Skip to content

[SPARK-11583] [Core]MapStatus Using RoaringBitmap More Properly #9661

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
yaooqinn wants to merge 7 commits into from
Closed

[SPARK-11583] [Core]MapStatus Using RoaringBitmap More Properly #9661

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
75c3209
MapStatus Using RoaringBitmap More Properly
yaooqinn Nov 12, 2015
8f3eb7d
update roaringbitmap to 0.5.10 and runOptimize
yaooqinn Nov 13, 2015
b4b7a1c
add roaringbitmap runOptimise tests
yaooqinn Nov 13, 2015
372dc4c
Fix RoaringBitmap serialization with Kryo
gssiyankai Nov 15, 2015
b9cc414
runOptimize and trim only
yaooqinn Nov 16, 2015
64e7295
fix style
yaooqinn Nov 16, 2015
0c4bcba
Merge pull request #1 from gssiyankai/mapstatus-roaring
yaooqinn Nov 17, 2015
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
Prev Previous commit
Next Next commit
update roaringbitmap to 0.5.10 and runOptimize
  • Loading branch information
@yaooqinn
yaooqinn committed Nov 13, 2015
commit 8f3eb7d25d0ca4465a5dc50aca5954a444ef82e0
1 change: 0 additions & 1 deletion core/pom.xml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -176,7 +176,6 @@
<dependency>
<groupId>org.roaringbitmap</groupId>
<artifactId>RoaringBitmap</artifactId>
<version>0.4.5</version>
</dependency>
<dependency>
<groupId>commons-net</groupId>
Expand Down
13 changes: 7 additions & 6 deletions core/src/main/scala/org/apache/spark/scheduler/MapStatus.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -121,12 +121,11 @@ private[spark] class CompressedMapStatus(

/**
* A [[MapStatus]] implementation that only stores the average size of non-empty blocks,
* plus a bitmap for tracking which blocks are empty. During serialization, this bitmap
* is compressed.
* plus a bitmap for tracking which blocks are empty(isSparse)/non-empty(!isSparse).
*
* @param loc location where the task is being executed
* @param numNonEmptyBlocks the number of non-empty blocks
* @param markedBlocks a bitmap tracking which blocks are empty
* @param markedBlocks a bitmap tracking which blocks are empty(isSparse)/non-empty(!isSparse)
* @param avgSize average size of the non-empty blocks
*/
private[spark] class HighlyCompressedMapStatus private (
Expand Down Expand Up @@ -176,9 +175,7 @@ private[spark] object HighlyCompressedMapStatus {
var i = 0
var numNonEmptyBlocks: Int = 0
var totalSize: Long = 0
// From a compression standpoint, it shouldn't matter whether we track empty or non-empty
// blocks. From a performance standpoint, we benefit from tracking empty blocks because
// we expect that there will be far fewer of them, so we will perform fewer bitmap insertions.

val emptyBlocks = new RoaringBitmap()
val nonEmptyBlocks = new RoaringBitmap()
val totalNumBlocks = uncompressedSizes.length
Copy link
Contributor

@davies davies Nov 16, 2015

Choose a reason for hiding this comment

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

This comment still make sense, could you keep it?

Copy link
Member Author

@yaooqinn yaooqinn Nov 17, 2015

Choose a reason for hiding this comment

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

OK

Expand All @@ -199,8 +196,12 @@ private[spark] object HighlyCompressedMapStatus {
0
}
if (numNonEmptyBlocks < totalNumBlocks / 2) {
// If non-empty blocks are sparse, we track non-empty blocks and set `isSparse` to true.
nonEmptyBlocks.runOptimize()
new HighlyCompressedMapStatus(loc, numNonEmptyBlocks, nonEmptyBlocks, avgSize, isSparse = true)
} else {
// If non-empty blocks are dense, we track empty blocks and set `isSparse` to false.
emptyBlocks.runOptimize()
new HighlyCompressedMapStatus(loc, numNonEmptyBlocks, emptyBlocks, avgSize, isSparse = false)
Copy link
Contributor

@squito squito Nov 12, 2015

Choose a reason for hiding this comment

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

(a) isSparse is not the right name and (b) we might not even need to bother with tracking empty vs. non-empty, after we see the results of calling runOptimize. In fact, if they are the same, I'd suggest we change this code to consistently track nonEmpty blocks, just for clarity.

}

Expand Down
21 changes: 7 additions & 14 deletions core/src/main/scala/org/apache/spark/serializer/KryoSerializer.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,26 +21,24 @@ import java.io.{EOFException, IOException, InputStream, OutputStream}
import java.nio.ByteBuffer
import javax.annotation.Nullable

import org.roaringbitmap.{BitmapContainer, RoaringBitmap, RoaringArray, ArrayContainer}

import scala.collection.JavaConverters._
import scala.collection.mutable.ArrayBuffer
import scala.reflect.ClassTag

import com.esotericsoftware.kryo.{Kryo, KryoException}
import com.esotericsoftware.kryo.io.{Input => KryoInput, Output => KryoOutput}
import com.esotericsoftware.kryo.serializers.{JavaSerializer => KryoJavaSerializer}
import com.esotericsoftware.kryo.{Kryo, KryoException}
import com.twitter.chill.{AllScalaRegistrar, EmptyScalaKryoInstantiator}
import org.apache.avro.generic.{GenericData, GenericRecord}

import org.apache.spark._
import org.apache.spark.api.python.PythonBroadcast
import org.apache.spark.broadcast.HttpBroadcast
import org.apache.spark.network.util.ByteUnit
import org.apache.spark.scheduler.{CompressedMapStatus, HighlyCompressedMapStatus}
import org.apache.spark.storage._
import org.apache.spark.util.{Utils, BoundedPriorityQueue, SerializableConfiguration, SerializableJobConf}
import org.apache.spark.util.collection.{BitSet, CompactBuffer}
import org.apache.spark.util.{BoundedPriorityQueue, SerializableConfiguration, SerializableJobConf, Utils}
import org.roaringbitmap.RoaringBitmap

import scala.collection.JavaConverters._
import scala.collection.mutable.ArrayBuffer
Copy link
Contributor

@davies davies Nov 16, 2015

Choose a reason for hiding this comment

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

Could you update your import rules to place scale package before third parties (as before this PR)?

Copy link
Member Author

@yaooqinn yaooqinn Nov 17, 2015

Choose a reason for hiding this comment

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

OK

import scala.reflect.ClassTag

/**
* A Spark serializer that uses the [[https://code.google.com/p/kryo/ Kryo serialization library]].
Expand Down Expand Up @@ -365,11 +363,6 @@ private[serializer] object KryoSerializer {
classOf[CompressedMapStatus],
classOf[HighlyCompressedMapStatus],
classOf[RoaringBitmap],
classOf[RoaringArray],
classOf[RoaringArray.Element],
classOf[Array[RoaringArray.Element]],
classOf[ArrayContainer],
classOf[BitmapContainer],
classOf[BitSet],
classOf[CompactBuffer[_]],
classOf[BlockManagerId],
Expand Down
2 changes: 1 addition & 1 deletion pom.xml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -626,7 +626,7 @@
<dependency>
<groupId>org.roaringbitmap</groupId>
<artifactId>RoaringBitmap</artifactId>
<version>0.4.5</version>
<version>0.5.10</version>
</dependency>
<dependency>
<groupId>commons-net</groupId>
Expand Down