[SPARK-48901][SPARK-48916][SS][PYTHON] Introduce clusterBy DataStream…

…Writer API

### What changes were proposed in this pull request?
Introduce a new clusterBy DataStreamWriter API in Scala, Spark Connect, and Pyspark.

### Why are the changes needed?
Provides another way for users to create clustered tables for streaming writes.

### Does this PR introduce _any_ user-facing change?
Yes, it adds a new clusterBy DataStreamWriter API in Scala, Spark Connect, and Pyspark to allow specifying the clustering columns when writing streaming DataFrames.

### How was this patch tested?
See new unit tests.

### Was this patch authored or co-authored using generative AI tooling?
No.

Closes apache#47376 from chirag-s-db/cluster-by-stream.

Lead-authored-by: Chirag Singh <[email protected]>
Co-authored-by: Chirag Singh <[email protected]>
Signed-off-by: Jungtaek Lim <[email protected]>

connect/common/src/main/protobuf/spark/connect/commands.proto

@@ -240,6 +240,9 @@ message WriteStreamOperationStart {
 
   StreamingForeachFunction foreach_writer = 13;
   StreamingForeachFunction foreach_batch = 14;
+
+  // (Optional) Columns used for clustering the table.
+  repeated string clustering_column_names = 15;
 }
 
 message StreamingForeachFunction {

connect/server/src/main/scala/org/apache/spark/sql/connect/planner/SparkConnectPlanner.scala

@@ -3139,6 +3139,10 @@ class SparkConnectPlanner(
       writer.partitionBy(writeOp.getPartitioningColumnNamesList.asScala.toList: _*)
     }
 
+    if (writeOp.getClusteringColumnNamesCount > 0) {
+      writer.clusterBy(writeOp.getClusteringColumnNamesList.asScala.toList: _*)
+    }
+
     writeOp.getTriggerCase match {
       case TriggerCase.PROCESSING_TIME_INTERVAL =>
         writer.trigger(Trigger.ProcessingTime(writeOp.getProcessingTimeInterval))

connector/connect/client/jvm/src/main/scala/org/apache/spark/sql/streaming/DataStreamWriter.scala

@@ -159,6 +159,23 @@ final class DataStreamWriter[T] private[sql] (ds: Dataset[T]) extends Logging {
     this
   }
 
+  /**
+   * Clusters the output by the given columns. If specified, the output is laid out such that
+   * records with similar values on the clustering column are grouped together in the same file.
+   *
+   * Clustering improves query efficiency by allowing queries with predicates on the clustering
+   * columns to skip unnecessary data. Unlike partitioning, clustering can be used on very high
+   * cardinality columns.
+   *
+   * @since 4.0.0
+   */
+  @scala.annotation.varargs
+  def clusterBy(colNames: String*): DataStreamWriter[T] = {
+    sinkBuilder.clearClusteringColumnNames()
+    sinkBuilder.addAllClusteringColumnNames(colNames.asJava)
+    this
+  }
+
   /**
    * Adds an output option for the underlying data source.
    *

connector/connect/client/jvm/src/test/scala/org/apache/spark/sql/streaming/ClientStreamingQuerySuite.scala

@@ -268,6 +268,42 @@ class ClientStreamingQuerySuite extends QueryTest with RemoteSparkSession with L
     }
   }
 
+  test("clusterBy") {
+    withSQLConf(
+      "spark.sql.shuffle.partitions" -> "1" // Avoid too many reducers.
+    ) {
+      spark.sql("DROP TABLE IF EXISTS my_table").collect()
+
+      withTempPath { ckpt =>
+        val q1 = spark.readStream
+          .format("rate")
+          .load()
+          .writeStream
+          .clusterBy("value")
+          .option("checkpointLocation", ckpt.getCanonicalPath)
+          .toTable("my_table")
+
+        try {
+          q1.processAllAvailable()
+          eventually(timeout(30.seconds)) {
+            checkAnswer(
+              spark.sql("DESCRIBE my_table"),
+              Seq(
+                Row("timestamp", "timestamp", null),
+                Row("value", "bigint", null),
+                Row("# Clustering Information", "", ""),
+                Row("# col_name", "data_type", "comment"),
+                Row("value", "bigint", null)))
+            assert(spark.table("my_sink").count() > 0)
+          }
+        } finally {
+          q1.stop()
+          spark.sql("DROP TABLE my_table")
+        }
+      }
+    }
+  }
+
   test("throw exception in streaming") {
     try {
       val session = spark

project/MimaExcludes.scala

@@ -102,7 +102,9 @@ object MimaExcludes {
     ProblemFilters.exclude[ReversedMissingMethodProblem]("org.apache.spark.sql.SQLImplicits.session"),
     ProblemFilters.exclude[DirectMissingMethodProblem]("org.apache.spark.sql.SparkSession#implicits._sqlContext"),
     // SPARK-48761: Add clusterBy() to CreateTableWriter.
-    ProblemFilters.exclude[ReversedMissingMethodProblem]("org.apache.spark.sql.CreateTableWriter.clusterBy")
+    ProblemFilters.exclude[ReversedMissingMethodProblem]("org.apache.spark.sql.CreateTableWriter.clusterBy"),
+    // SPARK-48901: Add clusterBy() to DataStreamWriter.
+    ProblemFilters.exclude[ReversedMissingMethodProblem]("org.apache.spark.sql.DataStreamWriter.clusterBy")
   )
 
   // Default exclude rules

python/pyspark/sql/connect/proto/commands_pb2.pyi

@@ -905,6 +905,7 @@ class WriteStreamOperationStart(google.protobuf.message.Message):
     TABLE_NAME_FIELD_NUMBER: builtins.int
     FOREACH_WRITER_FIELD_NUMBER: builtins.int
     FOREACH_BATCH_FIELD_NUMBER: builtins.int
+    CLUSTERING_COLUMN_NAMES_FIELD_NUMBER: builtins.int
     @property
     def input(self) -> pyspark.sql.connect.proto.relations_pb2.Relation:
         """(Required) The output of the `input` streaming relation will be written."""
@@ -932,6 +933,11 @@ class WriteStreamOperationStart(google.protobuf.message.Message):
     def foreach_writer(self) -> global___StreamingForeachFunction: ...
     @property
     def foreach_batch(self) -> global___StreamingForeachFunction: ...
+    @property
+    def clustering_column_names(
+        self,
+    ) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.str]:
+        """(Optional) Columns used for clustering the table."""
     def __init__(
         self,
         *,
@@ -949,6 +955,7 @@ class WriteStreamOperationStart(google.protobuf.message.Message):
         table_name: builtins.str = ...,
         foreach_writer: global___StreamingForeachFunction | None = ...,
         foreach_batch: global___StreamingForeachFunction | None = ...,
+        clustering_column_names: collections.abc.Iterable[builtins.str] | None = ...,
     ) -> None: ...
     def HasField(
         self,
@@ -982,6 +989,8 @@ class WriteStreamOperationStart(google.protobuf.message.Message):
         field_name: typing_extensions.Literal[
             "available_now",
             b"available_now",
+            "clustering_column_names",
+            b"clustering_column_names",
             "continuous_checkpoint_interval",
             b"continuous_checkpoint_interval",
             "foreach_batch",

python/pyspark/sql/connect/streaming/readwriter.py

@@ -445,6 +445,25 @@ def partitionBy(self, *cols: str) -> "DataStreamWriter":  # type: ignore[misc]
 
     partitionBy.__doc__ = PySparkDataStreamWriter.partitionBy.__doc__
 
+    @overload
+    def clusterBy(self, *cols: str) -> "DataStreamWriter":
+        ...
+
+    @overload
+    def clusterBy(self, __cols: List[str]) -> "DataStreamWriter":
+        ...
+
+    def clusterBy(self, *cols: str) -> "DataStreamWriter":  # type: ignore[misc]
+        if len(cols) == 1 and isinstance(cols[0], (list, tuple)):
+            cols = cols[0]
+        # Clear any existing columns (if any).
+        while len(self._write_proto.clustering_column_names) > 0:
+            self._write_proto.clustering_column_names.pop()
+        self._write_proto.clustering_column_names.extend(cast(List[str], cols))
+        return self
+
+    clusterBy.__doc__ = PySparkDataStreamWriter.clusterBy.__doc__
+
     def queryName(self, queryName: str) -> "DataStreamWriter":
         if not queryName or type(queryName) != str or len(queryName.strip()) == 0:
             raise PySparkValueError(

python/pyspark/sql/streaming/readwriter.py

@@ -1123,6 +1123,65 @@ def partitionBy(self, *cols: str) -> "DataStreamWriter":  # type: ignore[misc]
         self._jwrite = self._jwrite.partitionBy(_to_seq(self._spark._sc, cols))
         return self
 
+    @overload
+    def clusterBy(self, *cols: str) -> "DataStreamWriter":
+        ...
+
+    @overload
+    def clusterBy(self, __cols: List[str]) -> "DataStreamWriter":
+        ...
+
+    def clusterBy(self, *cols: str) -> "DataStreamWriter":  # type: ignore[misc]
+        """Clusters the output by the given columns.
+
+        If specified, the output is laid out such that records with similar values on the clustering
+        column(s) are grouped together in the same file.
+
+        Clustering improves query efficiency by allowing queries with predicates on the clustering
+        columns to skip unnecessary data. Unlike partitioning, clustering can be used on very high
+        cardinality columns.
+
+        .. versionadded:: 4.0.0
+
+        Parameters
+        ----------
+        cols : str or list
+            name of columns
+
+        Notes
+        -----
+        This API is evolving.
+
+        Examples
+        --------
+        >>> df = spark.readStream.format("rate").load()
+        >>> df.writeStream.clusterBy("value")
+        <...streaming.readwriter.DataStreamWriter object ...>
+
+        Cluster-by timestamp column from Rate source.
+
+        >>> import tempfile
+        >>> import time
+        >>> with tempfile.TemporaryDirectory(prefix="partitionBy1") as d:
+        ...     with tempfile.TemporaryDirectory(prefix="partitionBy2") as cp:
+        ...         df = spark.readStream.format("rate").option("rowsPerSecond", 10).load()
+        ...         q = df.writeStream.clusterBy(
+        ...             "timestamp").format("parquet").option("checkpointLocation", cp).start(d)
+        ...         time.sleep(5)
+        ...         q.stop()
+        ...         spark.read.schema(df.schema).parquet(d).show()
+        +...---------+-----+
+        |...timestamp|value|
+        +...---------+-----+
+        ...
+        """
+        from pyspark.sql.classic.column import _to_seq
+
+        if len(cols) == 1 and isinstance(cols[0], (list, tuple)):
+            cols = cols[0]
+        self._jwrite = self._jwrite.clusterBy(_to_seq(self._spark._sc, cols))
+        return self
+
     def queryName(self, queryName: str) -> "DataStreamWriter":
         """Specifies the name of the :class:`StreamingQuery` that can be started with
         :func:`start`. This name must be unique among all the currently active queries

python/pyspark/sql/tests/streaming/test_streaming.py

@@ -429,6 +429,31 @@ def test_streaming_write_to_table(self):
             result = self.spark.sql("SELECT value FROM output_table").collect()
             self.assertTrue(len(result) > 0)
 
+    def test_streaming_write_to_table_cluster_by(self):
+        with self.table("output_table"), tempfile.TemporaryDirectory(prefix="to_table") as tmpdir:
+            df = self.spark.readStream.format("rate").option("rowsPerSecond", 10).load()
+            q = df.writeStream.clusterBy("value").toTable(
+                "output_table", format="parquet", checkpointLocation=tmpdir
+            )
+            self.assertTrue(q.isActive)
+            time.sleep(10)
+            q.stop()
+            result = self.spark.sql("DESCRIBE output_table").collect()
+            self.assertEqual(
+                set(
+                    [
+                        Row(col_name="timestamp", data_type="timestamp", comment=None),
+                        Row(col_name="value", data_type="bigint", comment=None),
+                        Row(col_name="# Clustering Information", data_type="", comment=""),
+                        Row(col_name="# col_name", data_type="data_type", comment="comment"),
+                        Row(col_name="value", data_type="bigint", comment=None),
+                    ]
+                ),
+                set(result),
+            )
+            result = self.spark.sql("SELECT value FROM output_table").collect()
+            self.assertTrue(len(result) > 0)
+
     def test_streaming_with_temporary_view(self):
         """
         This verifies createOrReplaceTempView() works with a streaming dataframe. An SQL

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

@@ -36,9 +36,10 @@ import org.apache.spark.sql.catalyst.types.DataTypeUtils
 import org.apache.spark.sql.catalyst.util.CaseInsensitiveMap
 import org.apache.spark.sql.connector.catalog.{Identifier, SupportsWrite, Table, TableCatalog, TableProvider, V1Table, V2TableWithV1Fallback}
 import org.apache.spark.sql.connector.catalog.TableCapability._
+import org.apache.spark.sql.connector.expressions.{ClusterByTransform, FieldReference}
 import org.apache.spark.sql.errors.QueryCompilationErrors
 import org.apache.spark.sql.execution.command.DDLUtils
-import org.apache.spark.sql.execution.datasources.DataSource
+import org.apache.spark.sql.execution.datasources.{DataSource, DataSourceUtils}
 import org.apache.spark.sql.execution.datasources.v2.{DataSourceV2Utils, FileDataSourceV2}
 import org.apache.spark.sql.execution.datasources.v2.python.PythonDataSourceV2
 import org.apache.spark.sql.execution.streaming._
@@ -166,6 +167,24 @@ final class DataStreamWriter[T] private[sql](ds: Dataset[T]) {
   @scala.annotation.varargs
   def partitionBy(colNames: String*): DataStreamWriter[T] = {
     this.partitioningColumns = Option(colNames)
+    validatePartitioningAndClustering()
+    this
+  }
+
+  /**
+   * Clusters the output by the given columns. If specified, the output is laid out such that
+   * records with similar values on the clustering column are grouped together in the same file.
+   *
+   * Clustering improves query efficiency by allowing queries with predicates on the clustering
+   * columns to skip unnecessary data. Unlike partitioning, clustering can be used on very high
+   * cardinality columns.
+   *
+   * @since 4.0.0
+   */
+  @scala.annotation.varargs
+  def clusterBy(colNames: String*): DataStreamWriter[T] = {
+    this.clusteringColumns = Option(colNames)
+    validatePartitioningAndClustering()
     this
   }
 
@@ -288,12 +307,21 @@ final class DataStreamWriter[T] private[sql](ds: Dataset[T]) {
 
     if (!catalog.asTableCatalog.tableExists(identifier)) {
       import org.apache.spark.sql.connector.catalog.CatalogV2Implicits._
+
+      val properties = normalizedClusteringCols.map { cols =>
+        Map(
+          DataSourceUtils.CLUSTERING_COLUMNS_KEY -> DataSourceUtils.encodePartitioningColumns(cols))
+      }.getOrElse(Map.empty)
+      val partitioningOrClusteringTransform = normalizedClusteringCols.map { colNames =>
+        Array(ClusterByTransform(colNames.map(col => FieldReference(col)))).toImmutableArraySeq
+      }.getOrElse(partitioningColumns.getOrElse(Nil).asTransforms.toImmutableArraySeq)
+
       /**
        * Note, currently the new table creation by this API doesn't fully cover the V2 table.
        * TODO (SPARK-33638): Full support of v2 table creation
        */
       val tableSpec = UnresolvedTableSpec(
-        Map.empty[String, String],
+        properties,
         Some(source),
         OptionList(Seq.empty),
         extraOptions.get("path"),
@@ -303,7 +331,7 @@ final class DataStreamWriter[T] private[sql](ds: Dataset[T]) {
       val cmd = CreateTable(
         UnresolvedIdentifier(originalMultipartIdentifier),
         df.schema.asNullable.map(ColumnDefinition.fromV1Column(_, parser)),
-        partitioningColumns.getOrElse(Nil).asTransforms.toImmutableArraySeq,
+        partitioningOrClusteringTransform,
         tableSpec,
         ignoreIfExists = false)
       Dataset.ofRows(df.sparkSession, cmd)
@@ -439,10 +467,22 @@ final class DataStreamWriter[T] private[sql](ds: Dataset[T]) {
   }
 
   private def createV1Sink(optionsWithPath: CaseInsensitiveMap[String]): Sink = {
+    // Do not allow the user to specify clustering columns in the options. Ignoring this option is
+    // consistent with the behavior of DataFrameWriter on non Path-based tables and with the
+    // behavior of DataStreamWriter on partitioning columns specified in options.
+    val optionsWithoutClusteringKey =
+      optionsWithPath.originalMap - DataSourceUtils.CLUSTERING_COLUMNS_KEY
+
+    val optionsWithClusteringColumns = normalizedClusteringCols match {
+      case Some(cols) => optionsWithoutClusteringKey + (
+        DataSourceUtils.CLUSTERING_COLUMNS_KEY ->
+          DataSourceUtils.encodePartitioningColumns(cols))
+      case None => optionsWithoutClusteringKey
+    }
     val ds = DataSource(
       df.sparkSession,
       className = source,
-      options = optionsWithPath.originalMap,
+      options = optionsWithClusteringColumns,
       partitionColumns = normalizedParCols.getOrElse(Nil))
     ds.createSink(outputMode)
   }
@@ -514,6 +554,10 @@ final class DataStreamWriter[T] private[sql](ds: Dataset[T]) {
     cols.map(normalize(_, "Partition"))
   }
 
+  private def normalizedClusteringCols: Option[Seq[String]] = clusteringColumns.map { cols =>
+    cols.map(normalize(_, "Clustering"))
+  }
+
   /**
    * The given column name may not be equal to any of the existing column names if we were in
    * case-insensitive context. Normalize the given column name to the real one so that we don't
@@ -532,6 +576,13 @@ final class DataStreamWriter[T] private[sql](ds: Dataset[T]) {
     }
   }
 
+  // Validate that partitionBy isn't used with clusterBy.
+  private def validatePartitioningAndClustering(): Unit = {
+    if (clusteringColumns.nonEmpty && partitioningColumns.nonEmpty) {
+      throw QueryCompilationErrors.clusterByWithPartitionedBy()
+    }
+  }
+
   ///////////////////////////////////////////////////////////////////////////////////////
   // Builder pattern config options
   ///////////////////////////////////////////////////////////////////////////////////////
@@ -554,6 +605,8 @@ final class DataStreamWriter[T] private[sql](ds: Dataset[T]) {
   private var foreachBatchWriter: (Dataset[T], Long) => Unit = null
 
   private var partitioningColumns: Option[Seq[String]] = None
+
+  private var clusteringColumns: Option[Seq[String]] = None
 }
 
 object DataStreamWriter {