Github user srowen commented on a diff in the pull request:
https://github.com/apache/spark/pull/18441#discussion_r125269807
--- Diff: core/src/main/scala/org/apache/spark/rdd/BinaryFileRDD.scala ---
@@ -35,8 +36,12 @@ private[spark] class BinaryFileRDD[T](
extends NewHadoopRDD[String, T](sc, inputFormatClass, keyClass,
valueClass, conf) {
override def getPartitions: Array[Partition] = {
- val inputFormat = inputFormatClass.newInstance
val conf = getConf
+ // setMinPartitions below will call FileInputFormat.listStatus(),
which can be quite slow when
+ // traversing a large number of directories and files. Parallelize it.
+ conf.setIfUnset(FileInputFormat.LIST_STATUS_NUM_THREADS,
+ Runtime.getRuntime.availableProcessors().toString)
--- End diff --
@cloud-fan @kiszk this will happen on the driver, where it goes to assess
the size of the data in order to compute the desired number of partitions. I
don't know what the right value is here, to be honest. It won't actually use a
ton of CPU because these threads will mostly be waiting for the external FS
processes.
I didn't think it's worth yet another config, and didn't know what else to
base it on. The number of driver cores maybe?
What is `CPU_CORES_PER_EXECUTOR` by the way, I don't see that in the code?
I can check for `spark.driver.cores` in the Spark config here.
---
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]
