[SPARK-34960][SQL] Aggregate push down for ORC

### What changes were proposed in this pull request?

This PR is to add aggregate push down feature for ORC data source v2 reader.

At a high level, the PR does:

* The supported aggregate expression is MIN/MAX/COUNT same as [Parquet aggregate push down](apache#33639).
* BooleanType, ByteType, ShortType, IntegerType, LongType, FloatType, DoubleType, DateType are allowed in MIN/MAXX aggregate push down. All other columns types are not allowed in MIN/MAX aggregate push down.
* All columns types are supported in COUNT aggregate push down.
* Nested column's sub-fields are disallowed in aggregate push down.
* If the file does not have valid statistics, Spark will throw exception and fail query.
* If aggregate has filter or group-by column, aggregate will not be pushed down.

At code level, the PR does:
* `OrcScanBuilder`: `pushAggregation()` checks whether the aggregation can be pushed down. The most checking logic is shared between Parquet and ORC, extracted into `AggregatePushDownUtils.getSchemaForPushedAggregation()`. `OrcScanBuilder` will create a `OrcScan` with aggregation and aggregation data schema.
* `OrcScan`: `createReaderFactory` creates a ORC reader factory with aggregation and schema. Similar change with `ParquetScan`.
* `OrcPartitionReaderFactory`: `buildReaderWithAggregates` creates a ORC reader with aggregate push down (i.e. read ORC file footer to process columns statistics, instead of reading actual data in the file). `buildColumnarReaderWithAggregates` creates a columnar ORC reader similarly. Both delegate the real work to read footer in `OrcUtils.createAggInternalRowFromFooter`.
* `OrcUtils.createAggInternalRowFromFooter`: reads ORC file footer to process columns statistics (real heavy lift happens here). Similar to `ParquetUtils.createAggInternalRowFromFooter`. Leverage utility method such as `OrcFooterReader.readStatistics`.
* `OrcFooterReader`: `readStatistics` reads the ORC `ColumnStatistics[]` into Spark `OrcColumnStatistics`. The transformation is needed here, because ORC `ColumnStatistics[]` stores all columns statistics in a flatten array style, and hard to process. Spark `OrcColumnStatistics` stores the statistics in nested tree structure (e.g. like `StructType`). This is used by `OrcUtils.createAggInternalRowFromFooter`
* `OrcColumnStatistics`: the easy-to-manipulate structure for ORC `ColumnStatistics`. This is used by `OrcFooterReader.readStatistics`.

### Why are the changes needed?

To improve the performance of query with aggregate.

### Does this PR introduce _any_ user-facing change?

Yes. A user-facing config `spark.sql.orc.aggregatePushdown` is added to control enabling/disabling the aggregate push down for ORC. By default the feature is disabled.

### How was this patch tested?

Added unit test in `FileSourceAggregatePushDownSuite.scala`. Refactored all unit tests in apache#33639, and it now works for both Parquet and ORC.

Closes apache#34298 from c21/orc-agg.

Authored-by: Cheng Su <[email protected]>
Signed-off-by: Liang-Chi Hsieh <[email protected]>

sql/catalyst/src/main/scala/org/apache/spark/sql/internal/SQLConf.scala

@@ -950,6 +950,14 @@ object SQLConf {
     .booleanConf
     .createWithDefault(true)
 
+  val ORC_AGGREGATE_PUSHDOWN_ENABLED = buildConf("spark.sql.orc.aggregatePushdown")
+    .doc("If true, aggregates will be pushed down to ORC for optimization. Support MIN, MAX and " +
+      "COUNT as aggregate expression. For MIN/MAX, support boolean, integer, float and date " +
+      "type. For COUNT, support all data types.")
+    .version("3.3.0")
+    .booleanConf
+    .createWithDefault(false)
+
   val ORC_SCHEMA_MERGING_ENABLED = buildConf("spark.sql.orc.mergeSchema")
     .doc("When true, the Orc data source merges schemas collected from all data files, " +
       "otherwise the schema is picked from a random data file.")
@@ -3691,6 +3699,8 @@ class SQLConf extends Serializable with Logging {
 
   def orcFilterPushDown: Boolean = getConf(ORC_FILTER_PUSHDOWN_ENABLED)
 
+  def orcAggregatePushDown: Boolean = getConf(ORC_AGGREGATE_PUSHDOWN_ENABLED)
+
   def isOrcSchemaMergingEnabled: Boolean = getConf(ORC_SCHEMA_MERGING_ENABLED)
 
   def verifyPartitionPath: Boolean = getConf(HIVE_VERIFY_PARTITION_PATH)

sql/catalyst/src/main/scala/org/apache/spark/sql/types/StructType.scala

@@ -115,7 +115,7 @@ case class StructType(fields: Array[StructField]) extends DataType with Seq[Stru
   def names: Array[String] = fieldNames
 
   private lazy val fieldNamesSet: Set[String] = fieldNames.toSet
-  private[sql] lazy val nameToField: Map[String, StructField] = fields.map(f => f.name -> f).toMap
+  private lazy val nameToField: Map[String, StructField] = fields.map(f => f.name -> f).toMap
   private lazy val nameToIndex: Map[String, Int] = fieldNames.zipWithIndex.toMap
 
   override def equals(that: Any): Boolean = {

sql/core/src/main/java/org/apache/spark/sql/execution/datasources/orc/OrcColumnStatistics.java

@@ -0,0 +1,80 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.sql.execution.datasources.orc;
+
+import org.apache.orc.ColumnStatistics;
+
+import java.util.ArrayList;
+import java.util.List;
+
+/**
+ * Columns statistics interface wrapping ORC {@link ColumnStatistics}s.
+ *
+ * Because ORC {@link ColumnStatistics}s are stored as an flatten array in ORC file footer,
+ * this class is used to covert ORC {@link ColumnStatistics}s from array to nested tree structure,
+ * according to data types. The flatten array stores all data types (including nested types) in
+ * tree pre-ordering. This is used for aggregate push down in ORC.
+ *
+ * For nested data types (array, map and struct), the sub-field statistics are stored recursively
+ * inside parent column's children field. Here is an example of {@link OrcColumnStatistics}:
+ *
+ * Data schema:
+ * c1: int
+ * c2: struct<f1: int, f2: float>
+ * c3: map<key: int, value: string>
+ * c4: array<int>
+ *
+ *                        OrcColumnStatistics
+ *                                | (children)
+ *             ---------------------------------------------
+ *            /         |                 \                 \
+ *           c1        c2                 c3                c4
+ *      (integer)    (struct)            (map)             (array)
+*        (min:1,        | (children)       | (children)      | (children)
+ *       max:10)      -----              -----             element
+ *                   /     \            /     \           (integer)
+ *                c2.f1    c2.f2      key     value
+ *              (integer) (float)  (integer) (string)
+ *                       (min:0.1,           (min:"a",
+ *                       max:100.5)          max:"zzz")
+ */
+public class OrcColumnStatistics {
+  private final ColumnStatistics statistics;
+  private final List<OrcColumnStatistics> children;
+
+  public OrcColumnStatistics(ColumnStatistics statistics) {
+    this.statistics = statistics;
+    this.children = new ArrayList<>();
+  }
+
+  public ColumnStatistics getStatistics() {
+    return statistics;
+  }
+
+  public OrcColumnStatistics get(int ordinal) {
+    if (ordinal < 0 || ordinal >= children.size()) {
+      throw new IndexOutOfBoundsException(
+        String.format("Ordinal %d out of bounds of statistics size %d", ordinal, children.size()));
+    }
+    return children.get(ordinal);
+  }
+
+  public void add(OrcColumnStatistics newChild) {
+    children.add(newChild);
+  }
+}

sql/core/src/main/java/org/apache/spark/sql/execution/datasources/orc/OrcFooterReader.java

@@ -0,0 +1,67 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.sql.execution.datasources.orc;
+
+import org.apache.orc.ColumnStatistics;
+import org.apache.orc.Reader;
+import org.apache.orc.TypeDescription;
+import org.apache.spark.sql.types.*;
+
+import java.util.Arrays;
+import java.util.LinkedList;
+import java.util.Queue;
+
+/**
+ * {@link OrcFooterReader} is a util class which encapsulates the helper
+ * methods of reading ORC file footer.
+ */
+public class OrcFooterReader {
+
+  /**
+   * Read the columns statistics from ORC file footer.
+   *
+   * @param orcReader the reader to read ORC file footer.
+   * @return Statistics for all columns in the file.
+   */
+  public static OrcColumnStatistics readStatistics(Reader orcReader) {
+    TypeDescription orcSchema = orcReader.getSchema();
+    ColumnStatistics[] orcStatistics = orcReader.getStatistics();
+    StructType sparkSchema = OrcUtils.toCatalystSchema(orcSchema);
+    return convertStatistics(sparkSchema, new LinkedList<>(Arrays.asList(orcStatistics)));
+  }
+
+  /**
+   * Convert a queue of ORC {@link ColumnStatistics}s into Spark {@link OrcColumnStatistics}.
+   * The queue of ORC {@link ColumnStatistics}s are assumed to be ordered as tree pre-order.
+   */
+  private static OrcColumnStatistics convertStatistics(
+      DataType sparkSchema, Queue<ColumnStatistics> orcStatistics) {
+    OrcColumnStatistics statistics = new OrcColumnStatistics(orcStatistics.remove());
+    if (sparkSchema instanceof StructType) {
+      for (StructField field : ((StructType) sparkSchema).fields()) {
+        statistics.add(convertStatistics(field.dataType(), orcStatistics));
+      }
+    } else if (sparkSchema instanceof MapType) {
+      statistics.add(convertStatistics(((MapType) sparkSchema).keyType(), orcStatistics));
+      statistics.add(convertStatistics(((MapType) sparkSchema).valueType(), orcStatistics));
+    } else if (sparkSchema instanceof ArrayType) {
+      statistics.add(convertStatistics(((ArrayType) sparkSchema).elementType(), orcStatistics));
+    }
+    return statistics;
+  }
+}

sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/AggregatePushDownUtils.scala

@@ -0,0 +1,141 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.sql.execution.datasources
+
+import org.apache.spark.sql.catalyst.InternalRow
+import org.apache.spark.sql.catalyst.expressions.Expression
+import org.apache.spark.sql.connector.expressions.NamedReference
+import org.apache.spark.sql.connector.expressions.aggregate.{AggregateFunc, Aggregation, Count, CountStar, Max, Min}
+import org.apache.spark.sql.execution.RowToColumnConverter
+import org.apache.spark.sql.execution.vectorized.{OffHeapColumnVector, OnHeapColumnVector}
+import org.apache.spark.sql.types.{BooleanType, ByteType, DateType, DoubleType, FloatType, IntegerType, LongType, ShortType, StructField, StructType}
+import org.apache.spark.sql.vectorized.{ColumnarBatch, ColumnVector}
+
+/**
+ * Utility class for aggregate push down to Parquet and ORC.
+ */
+object AggregatePushDownUtils {
+
+  /**
+   * Get the data schema for aggregate to be pushed down.
+   */
+  def getSchemaForPushedAggregation(
+      aggregation: Aggregation,
+      schema: StructType,
+      partitionNames: Set[String],
+      dataFilters: Seq[Expression]): Option[StructType] = {
+
+    var finalSchema = new StructType()
+
+    def getStructFieldForCol(col: NamedReference): StructField = {
+      schema.apply(col.fieldNames.head)
+    }
+
+    def isPartitionCol(col: NamedReference) = {
+      partitionNames.contains(col.fieldNames.head)
+    }
+
+    def processMinOrMax(agg: AggregateFunc): Boolean = {
+      val (column, aggType) = agg match {
+        case max: Max => (max.column, "max")
+        case min: Min => (min.column, "min")
+        case _ =>
+          throw new IllegalArgumentException(s"Unexpected type of AggregateFunc ${agg.describe}")
+      }
+
+      if (isPartitionCol(column)) {
+        // don't push down partition column, footer doesn't have max/min for partition column
+        return false
+      }
+      val structField = getStructFieldForCol(column)
+
+      structField.dataType match {
+        // not push down complex type
+        // not push down Timestamp because INT96 sort order is undefined,
+        // Parquet doesn't return statistics for INT96
+        // not push down Parquet Binary because min/max could be truncated
+        // (https://issues.apache.org/jira/browse/PARQUET-1685), Parquet Binary
+        // could be Spark StringType, BinaryType or DecimalType.
+        // not push down for ORC with same reason.
+        case BooleanType | ByteType | ShortType | IntegerType
+             | LongType | FloatType | DoubleType | DateType =>
+          finalSchema = finalSchema.add(structField.copy(s"$aggType(" + structField.name + ")"))
+          true
+        case _ =>
+          false
+      }
+    }
+
+    if (aggregation.groupByColumns.nonEmpty || dataFilters.nonEmpty) {
+      // Parquet/ORC footer has max/min/count for columns
+      // e.g. SELECT COUNT(col1) FROM t
+      // but footer doesn't have max/min/count for a column if max/min/count
+      // are combined with filter or group by
+      // e.g. SELECT COUNT(col1) FROM t WHERE col2 = 8
+      //      SELECT COUNT(col1) FROM t GROUP BY col2
+      // However, if the filter is on partition column, max/min/count can still be pushed down
+      // Todo:  add support if groupby column is partition col
+      //        (https://issues.apache.org/jira/browse/SPARK-36646)
+      return None
+    }
+
+    aggregation.aggregateExpressions.foreach {
+      case max: Max =>
+        if (!processMinOrMax(max)) return None
+      case min: Min =>
+        if (!processMinOrMax(min)) return None
+      case count: Count =>
+        if (count.column.fieldNames.length != 1 || count.isDistinct) return None
+        finalSchema =
+          finalSchema.add(StructField(s"count(" + count.column.fieldNames.head + ")", LongType))
+      case _: CountStar =>
+        finalSchema = finalSchema.add(StructField("count(*)", LongType))
+      case _ =>
+        return None
+    }
+
+    Some(finalSchema)
+  }
+
+  /**
+   * Check if two Aggregation `a` and `b` is equal or not.
+   */
+  def equivalentAggregations(a: Aggregation, b: Aggregation): Boolean = {
+    a.aggregateExpressions.sortBy(_.hashCode())
+      .sameElements(b.aggregateExpressions.sortBy(_.hashCode())) &&
+      a.groupByColumns.sortBy(_.hashCode()).sameElements(b.groupByColumns.sortBy(_.hashCode()))
+  }
+
+  /**
+   * Convert the aggregates result from `InternalRow` to `ColumnarBatch`.
+   * This is used for columnar reader.
+   */
+  def convertAggregatesRowToBatch(
+      aggregatesAsRow: InternalRow,
+      aggregatesSchema: StructType,
+      offHeap: Boolean): ColumnarBatch = {
+    val converter = new RowToColumnConverter(aggregatesSchema)
+    val columnVectors = if (offHeap) {
+      OffHeapColumnVector.allocateColumns(1, aggregatesSchema)
+    } else {
+      OnHeapColumnVector.allocateColumns(1, aggregatesSchema)
+    }
+    converter.convert(aggregatesAsRow, columnVectors.toArray)
+    new ColumnarBatch(columnVectors.asInstanceOf[Array[ColumnVector]], 1)
+  }
+}

sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/orc/OrcDeserializer.scala

@@ -68,6 +68,22 @@ class OrcDeserializer(
     resultRow
   }
 
+  def deserializeFromValues(orcValues: Seq[WritableComparable[_]]): InternalRow = {
+    var targetColumnIndex = 0
+    while (targetColumnIndex < fieldWriters.length) {
+      if (fieldWriters(targetColumnIndex) != null) {
+        val value = orcValues(requestedColIds(targetColumnIndex))
+        if (value == null) {
+          resultRow.setNullAt(targetColumnIndex)
+        } else {
+          fieldWriters(targetColumnIndex)(value)
+        }
+      }
+      targetColumnIndex += 1
+    }
+    resultRow
+  }
+
   /**
    * Creates a writer to write ORC values to Catalyst data structure at the given ordinal.
    */