will have its own hadoopConf and then own fs, and nothing
is shared in this method.
sparkContext
.parallelize(serializedPaths, numParallelism)
.mapPartitions { pathStrings =>
val hadoopConf = serializableConfiguration.value
pathStrings.map(new Path(_)).toSeq.map { path =>
(path, listLeafFiles(path, hadoopConf, filter, None))
}.iterator
}.map { case (path, statuses) =>
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
