[SPARK-25510][TEST] Create new trait replace BenchmarkWithCodegen

[wangyum]

What changes were proposed in this pull request?

We need create a new trait to replace BenchmarkWithCodegen as BenchmarkWithCodegen extends from SparkFunSuite.

For example. when doing AggregateBenchmark refactor.
Before this change:

objectAggregateBenchmarkextendsBenchmarkBase {
lazyvalsparkSession=SparkSession.builder
.master("local[1]")
.appName(this.getClass.getSimpleName)
.config("spark.sql.shuffle.partitions", 1)
.config("spark.sql.autoBroadcastJoinThreshold", 1)
.getOrCreate()
/** Runs function `f` with whole stage codegen on and off. */defrunBenchmark(name: String, cardinality: Long)(f: =>Unit):Unit= {
valbenchmark=newBenchmark(name, cardinality, output = output)
benchmark.addCase(s"$name wholestage off", numIters =2) { iter =>
sparkSession.conf.set("spark.sql.codegen.wholeStage", value =false)
f
}
benchmark.addCase(s"$name wholestage on", numIters =5) { iter =>
sparkSession.conf.set("spark.sql.codegen.wholeStage", value =true)
f
}
benchmark.run()
}
overridedefbenchmark():Unit= {
runBenchmark("aggregate without grouping") {
valN=500L<<22
runBenchmark("agg w/o group", N) {
sparkSession.range(N).selectExpr("sum(id)").collect()
}
}
...

After this change:

objectAggregateBenchmarkextendsSqlBasedBenchmark {
overridedefbenchmark():Unit= {
runBenchmark("aggregate without grouping") {
valN=500L<<22
runBenchmark("agg w/o group", N) {
sparkSession.range(N).selectExpr("sum(id)").collect()
}
}
...

All these benchmarks will use this trait:

AggregateBenchmark
BenchmarkWideTable
JoinBenchmark
MiscBenchmark
ObjectHashAggregateExecBenchmark
SortBenchmark
UnsafeArrayDataBenchmark

How was this patch tested?

manual tests

[wangyum]

cc @cloud-fan@gengliangwang@dongjoon-hyun

[SparkQA]

Test build #96452 has finished for PR 22522 at commit 275cc6c.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[cloud-fan]

shall this extend BenchmarkBase?

[cloud-fan]

I'd remove private[benchmark] to be consistent with other benchmark classes.

[wangyum]

extends BenchmarkBase and add getSparkSession function thus subclass can build their own SparkSession:

spark/sql/core/src/test/scala/org/apache/spark/sql/execution/benchmark/RunBenchmarkWithCodegen.scala

Lines 28 to 58 in 42230b6

	traitRunBenchmarkWithCodegenextendsBenchmarkBase {
	valspark:SparkSession= getSparkSession
	/** Subclass can override this function to build their own SparkSession */
	defgetSparkSession:SparkSession= {
	SparkSession.builder()
	.master("local[1]")
	.appName(this.getClass.getCanonicalName)
	.config(SQLConf.SHUFFLE_PARTITIONS.key, 1)
	.config(SQLConf.AUTO_BROADCASTJOIN_THRESHOLD.key, 1)
	.getOrCreate()
	}
	/** Runs function `f` with whole stage codegen on and off. */
	defrunBenchmark(name: String, cardinality: Long)(f: =>Unit):Unit= {
	valbenchmark=newBenchmark(name, cardinality, output = output)
	benchmark.addCase(s"$name wholestage off", numIters =2) { iter =>
	spark.sqlContext.conf.setConf(SQLConf.WHOLESTAGE_CODEGEN_ENABLED, value =false)
	f
	}
	benchmark.addCase(s"$name wholestage on", numIters =5) { iter =>
	spark.sqlContext.conf.setConf(SQLConf.WHOLESTAGE_CODEGEN_ENABLED, value =true)
	f
	}
	benchmark.run()
	}
	}

[cloud-fan]

I think this change is necessary, but I'd like to migrate one benchmark to use this new trait as an example. We can migrate others in follow up PRs.

[wangyum]

Thanks @cloud-fan I have migrate AggregateBenchmark to use new trait.

[gengliangwang]

@wangyum I have left my comment in #22484 .
Also, should we close this one and move to #22484 ?

[wangyum]

cc @dongjoon-hyun

[SparkQA]

Test build #96783 has finished for PR 22522 at commit 20668ad.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
• trait SqlBasedBenchmark extends BenchmarkBase with SQLHelper