[SPARK-24525][SS] Provide an option to limit number of rows in a MemorySink

sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/memory.scala

@@ -33,6 +33,7 @@ import org.apache.spark.sql.catalyst.expressions.{Attribute, UnsafeRow}
 import org.apache.spark.sql.catalyst.plans.logical.{LeafNode, LogicalPlan, Statistics}
 import org.apache.spark.sql.catalyst.plans.logical.statsEstimation.EstimationUtils
 import org.apache.spark.sql.catalyst.streaming.InternalOutputModes._
+import org.apache.spark.sql.sources.v2.DataSourceOptions
 import org.apache.spark.sql.sources.v2.reader.{InputPartition, InputPartitionReader, SupportsScanUnsafeRow}
 import org.apache.spark.sql.sources.v2.reader.streaming.{MicroBatchReader, Offset => OffsetV2}
 import org.apache.spark.sql.streaming.OutputMode
@@ -228,19 +229,45 @@ trait MemorySinkBase extends BaseStreamingSink {
   def latestBatchId: Option[Long]
 }
 
+/**
+ * Companion object to MemorySinkBase.
+ */
+object MemorySinkBase {
+  val MAX_MEMORY_SINK_ROWS = "maxMemorySinkRows"
+  val MAX_MEMORY_SINK_ROWS_DEFAULT = -1
+
+  /**
+   * Gets the max number of rows a MemorySink should store. This number is based on the memory
+   * sink row limit if it is set. If not, there is no limit.
+   * @param options Options for writing from which we get the max rows option
+   * @return The maximum number of rows a memorySink should store, or None for no limit.
+   */
+  def getMemorySinkCapacity(options: DataSourceOptions): Option[Int] = {
+    val maxRows = options.getInt(MAX_MEMORY_SINK_ROWS, MAX_MEMORY_SINK_ROWS_DEFAULT)
+    if (maxRows >= 0) Some(maxRows) else None
+  }
+}
+
+
 /**
  * A sink that stores the results in memory. This [[Sink]] is primarily intended for use in unit
  * tests and does not provide durability.
  */
-class MemorySink(val schema: StructType, outputMode: OutputMode) extends Sink
-  with MemorySinkBase with Logging {
+class MemorySink(val schema: StructType, outputMode: OutputMode, options: DataSourceOptions)
+  extends Sink with MemorySinkBase with Logging {
 
   private case class AddedData(batchId: Long, data: Array[Row])
 
   /** An order list of batches that have been written to this [[Sink]]. */
   @GuardedBy("this")
   private val batches = new ArrayBuffer[AddedData]()
 
+  /** The number of rows in this MemorySink. */
+  private var numRows = 0
+
+  /** The capacity in rows of this sink. */
+  val sinkCapacity: Option[Int] = MemorySinkBase.getMemorySinkCapacity(options)
+
   /** Returns all rows that are stored in this [[Sink]]. */
   def allData: Seq[Row] = synchronized {
     batches.flatMap(_.data)
@@ -273,14 +300,26 @@ class MemorySink(val schema: StructType, outputMode: OutputMode) extends Sink
       logDebug(s"Committing batch $batchId to $this")
       outputMode match {
         case Append | Update =>
-          val rows = AddedData(batchId, data.collect())
-          synchronized { batches += rows }
+          var rowsToAdd = data.collect()
+          synchronized {
+            if (sinkCapacity.isDefined) {
+              rowsToAdd = truncateRowsIfNeeded(rowsToAdd, sinkCapacity.get - numRows, batchId)
+            }
+            val rows = AddedData(batchId, rowsToAdd)
+            batches += rows
+            numRows += rowsToAdd.length
+          }
 
         case Complete =>
-          val rows = AddedData(batchId, data.collect())
+          var rowsToAdd = data.collect()
           synchronized {
+            if (sinkCapacity.isDefined) {
+              rowsToAdd = truncateRowsIfNeeded(rowsToAdd, sinkCapacity.get, batchId)
+            }
+            val rows = AddedData(batchId, rowsToAdd)
             batches.clear()
             batches += rows
+            numRows = rowsToAdd.length
           }
 
         case _ =>
@@ -294,6 +333,16 @@ class MemorySink(val schema: StructType, outputMode: OutputMode) extends Sink
 
   def clear(): Unit = synchronized {
     batches.clear()
+    numRows = 0
+  }
+
+  private def truncateRowsIfNeeded(rows: Array[Row], maxRows: Int, batchId: Long): Array[Row] = {
+    if (rows.length > maxRows) {
+      logWarning(s"Truncating batch $batchId to $maxRows rows")
+      rows.take(maxRows)
+    } else {
+      rows
+    }
   }
 
   override def toString(): String = "MemorySink"

sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/sources/memoryV2.scala

@@ -46,7 +46,7 @@ class MemorySinkV2 extends DataSourceV2 with StreamWriteSupport with MemorySinkB
       schema: StructType,
       mode: OutputMode,
       options: DataSourceOptions): StreamWriter = {
-    new MemoryStreamWriter(this, mode)
+    new MemoryStreamWriter(this, mode, options)
   }
 
   private case class AddedData(batchId: Long, data: Array[Row])
@@ -55,6 +55,9 @@ class MemorySinkV2 extends DataSourceV2 with StreamWriteSupport with MemorySinkB
   @GuardedBy("this")
   private val batches = new ArrayBuffer[AddedData]()
 
+  /** The number of rows in this MemorySink. */
+  private var numRows = 0
+
   /** Returns all rows that are stored in this [[Sink]]. */
   def allData: Seq[Row] = synchronized {
     batches.flatMap(_.data)
@@ -81,22 +84,35 @@ class MemorySinkV2 extends DataSourceV2 with StreamWriteSupport with MemorySinkB
     }.mkString("\n")
   }
 
-  def write(batchId: Long, outputMode: OutputMode, newRows: Array[Row]): Unit = {
+  def write(batchId: Long, outputMode: OutputMode, newRows: Array[Row], sinkCapacity: Option[Int])
+  : Unit = {
     val notCommitted = synchronized {
       latestBatchId.isEmpty || batchId > latestBatchId.get
     }
     if (notCommitted) {
       logDebug(s"Committing batch $batchId to $this")
       outputMode match {
         case Append | Update =>
-          val rows = AddedData(batchId, newRows)
-          synchronized { batches += rows }
+          synchronized {
+            var rowsToAdd = newRows
+            if (sinkCapacity.isDefined) {
+              rowsToAdd = truncateRowsIfNeeded(rowsToAdd, sinkCapacity.get - numRows, batchId)
+            }
+            val rows = AddedData(batchId, rowsToAdd)
+            batches += rows
+            numRows += rowsToAdd.length
+          }
 
         case Complete =>
-          val rows = AddedData(batchId, newRows)
           synchronized {
+            var rowsToAdd = newRows
+            if (sinkCapacity.isDefined) {
+              rowsToAdd = truncateRowsIfNeeded(rowsToAdd, sinkCapacity.get, batchId)
+            }
+            val rows = AddedData(batchId, rowsToAdd)
             batches.clear()
             batches += rows
+            numRows = rowsToAdd.length
           }
 
         case _ =>
@@ -110,40 +126,61 @@ class MemorySinkV2 extends DataSourceV2 with StreamWriteSupport with MemorySinkB
 
   def clear(): Unit = synchronized {
     batches.clear()
+    numRows = 0
+  }
+
+  private def truncateRowsIfNeeded(rows: Array[Row], maxRows: Int, batchId: Long): Array[Row] = {
+    if (rows.length > maxRows) {
+      logWarning(s"Truncating batch $batchId to $maxRows rows")
+      rows.take(maxRows)
+    } else {
+      rows
+    }
   }
 
   override def toString(): String = "MemorySinkV2"
 }
 
 case class MemoryWriterCommitMessage(partition: Int, data: Seq[Row]) extends WriterCommitMessage {}
 
-class MemoryWriter(sink: MemorySinkV2, batchId: Long, outputMode: OutputMode)
+class MemoryWriter(
+    sink: MemorySinkV2,
+    batchId: Long,
+    outputMode: OutputMode,
+    options: DataSourceOptions)
   extends DataSourceWriter with Logging {
 
+  val sinkCapacity: Option[Int] = MemorySinkBase.getMemorySinkCapacity(options)
+
   override def createWriterFactory: MemoryWriterFactory = MemoryWriterFactory(outputMode)
 
   def commit(messages: Array[WriterCommitMessage]): Unit = {
     val newRows = messages.flatMap {
       case message: MemoryWriterCommitMessage => message.data
     }
-    sink.write(batchId, outputMode, newRows)
+    sink.write(batchId, outputMode, newRows, sinkCapacity)
   }
 
   override def abort(messages: Array[WriterCommitMessage]): Unit = {
     // Don't accept any of the new input.
   }
 }
 
-class MemoryStreamWriter(val sink: MemorySinkV2, outputMode: OutputMode)
+class MemoryStreamWriter(
+    val sink: MemorySinkV2,
+    outputMode: OutputMode,
+    options: DataSourceOptions)
   extends StreamWriter {
 
+  val sinkCapacity: Option[Int] = MemorySinkBase.getMemorySinkCapacity(options)
+
   override def createWriterFactory: MemoryWriterFactory = MemoryWriterFactory(outputMode)
 
   override def commit(epochId: Long, messages: Array[WriterCommitMessage]): Unit = {
     val newRows = messages.flatMap {
       case message: MemoryWriterCommitMessage => message.data
     }
-    sink.write(epochId, outputMode, newRows)
+    sink.write(epochId, outputMode, newRows, sinkCapacity)
   }
 
   override def abort(epochId: Long, messages: Array[WriterCommitMessage]): Unit = {

sql/core/src/main/scala/org/apache/spark/sql/streaming/DataStreamWriter.scala

@@ -29,7 +29,7 @@ import org.apache.spark.sql.execution.datasources.DataSource
 import org.apache.spark.sql.execution.streaming._
 import org.apache.spark.sql.execution.streaming.continuous.ContinuousTrigger
 import org.apache.spark.sql.execution.streaming.sources.{ForeachWriterProvider, MemoryPlanV2, MemorySinkV2}
-import org.apache.spark.sql.sources.v2.StreamWriteSupport
+import org.apache.spark.sql.sources.v2.{DataSourceOptions, StreamWriteSupport}
 
 /**
  * Interface used to write a streaming `Dataset` to external storage systems (e.g. file systems,
@@ -249,7 +249,7 @@ final class DataStreamWriter[T] private[sql](ds: Dataset[T]) {
           val r = Dataset.ofRows(df.sparkSession, new MemoryPlanV2(s, df.schema.toAttributes))
           (s, r)
         case _ =>
-          val s = new MemorySink(df.schema, outputMode)
+          val s = new MemorySink(df.schema, outputMode, new DataSourceOptions(extraOptions.asJava))
           val r = Dataset.ofRows(df.sparkSession, new MemoryPlan(s))
           (s, r)
       }

sql/core/src/test/scala/org/apache/spark/sql/execution/streaming/MemorySinkSuite.scala

@@ -17,11 +17,13 @@
 
 package org.apache.spark.sql.execution.streaming
 
+import scala.collection.JavaConverters._
 import scala.language.implicitConversions
 
 import org.scalatest.BeforeAndAfter
 
 import org.apache.spark.sql._
+import org.apache.spark.sql.sources.v2.DataSourceOptions
 import org.apache.spark.sql.streaming.{OutputMode, StreamTest}
 import org.apache.spark.sql.types.{IntegerType, StructField, StructType}
 import org.apache.spark.util.Utils
@@ -36,7 +38,7 @@ class MemorySinkSuite extends StreamTest with BeforeAndAfter {
 
   test("directly add data in Append output mode") {
     implicit val schema = new StructType().add(new StructField("value", IntegerType))
-    val sink = new MemorySink(schema, OutputMode.Append)
+    val sink = new MemorySink(schema, OutputMode.Append, DataSourceOptions.empty())
 
     // Before adding data, check output
     assert(sink.latestBatchId === None)
@@ -68,9 +70,35 @@ class MemorySinkSuite extends StreamTest with BeforeAndAfter {
     checkAnswer(sink.allData, 1 to 9)
   }
 
+  test("directly add data in Append output mode with row limit") {
+    implicit val schema = new StructType().add(new StructField("value", IntegerType))
+
+    var optionsMap = new scala.collection.mutable.HashMap[String, String]
+    optionsMap.put(MemorySinkBase.MAX_MEMORY_SINK_ROWS, 5.toString())
+    var options = new DataSourceOptions(optionsMap.toMap.asJava)
+    val sink = new MemorySink(schema, OutputMode.Append, options)
+
+    // Before adding data, check output
+    assert(sink.latestBatchId === None)
+    checkAnswer(sink.latestBatchData, Seq.empty)
+    checkAnswer(sink.allData, Seq.empty)
+
+    // Add batch 0 and check outputs
+    sink.addBatch(0, 1 to 3)
+    assert(sink.latestBatchId === Some(0))
+    checkAnswer(sink.latestBatchData, 1 to 3)
+    checkAnswer(sink.allData, 1 to 3)
+
+    // Add batch 1 and check outputs
+    sink.addBatch(1, 4 to 6)
+    assert(sink.latestBatchId === Some(1))
+    checkAnswer(sink.latestBatchData, 4 to 5)
+    checkAnswer(sink.allData, 1 to 5)     // new data should not go over the limit
+  }
+
   test("directly add data in Update output mode") {
     implicit val schema = new StructType().add(new StructField("value", IntegerType))
-    val sink = new MemorySink(schema, OutputMode.Update)
+    val sink = new MemorySink(schema, OutputMode.Update, DataSourceOptions.empty())
 
     // Before adding data, check output
     assert(sink.latestBatchId === None)
@@ -104,7 +132,7 @@ class MemorySinkSuite extends StreamTest with BeforeAndAfter {
 
   test("directly add data in Complete output mode") {
     implicit val schema = new StructType().add(new StructField("value", IntegerType))
-    val sink = new MemorySink(schema, OutputMode.Complete)
+    val sink = new MemorySink(schema, OutputMode.Complete, DataSourceOptions.empty())
 
     // Before adding data, check output
     assert(sink.latestBatchId === None)
@@ -136,6 +164,32 @@ class MemorySinkSuite extends StreamTest with BeforeAndAfter {
     checkAnswer(sink.allData, 7 to 9)
   }
 
+  test("directly add data in Complete output mode with row limit") {
+    implicit val schema = new StructType().add(new StructField("value", IntegerType))
+
+    var optionsMap = new scala.collection.mutable.HashMap[String, String]
+    optionsMap.put(MemorySinkBase.MAX_MEMORY_SINK_ROWS, 5.toString())
+    var options = new DataSourceOptions(optionsMap.toMap.asJava)
+    val sink = new MemorySink(schema, OutputMode.Complete, options)
+
+    // Before adding data, check output
+    assert(sink.latestBatchId === None)
+    checkAnswer(sink.latestBatchData, Seq.empty)
+    checkAnswer(sink.allData, Seq.empty)
+
+    // Add batch 0 and check outputs
+    sink.addBatch(0, 1 to 3)
+    assert(sink.latestBatchId === Some(0))
+    checkAnswer(sink.latestBatchData, 1 to 3)
+    checkAnswer(sink.allData, 1 to 3)
+
+    // Add batch 1 and check outputs
+    sink.addBatch(1, 4 to 10)
+    assert(sink.latestBatchId === Some(1))
+    checkAnswer(sink.latestBatchData, 4 to 8)
+    checkAnswer(sink.allData, 4 to 8)     // new data should replace old data
+  }
+
 
   test("registering as a table in Append output mode") {
     val input = MemoryStream[Int]
@@ -211,7 +265,7 @@ class MemorySinkSuite extends StreamTest with BeforeAndAfter {
 
   test("MemoryPlan statistics") {
     implicit val schema = new StructType().add(new StructField("value", IntegerType))
-    val sink = new MemorySink(schema, OutputMode.Append)
+    val sink = new MemorySink(schema, OutputMode.Append, DataSourceOptions.empty())
     val plan = new MemoryPlan(sink)
 
     // Before adding data, check output