Github user squito commented on a diff in the pull request:
https://github.com/apache/spark/pull/6648#discussion_r32024100
--- Diff:
core/src/main/scala/org/apache/spark/shuffle/hash/HashShuffleManager.scala ---
@@ -50,15 +53,30 @@ private[spark] class HashShuffleManager(conf:
SparkConf) extends ShuffleManager
}
/** Get a writer for a given partition. Called on executors by map
tasks. */
- override def getWriter[K, V](handle: ShuffleHandle, mapId: Int, context:
TaskContext)
- : ShuffleWriter[K, V] = {
- new HashShuffleWriter(
- shuffleBlockResolver, handle.asInstanceOf[BaseShuffleHandle[K, V,
_]], mapId, context)
+ override def getWriter[K, V](
+ handle: ShuffleHandle,
+ mapId: Int,
+ stageAttemptId: Int,
+ context: TaskContext): ShuffleWriter[K, V] = {
+ addShuffleAttempt(handle.shuffleId, stageAttemptId)
+ new HashShuffleWriter(shuffleBlockResolver,
handle.asInstanceOf[BaseShuffleHandle[K, V, _]],
--- End diff --
each ShuffleManager must call `addShuffleAttempt` in `getWriter`, to ensure
that all attempts get cleaned up when the shuffle is unregistered.
now that I'm writing this I'm wondering ... should I make a new abstract
method `getWriterHelper` or something, which only has to get the writer, and
change `ShuffleManger.getWriter` to be a concrete implementation like:
```
def getWriter[K, V](
handle: ShuffleHandle,
mapId: Int,
stageAttemptId: Int,
context: TaskContext): ShuffleWriter[K, V] = {
addShuffleAttempt(handle.shuffleId, stageAttemptId)
getWriterHelper(handle, mapId, stageAttemptId, context)
}
```
?
though I'm not so sure it will really help b/c of the way
`UnsafeShuffleManager` delegates to `SortShuffleManager`, and each
implementation has to implement other cleanup logic in any case ...
---
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]
