Skip to content

[SPARK-18471][CORE] New treeAggregate overload for big large aggregators #10

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
AnthonyTruchet wants to merge 1 commit into from
Closed

[SPARK-18471][CORE] New treeAggregate overload for big large aggregators #10

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 18 additions & 3 deletions core/src/main/scala/org/apache/spark/rdd/RDD.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1111,21 +1111,24 @@ abstract class RDD[T: ClassTag](
/**
* Aggregates the elements of this RDD in a multi-level tree pattern.
*
* This variant with a function generating the zero, provide for efficiently
* running on big aggregation structure like large dense vectors
Copy link
Copy Markdown

@tibidoh tibidoh Nov 18, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think where should be no mentions of vector manipulations in core APIs

Copy link
Copy Markdown
Author

@AnthonyTruchet AnthonyTruchet Nov 28, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

*
* @param depth suggested depth of the tree (default: 2)
* @see [[org.apache.spark.rdd.RDD#aggregate]]
*/
def treeAggregate[U: ClassTag](zeroValue: U)(
def treeAggregateWithZeroGenerator[U: ClassTag](zeroValueGenerator: () => U)(
Copy link
Copy Markdown

@superbobry superbobry Nov 18, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You could use a lazy parameter here instead of an explicit closure, see e.g. this blog post.

Copy link
Copy Markdown
Author

@AnthonyTruchet AnthonyTruchet Nov 21, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Great suggestion ! I'm not sure however how lazy val A a/o => A behave when serialized to an executor... Any idea ?

Copy link
Copy Markdown

@tibidoh tibidoh Nov 21, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

According to the blog post, it should be exactly what we want. But yeah, we need to test

Copy link
Copy Markdown
Author

@AnthonyTruchet AnthonyTruchet Nov 21, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Let's discuss/test this live because I reckon that the value will be instantiated when thee serializer is called, which is NOT what we want.

seqOp: (U, T) => U,
combOp: (U, U) => U,
depth: Int = 2): U = withScope {
require(depth >= 1, s"Depth must be greater than or equal to 1 but got $depth.")
if (partitions.length == 0) {
Utils.clone(zeroValue, context.env.closureSerializer.newInstance())
Utils.clone(zeroValueGenerator(), context.env.closureSerializer.newInstance())
} else {
val cleanSeqOp = context.clean(seqOp)
val cleanCombOp = context.clean(combOp)
val aggregatePartition =
(it: Iterator[T]) => it.aggregate(zeroValue)(cleanSeqOp, cleanCombOp)
(it: Iterator[T]) => it.aggregate(zeroValueGenerator())(cleanSeqOp, cleanCombOp)
var partiallyAggregated = mapPartitions(it => Iterator(aggregatePartition(it)))
var numPartitions = partiallyAggregated.partitions.length
val scale = math.max(math.ceil(math.pow(numPartitions, 1.0 / depth)).toInt, 2)
Expand All @@ -1144,6 +1147,18 @@ abstract class RDD[T: ClassTag](
}
}

/**
* Aggregates the elements of this RDD in a multi-level tree pattern.
*
* @param depth suggested depth of the tree (default: 2)
* @see [[org.apache.spark.rdd.RDD#aggregate]]
*/
def treeAggregate[U: ClassTag](zeroValue: U)(
Copy link
Copy Markdown

@tibidoh tibidoh Nov 18, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

OPT: Please keep BFS order of method declarations

Copy link
Copy Markdown
Author

@AnthonyTruchet AnthonyTruchet Nov 21, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

BFS ? What do you mean ?

Copy link
Copy Markdown

@tibidoh tibidoh Nov 21, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Breadth first (search) order

Copy link
Copy Markdown
Author

@AnthonyTruchet AnthonyTruchet Nov 28, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

seqOp: (U, T) => U,
combOp: (U, U) => U,
depth: Int = 2): U =
treeAggregateWithZeroGenerator(() => zeroValue)(seqOp, combOp, depth)

/**
* Return the number of elements in the RDD.
*/
Expand Down