Skip to content
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

[SPARK-23195] [SQL] Keep the Hint of Cached Data #20368

Closed
wants to merge 1 commit into from
Closed

[SPARK-23195] [SQL] Keep the Hint of Cached Data #20368

Show file tree
Hide file tree
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
4 changes: 2 additions & 2 deletions sql/core/src/main/scala/org/apache/spark/sql/execution/columnar/InMemoryRelation.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -63,7 +63,7 @@ case class InMemoryRelation(
tableName: Option[String])(
@transient var _cachedColumnBuffers: RDD[CachedBatch] = null,
val batchStats: LongAccumulator = child.sqlContext.sparkContext.longAccumulator,
statsOfPlanToCache: Statistics = null)
statsOfPlanToCache: Statistics)
Copy link
Contributor

Choose a reason for hiding this comment

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

leave no default value is fine, we do not any default value actually

Copy link
Member Author

Choose a reason for hiding this comment

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

Setting null by default is risky, because we might hit NullPointerException .

extends logical.LeafNode with MultiInstanceRelation {

override protected def innerChildren: Seq[SparkPlan] = Seq(child)
Expand All @@ -77,7 +77,7 @@ case class InMemoryRelation(
// Underlying columnar RDD hasn't been materialized, use the stats from the plan to cache
statsOfPlanToCache
} else {
Statistics(sizeInBytes = batchStats.value.longValue)
Statistics(sizeInBytes = batchStats.value.longValue, hints = statsOfPlanToCache.hints)
Copy link
Contributor

Choose a reason for hiding this comment

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

Why don't you simply statsOfPlanToCache.copy(sizeInBytes = batchStats.value.longValue)?

Copy link
Member Author

@gatorsmile gatorsmile Jan 23, 2018 •
edited
Loading

Choose a reason for hiding this comment

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

That is misleading. Conceptually, that is wrong. The values should be filled by the actual values from the materialized results.

}
}

Expand Down
16 changes: 16 additions & 0 deletions sql/core/src/test/scala/org/apache/spark/sql/execution/joins/BroadcastJoinSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -126,6 +126,22 @@ class BroadcastJoinSuite extends QueryTest with SQLTestUtils {
}
}

test("broadcast hint is retained in a cached plan") {
Seq(true, false).foreach { materialized =>
withSQLConf(SQLConf.AUTO_BROADCASTJOIN_THRESHOLD.key -> "-1") {
val df1 = spark.createDataFrame(Seq((1, "4"), (2, "2"))).toDF("key", "value")
Copy link
Contributor

Choose a reason for hiding this comment

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

Is spark.createDataFrame(...) wrapper really required? I thought Seq((1, "4"), (2, "2")).toDF("key", "value") would just work fine.

Copy link
Member Author

Choose a reason for hiding this comment

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

That should not matter.

val df2 = spark.createDataFrame(Seq((1, "1"), (2, "2"))).toDF("key", "value")
broadcast(df2).cache()
if (materialized) df2.collect()
Copy link
Member Author

Choose a reason for hiding this comment

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

This PR #19864 accidentally fixes the issue when the plan is not materialized. However, it does not resolve the issue when the cached plan is materialized.

val df3 = df1.join(df2, Seq("key"), "inner")
Copy link
Contributor

Choose a reason for hiding this comment

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

val df3 = df1.join(df2, "key")? inner is implied, isn't it? (I'm proposing the change as this and other tests could be easily used as a learning tool to master Spark SQL's API)

Copy link
Member Author

Choose a reason for hiding this comment

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

That should not matter.

Copy link
Contributor

Choose a reason for hiding this comment

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

I tend to agree that tests are also examples for Spark users, we should pick the recommended usages.

Copy link
Member Author

Choose a reason for hiding this comment

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

All the other cases are creating Dataframes like this. Anyway, I changed all of them in the new PR.

val numBroadCastHashJoin = df3.queryExecution.executedPlan.collect {
case b: BroadcastHashJoinExec => b
}.size
assert(numBroadCastHashJoin === 1)
}
}
}

private def assertBroadcastJoin(df : Dataset[Row]) : Unit = {
val df1 = spark.createDataFrame(Seq((1, "4"), (2, "2"))).toDF("key", "value")
val joined = df1.join(df, Seq("key"), "inner")
Expand Down