Clean shuffle data (#312) (#322)

* Cleanup Spark shuffle data after data is consumed



* update comments



---------


(cherry picked from commit 222e473)

Signed-off-by: Peng Huo <penghuo@gmail.com>
Signed-off-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

Author: opensearch-trigger-bot[bot]

spark-sql-application/src/main/scala/org/apache/spark/sql/FlintJob.scala

@@ -6,24 +6,15 @@
 // defined in spark package so that I can use ThreadUtils
 package org.apache.spark.sql
 
-import java.util.Locale
 import java.util.concurrent.atomic.AtomicInteger
 
-import org.opensearch.client.{RequestOptions, RestHighLevelClient}
-import org.opensearch.cluster.metadata.MappingMetadata
-import org.opensearch.common.settings.Settings
-import org.opensearch.common.xcontent.XContentType
-import org.opensearch.flint.core.{FlintClient, FlintClientBuilder, FlintOptions}
-import org.opensearch.flint.core.metadata.FlintMetadata
 import org.opensearch.flint.core.metrics.MetricConstants
 import org.opensearch.flint.core.metrics.MetricsUtil.registerGauge
 import play.api.libs.json._
 
-import org.apache.spark.SparkConf
 import org.apache.spark.internal.Logging
-import org.apache.spark.sql.catalyst.parser.ParseException
 import org.apache.spark.sql.flint.config.FlintSparkConf
-import org.apache.spark.sql.types.{StructField, _}
+import org.apache.spark.sql.types._
 
 /**
  * Spark SQL Application entrypoint

spark-sql-application/src/main/scala/org/apache/spark/sql/FlintJobExecutor.scala

@@ -7,26 +7,18 @@ package org.apache.spark.sql
 
 import java.util.Locale
 
-import scala.concurrent.{ExecutionContext, Future, TimeoutException}
-import scala.concurrent.duration.{Duration, MINUTES}
-
 import com.amazonaws.services.s3.model.AmazonS3Exception
 import org.apache.commons.text.StringEscapeUtils.unescapeJava
-import org.opensearch.flint.core.{FlintClient, IRestHighLevelClient}
-import org.opensearch.flint.core.metadata.FlintMetadata
+import org.opensearch.flint.core.IRestHighLevelClient
 import org.opensearch.flint.core.metrics.MetricConstants
 import org.opensearch.flint.core.metrics.MetricsUtil.incrementCounter
-import play.api.libs.json.{JsArray, JsBoolean, JsObject, Json, JsString, JsValue}
+import play.api.libs.json._
 
 import org.apache.spark.{SparkConf, SparkException}
 import org.apache.spark.internal.Logging
-import org.apache.spark.sql.FlintREPL.envinromentProvider
 import org.apache.spark.sql.catalyst.parser.ParseException
-import org.apache.spark.sql.execution.datasources.DataSource
-import org.apache.spark.sql.flint.config.FlintSparkConf
-import org.apache.spark.sql.types.{ArrayType, LongType, StringType, StructField, StructType}
-import org.apache.spark.sql.util.{DefaultThreadPoolFactory, EnvironmentProvider, RealEnvironment, RealTimeProvider, ThreadPoolFactory, TimeProvider}
-import org.apache.spark.util.ThreadUtils
+import org.apache.spark.sql.types._
+import org.apache.spark.sql.util._
 
 trait FlintJobExecutor {
   this: Logging =>
@@ -157,7 +149,8 @@ trait FlintJobExecutor {
       query: String,
       sessionId: String,
       startTime: Long,
-      timeProvider: TimeProvider): DataFrame = {
+      timeProvider: TimeProvider,
+      cleaner: Cleaner): DataFrame = {
     // Create the schema dataframe
     val schemaRows = result.schema.fields.map { field =>
       Row(field.name, field.dataType.typeName)
@@ -192,6 +185,11 @@ trait FlintJobExecutor {
     val resultSchemaToSave = resultSchema.toJSON.collect.toList.map(_.replaceAll("\"", "'"))
     val endTime = timeProvider.currentEpochMillis()
 
+    // https://github.com/opensearch-project/opensearch-spark/issues/302. Clean shuffle data
+    // after consumed the query result. Streaming query shuffle data is cleaned after each
+    // microBatch execution.
+    cleaner.cleanUp(spark)
+
     // Create the data rows
     val rows = Seq(
       (
@@ -375,7 +373,8 @@ trait FlintJobExecutor {
       query: String,
       dataSource: String,
       queryId: String,
-      sessionId: String): DataFrame = {
+      sessionId: String,
+      streaming: Boolean): DataFrame = {
     // Execute SQL query
     val startTime = System.currentTimeMillis()
     // we have to set job group in the same thread that started the query according to spark doc
@@ -390,7 +389,8 @@ trait FlintJobExecutor {
       query,
       sessionId,
       startTime,
-      currentTimeProvider)
+      currentTimeProvider,
+      CleanerFactory.cleaner(streaming))
   }
 
   private def handleQueryException(

spark-sql-application/src/main/scala/org/apache/spark/sql/FlintREPL.scala

@@ -22,15 +22,13 @@ import org.opensearch.flint.app.{FlintCommand, FlintInstance}
 import org.opensearch.flint.app.FlintInstance.formats
 import org.opensearch.flint.core.FlintOptions
 import org.opensearch.flint.core.metrics.MetricConstants
-import org.opensearch.flint.core.metrics.MetricsUtil.{decrementCounter, getTimerContext, incrementCounter, registerGauge, stopTimer}
+import org.opensearch.flint.core.metrics.MetricsUtil.{getTimerContext, incrementCounter, registerGauge, stopTimer}
 import org.opensearch.flint.core.storage.{FlintReader, OpenSearchUpdater}
 import org.opensearch.search.sort.SortOrder
 
 import org.apache.spark.SparkConf
 import org.apache.spark.internal.Logging
-import org.apache.spark.sql.FlintJob.createSparkSession
 import org.apache.spark.sql.flint.config.FlintSparkConf
-import org.apache.spark.sql.flint.config.FlintSparkConf.REPL_INACTIVITY_TIMEOUT_MILLIS
 import org.apache.spark.sql.util.{DefaultShutdownHookManager, ShutdownHookManagerTrait}
 import org.apache.spark.util.ThreadUtils
 
@@ -829,7 +827,13 @@ object FlintREPL extends Logging with FlintJobExecutor {
         startTime)
     } else {
       val futureQueryExecution = Future {
-        executeQuery(spark, flintCommand.query, dataSource, flintCommand.queryId, sessionId)
+        executeQuery(
+          spark,
+          flintCommand.query,
+          dataSource,
+          flintCommand.queryId,
+          sessionId,
+          false)
       }(executionContext)
 
       // time out after 10 minutes

spark-sql-application/src/main/scala/org/apache/spark/sql/JobOperator.scala

@@ -14,13 +14,10 @@ import scala.util.{Failure, Success, Try}
 
 import org.opensearch.flint.core.metrics.MetricConstants
 import org.opensearch.flint.core.metrics.MetricsUtil.incrementCounter
-import org.opensearch.flint.core.storage.OpenSearchUpdater
 
-import org.apache.spark.SparkConf
 import org.apache.spark.internal.Logging
-import org.apache.spark.sql.FlintJob.createSparkSession
-import org.apache.spark.sql.FlintREPL.{executeQuery, logInfo, threadPoolFactory, updateFlintInstanceBeforeShutdown}
 import org.apache.spark.sql.flint.config.FlintSparkConf
+import org.apache.spark.sql.util.ShuffleCleaner
 import org.apache.spark.util.ThreadUtils
 
 case class JobOperator(
@@ -53,7 +50,7 @@ case class JobOperator(
       val futureMappingCheck = Future {
         checkAndCreateIndex(osClient, resultIndex)
       }
-      val data = executeQuery(spark, query, dataSource, "", "")
+      val data = executeQuery(spark, query, dataSource, "", "", streaming)
 
       val mappingCheckResult = ThreadUtils.awaitResult(futureMappingCheck, Duration(1, MINUTES))
       dataToWrite = Some(mappingCheckResult match {
@@ -92,6 +89,8 @@ case class JobOperator(
     try {
       // Wait for streaming job complete if no error and there is streaming job running
       if (!exceptionThrown && streaming && spark.streams.active.nonEmpty) {
+        // Clean Spark shuffle data after each microBatch.
+        spark.streams.addListener(new ShuffleCleaner(spark))
         // wait if any child thread to finish before the main thread terminates
         spark.streams.awaitAnyTermination()
       }
@@ -149,4 +148,5 @@ case class JobOperator(
       case false => incrementCounter(MetricConstants.STREAMING_SUCCESS_METRIC)
     }
   }
+
 }

spark-sql-application/src/main/scala/org/apache/spark/sql/util/ShuffleCleaner.scala

@@ -0,0 +1,57 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package org.apache.spark.sql.util
+
+import org.apache.spark.{MapOutputTrackerMaster, SparkEnv}
+import org.apache.spark.internal.Logging
+import org.apache.spark.sql.SparkSession
+import org.apache.spark.sql.streaming.StreamingQueryListener
+
+/**
+ * Clean Spark shuffle data after each microBatch.
+ * https://github.com/opensearch-project/opensearch-spark/issues/302
+ */
+class ShuffleCleaner(spark: SparkSession) extends StreamingQueryListener with Logging {
+
+  override def onQueryStarted(event: StreamingQueryListener.QueryStartedEvent): Unit = {}
+
+  override def onQueryProgress(event: StreamingQueryListener.QueryProgressEvent): Unit = {
+    ShuffleCleaner.cleanUp(spark)
+  }
+
+  override def onQueryTerminated(event: StreamingQueryListener.QueryTerminatedEvent): Unit = {}
+}
+
+trait Cleaner {
+  def cleanUp(spark: SparkSession)
+}
+
+object CleanerFactory {
+  def cleaner(streaming: Boolean): Cleaner = {
+    if (streaming) NoOpCleaner else ShuffleCleaner
+  }
+}
+
+/**
+ * No operation cleaner.
+ */
+object NoOpCleaner extends Cleaner {
+  override def cleanUp(spark: SparkSession): Unit = {}
+}
+
+/**
+ * Spark shuffle data cleaner.
+ */
+object ShuffleCleaner extends Cleaner with Logging {
+  def cleanUp(spark: SparkSession): Unit = {
+    logInfo("Before cleanUp Shuffle")
+    val cleaner = spark.sparkContext.cleaner
+    val masterTracker = SparkEnv.get.mapOutputTracker.asInstanceOf[MapOutputTrackerMaster]
+    val shuffleIds = masterTracker.shuffleStatuses.keys.toSet
+    shuffleIds.foreach(shuffleId => cleaner.foreach(c => c.doCleanupShuffle(shuffleId, true)))
+    logInfo("After cleanUp Shuffle")
+  }
+}

spark-sql-application/src/test/scala/org/apache/spark/sql/FlintJobTest.scala

@@ -7,7 +7,7 @@ package org.apache.spark.sql
 
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.sql.types._
-import org.apache.spark.sql.util.MockTimeProvider
+import org.apache.spark.sql.util.{CleanerFactory, MockTimeProvider}
 
 class FlintJobTest extends SparkFunSuite with JobMatchers {
 
@@ -76,7 +76,8 @@ class FlintJobTest extends SparkFunSuite with JobMatchers {
         "select 1",
         "20",
         currentTime - queryRunTime,
-        new MockTimeProvider(currentTime))
+        new MockTimeProvider(currentTime),
+        CleanerFactory.cleaner(false))
     assertEqualDataframe(expected, result)
   }
 

spark-sql-application/src/test/scala/org/apache/spark/sql/util/CleanerFactoryTest.scala

@@ -0,0 +1,23 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+package org.apache.spark.sql.util
+
+import org.scalatest.matchers.should.Matchers
+
+import org.apache.spark.SparkFunSuite
+
+class CleanerFactoryTest extends SparkFunSuite with Matchers {
+
+  test("CleanerFactory should return NoOpCleaner when streaming is true") {
+    val cleaner = CleanerFactory.cleaner(streaming = true)
+    cleaner shouldBe NoOpCleaner
+  }
+
+  test("CleanerFactory should return ShuffleCleaner when streaming is false") {
+    val cleaner = CleanerFactory.cleaner(streaming = false)
+    cleaner shouldBe ShuffleCleaner
+  }
+}