[SPARK-17073] [SQL] generate column-level statistics

sql/catalyst/src/main/antlr4/org/apache/spark/sql/catalyst/parser/SqlBase.g4

@@ -86,7 +86,7 @@ statement
     | CREATE TABLE (IF NOT EXISTS)? target=tableIdentifier
         LIKE source=tableIdentifier                                    #createTableLike
     | ANALYZE TABLE tableIdentifier partitionSpec? COMPUTE STATISTICS
-        (identifier | FOR COLUMNS identifierSeq?)?                     #analyze
+        (identifier | FOR COLUMNS identifierSeq)?                      #analyze
     | ALTER (TABLE | VIEW) from=tableIdentifier
         RENAME TO to=tableIdentifier                                   #renameTable
     | ALTER (TABLE | VIEW) tableIdentifier

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/Statistics.scala

@@ -17,6 +17,12 @@
 
 package org.apache.spark.sql.catalyst.plans.logical
 
+import org.apache.commons.codec.binary.Base64
+
+import org.apache.spark.sql.catalyst.InternalRow
+import org.apache.spark.sql.catalyst.expressions.UnsafeRow
+import org.apache.spark.sql.types._
+
 /**
  * Estimates of various statistics.  The default estimation logic simply lazily multiplies the
  * corresponding statistic produced by the children.  To override this behavior, override
@@ -32,19 +38,80 @@ package org.apache.spark.sql.catalyst.plans.logical
  * @param sizeInBytes Physical size in bytes. For leaf operators this defaults to 1, otherwise it
  *                    defaults to the product of children's `sizeInBytes`.
  * @param rowCount Estimated number of rows.
+ * @param colStats Column-level statistics.
  * @param isBroadcastable If true, output is small enough to be used in a broadcast join.
  */
 case class Statistics(
     sizeInBytes: BigInt,
     rowCount: Option[BigInt] = None,
+    colStats: Map[String, ColumnStat] = Map.empty,
     isBroadcastable: Boolean = false) {
+
   override def toString: String = "Statistics(" + simpleString + ")"
 
   /** Readable string representation for the Statistics. */
   def simpleString: String = {
     Seq(s"sizeInBytes=$sizeInBytes",
       if (rowCount.isDefined) s"rowCount=${rowCount.get}" else "",
       s"isBroadcastable=$isBroadcastable"
-    ).filter(_.nonEmpty).mkString("", ", ", "")
+    ).filter(_.nonEmpty).mkString(", ")
+  }
+}
+
+/**
+ * Statistics for a column.
+ */
+case class ColumnStat(statRow: InternalRow) {
+
+  def forNumeric[T <: AtomicType](dataType: T): NumericColumnStat[T] = {
+    NumericColumnStat(statRow, dataType)
+  }
+  def forString: StringColumnStat = StringColumnStat(statRow)
+  def forBinary: BinaryColumnStat = BinaryColumnStat(statRow)
+  def forBoolean: BooleanColumnStat = BooleanColumnStat(statRow)
+
+  override def toString: String = {
+    // use Base64 for encoding
+    Base64.encodeBase64String(statRow.asInstanceOf[UnsafeRow].getBytes)
   }
 }
+
+object ColumnStat {
+  def apply(numFields: Int, str: String): ColumnStat = {
+    // use Base64 for decoding
+    val bytes = Base64.decodeBase64(str)
+    val unsafeRow = new UnsafeRow(numFields)
+    unsafeRow.pointTo(bytes, bytes.length)
+    ColumnStat(unsafeRow)
+  }
+}
+
+case class NumericColumnStat[T <: AtomicType](statRow: InternalRow, dataType: T) {
+  // The indices here must be consistent with `ColumnStatStruct.numericColumnStat`.
+  val numNulls: Long = statRow.getLong(0)
+  val max: T#InternalType = statRow.get(1, dataType).asInstanceOf[T#InternalType]
+  val min: T#InternalType = statRow.get(2, dataType).asInstanceOf[T#InternalType]
+  val ndv: Long = statRow.getLong(3)
+}
+
+case class StringColumnStat(statRow: InternalRow) {
+  // The indices here must be consistent with `ColumnStatStruct.stringColumnStat`.
+  val numNulls: Long = statRow.getLong(0)
+  val avgColLen: Double = statRow.getDouble(1)
+  val maxColLen: Long = statRow.getLong(2)
+  val ndv: Long = statRow.getLong(3)
+}
+
+case class BinaryColumnStat(statRow: InternalRow) {
+  // The indices here must be consistent with `ColumnStatStruct.binaryColumnStat`.
+  val numNulls: Long = statRow.getLong(0)
+  val avgColLen: Double = statRow.getDouble(1)
+  val maxColLen: Long = statRow.getLong(2)
+}
+
+case class BooleanColumnStat(statRow: InternalRow) {
+  // The indices here must be consistent with `ColumnStatStruct.booleanColumnStat`.
+  val numNulls: Long = statRow.getLong(0)
+  val numTrues: Long = statRow.getLong(1)
+  val numFalses: Long = statRow.getLong(2)
+}

sql/core/src/main/scala/org/apache/spark/sql/execution/SparkSqlParser.scala

@@ -87,19 +87,27 @@ class SparkSqlAstBuilder(conf: SQLConf) extends AstBuilder {
   }
 
   /**
-   * Create an [[AnalyzeTableCommand]] command. This currently only implements the NOSCAN
-   * option (other options are passed on to Hive) e.g.:
+   * Create an [[AnalyzeTableCommand]] command or an [[AnalyzeColumnCommand]] command.
+   * Example SQL for analyzing table :
    * {{{
-   *   ANALYZE TABLE table COMPUTE STATISTICS NOSCAN;
+   *   ANALYZE TABLE table COMPUTE STATISTICS [NOSCAN];
+   * }}}
+   * Example SQL for analyzing columns :
+   * {{{
+   *   ANALYZE TABLE table COMPUTE STATISTICS FOR COLUMNS column1, column2;
    * }}}
    */
   override def visitAnalyze(ctx: AnalyzeContext): LogicalPlan = withOrigin(ctx) {
     if (ctx.partitionSpec == null &&
       ctx.identifier != null &&
       ctx.identifier.getText.toLowerCase == "noscan") {
-      AnalyzeTableCommand(visitTableIdentifier(ctx.tableIdentifier).toString)
+      AnalyzeTableCommand(visitTableIdentifier(ctx.tableIdentifier))
+    } else if (ctx.identifierSeq() == null) {
+      AnalyzeTableCommand(visitTableIdentifier(ctx.tableIdentifier), noscan = false)
     } else {
-      AnalyzeTableCommand(visitTableIdentifier(ctx.tableIdentifier).toString, noscan = false)
+      AnalyzeColumnCommand(
+        visitTableIdentifier(ctx.tableIdentifier),
+        visitIdentifierSeq(ctx.identifierSeq()))
     }
   }
 

sql/core/src/main/scala/org/apache/spark/sql/execution/command/AnalyzeColumnCommand.scala

@@ -0,0 +1,175 @@
+/*
+ * 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.command
+
+import scala.collection.mutable
+
+import org.apache.spark.sql._
+import org.apache.spark.sql.catalyst.TableIdentifier
+import org.apache.spark.sql.catalyst.analysis.EliminateSubqueryAliases
+import org.apache.spark.sql.catalyst.catalog.{CatalogRelation, CatalogTable}
+import org.apache.spark.sql.catalyst.expressions._
+import org.apache.spark.sql.catalyst.expressions.aggregate._
+import org.apache.spark.sql.catalyst.plans.logical.{Aggregate, ColumnStat, LogicalPlan, Statistics}
+import org.apache.spark.sql.execution.datasources.LogicalRelation
+import org.apache.spark.sql.types._
+
+
+/**
+ * Analyzes the given columns of the given table to generate statistics, which will be used in
+ * query optimizations.
+ */
+case class AnalyzeColumnCommand(
+    tableIdent: TableIdentifier,
+    columnNames: Seq[String]) extends RunnableCommand {
+
+  override def run(sparkSession: SparkSession): Seq[Row] = {
+    val sessionState = sparkSession.sessionState
+    val db = tableIdent.database.getOrElse(sessionState.catalog.getCurrentDatabase)
+    val tableIdentWithDB = TableIdentifier(tableIdent.table, Some(db))
+    val relation = EliminateSubqueryAliases(sessionState.catalog.lookupRelation(tableIdentWithDB))
+
+    relation match {
+      case catalogRel: CatalogRelation =>
+        updateStats(catalogRel.catalogTable,
+          AnalyzeTableCommand.calculateTotalSize(sessionState, catalogRel.catalogTable))
+
+      case logicalRel: LogicalRelation if logicalRel.catalogTable.isDefined =>
+        updateStats(logicalRel.catalogTable.get, logicalRel.relation.sizeInBytes)
+
+      case otherRelation =>
+        throw new AnalysisException("ANALYZE TABLE is not supported for " +
+          s"${otherRelation.nodeName}.")
+    }
+
+    def updateStats(catalogTable: CatalogTable, newTotalSize: Long): Unit = {
+      val (rowCount, columnStats) = computeColStats(sparkSession, relation)
+      val statistics = Statistics(
+        sizeInBytes = newTotalSize,
+        rowCount = Some(rowCount),
+        colStats = columnStats ++ catalogTable.stats.map(_.colStats).getOrElse(Map()))
+      sessionState.catalog.alterTable(catalogTable.copy(stats = Some(statistics)))
+      // Refresh the cached data source table in the catalog.
+      sessionState.catalog.refreshTable(tableIdentWithDB)
+    }
+
+    Seq.empty[Row]
+  }
+
+  def computeColStats(
+      sparkSession: SparkSession,
+      relation: LogicalPlan): (Long, Map[String, ColumnStat]) = {
+
+    // check correctness of column names
+    val attributesToAnalyze = mutable.MutableList[Attribute]()
+    val duplicatedColumns = mutable.MutableList[String]()
+    val resolver = sparkSession.sessionState.conf.resolver
+    columnNames.foreach { col =>
+      val exprOption = relation.output.find(attr => resolver(attr.name, col))
+      val expr = exprOption.getOrElse(throw new AnalysisException(s"Invalid column name: $col."))
+      // do deduplication
+      if (!attributesToAnalyze.contains(expr)) {
+        attributesToAnalyze += expr
+      } else {
+        duplicatedColumns += col
+      }
+    }
+    if (duplicatedColumns.nonEmpty) {
+      logWarning(s"Duplicated columns ${duplicatedColumns.mkString("(", ", ", ")")} detected " +
+        s"when analyzing columns ${columnNames.mkString("(", ", ", ")")}, ignoring them.")
+    }
+
+    // Collect statistics per column.
+    // The first element in the result will be the overall row count, the following elements
+    // will be structs containing all column stats.
+    // The layout of each struct follows the layout of the ColumnStats.
+    val ndvMaxErr = sparkSession.sessionState.conf.ndvMaxError
+    val expressions = Count(Literal(1)).toAggregateExpression() +:
+      attributesToAnalyze.map(ColumnStatStruct(_, ndvMaxErr))
+    val namedExpressions = expressions.map(e => Alias(e, e.toString)())
+    val statsRow = Dataset.ofRows(sparkSession, Aggregate(Nil, namedExpressions, relation))
+      .queryExecution.toRdd.collect().head
+
+    // unwrap the result
+    val rowCount = statsRow.getLong(0)
+    val columnStats = attributesToAnalyze.zipWithIndex.map { case (expr, i) =>
+      val numFields = ColumnStatStruct.numStatFields(expr.dataType)
+      (expr.name, ColumnStat(statsRow.getStruct(i + 1, numFields)))
+    }.toMap
+    (rowCount, columnStats)
+  }
+}
+
+object ColumnStatStruct {
+  val zero = Literal(0, LongType)
+  val one = Literal(1, LongType)
+
+  def numNulls(e: Expression): Expression = if (e.nullable) Sum(If(IsNull(e), one, zero)) else zero
+  def max(e: Expression): Expression = Max(e)
+  def min(e: Expression): Expression = Min(e)
+  def ndv(e: Expression, relativeSD: Double): Expression = {
+    // the approximate ndv should never be larger than the number of rows
+    Least(Seq(HyperLogLogPlusPlus(e, relativeSD), Count(one)))
+  }
+  def avgLength(e: Expression): Expression = Average(Length(e))
+  def maxLength(e: Expression): Expression = Max(Length(e))
+  def numTrues(e: Expression): Expression = Sum(If(e, one, zero))
+  def numFalses(e: Expression): Expression = Sum(If(Not(e), one, zero))
+
+  def getStruct(exprs: Seq[Expression]): CreateStruct = {
+    CreateStruct(exprs.map { expr: Expression =>
+      expr.transformUp {
+        case af: AggregateFunction => af.toAggregateExpression()
+      }
+    })
+  }
+
+  def numericColumnStat(e: Expression, relativeSD: Double): Seq[Expression] = {
+    Seq(numNulls(e), max(e), min(e), ndv(e, relativeSD))
+  }
+
+  def stringColumnStat(e: Expression, relativeSD: Double): Seq[Expression] = {
+    Seq(numNulls(e), avgLength(e), maxLength(e), ndv(e, relativeSD))
+  }
+
+  def binaryColumnStat(e: Expression): Seq[Expression] = {
+    Seq(numNulls(e), avgLength(e), maxLength(e))
+  }
+
+  def booleanColumnStat(e: Expression): Seq[Expression] = {
+    Seq(numNulls(e), numTrues(e), numFalses(e))
+  }
+
+  def numStatFields(dataType: DataType): Int = {
+    dataType match {
+      case BinaryType | BooleanType => 3
+      case _ => 4
+    }
+  }
+
+  def apply(e: Attribute, relativeSD: Double): CreateStruct = e.dataType match {
+    // Use aggregate functions to compute statistics we need.
+    case _: NumericType | TimestampType | DateType => getStruct(numericColumnStat(e, relativeSD))
+    case StringType => getStruct(stringColumnStat(e, relativeSD))
+    case BinaryType => getStruct(binaryColumnStat(e))
+    case BooleanType => getStruct(booleanColumnStat(e))
+    case otherType =>
+      throw new AnalysisException("Analyzing columns is not supported for column " +
+        s"${e.name} of data type: ${e.dataType}.")
+  }
+}