Github user mateiz commented on a diff in the pull request:
https://github.com/apache/spark/pull/50#discussion_r10243269
--- Diff: core/src/main/scala/org/apache/spark/CacheManager.scala ---
@@ -71,10 +71,21 @@ private[spark] class CacheManager(blockManager:
BlockManager) extends Logging {
val computedValues = rdd.computeOrReadCheckpoint(split, context)
// Persist the result, so long as the task is not running locally
if (context.runningLocally) { return computedValues }
- val elements = new ArrayBuffer[Any]
- elements ++= computedValues
- blockManager.put(key, elements, storageLevel, tellMaster = true)
- elements.iterator.asInstanceOf[Iterator[T]]
+ if (storageLevel.useDisk && !storageLevel.useMemory) {
+ blockManager.put(key, computedValues, storageLevel, tellMaster
= true)
+ return blockManager.get(key) match {
+ case Some(values) =>
+ return new InterruptibleIterator(context,
values.asInstanceOf[Iterator[T]])
+ case None =>
+ logInfo("Failure to store %s".format(key))
+ return null
+ }
+ } else {
+ val elements = new ArrayBuffer[Any]
+ elements ++= computedValues
+ blockManager.put(key, elements, storageLevel, tellMaster =
true)
+ return elements.iterator.asInstanceOf[Iterator[T]]
+ }
--- End diff --
Yes, performance and correctness are actually both reasons. The code path
for disk first writes the data to disk and then has to read and deserialize it
from there, which is slow. Also, if you used the memory store in the same way,
the store might drop it before you have a chance to call get(). See my comments
on the main discussion.
---
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 infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
