[SPARK-6638] [SQL] Improve performance of StringType in SQL

python/pyspark/sql/dataframe.py

@@ -456,7 +456,7 @@ def join(self, other, joinExprs=None, joinType=None):
             One of `inner`, `outer`, `left_outer`, `right_outer`, `semijoin`.
 
         >>> df.join(df2, df.name == df2.name, 'outer').select(df.name, df2.height).collect()
-        [Row(name=None, height=80), Row(name=u'Bob', height=85), Row(name=u'Alice', height=None)]
+        [Row(name=None, height=80), Row(name=u'Alice', height=None), Row(name=u'Bob', height=85)]
         """
 
         if joinExprs is None:
@@ -637,9 +637,9 @@ def groupBy(self, *cols):
         >>> df.groupBy().avg().collect()
         [Row(AVG(age)=3.5)]
         >>> df.groupBy('name').agg({'age': 'mean'}).collect()
-        [Row(name=u'Bob', AVG(age)=5.0), Row(name=u'Alice', AVG(age)=2.0)]
+        [Row(name=u'Alice', AVG(age)=2.0), Row(name=u'Bob', AVG(age)=5.0)]
         >>> df.groupBy(df.name).avg().collect()
-        [Row(name=u'Bob', AVG(age)=5.0), Row(name=u'Alice', AVG(age)=2.0)]
+        [Row(name=u'Alice', AVG(age)=2.0), Row(name=u'Bob', AVG(age)=5.0)]
         """
         jcols = ListConverter().convert([_to_java_column(c) for c in cols],
                                         self._sc._gateway._gateway_client)
@@ -867,11 +867,11 @@ def agg(self, *exprs):
 
         >>> gdf = df.groupBy(df.name)
         >>> gdf.agg({"*": "count"}).collect()
-        [Row(name=u'Bob', COUNT(1)=1), Row(name=u'Alice', COUNT(1)=1)]
+        [Row(name=u'Alice', COUNT(1)=1), Row(name=u'Bob', COUNT(1)=1)]
 
         >>> from pyspark.sql import functions as F
         >>> gdf.agg(F.min(df.age)).collect()
-        [Row(MIN(age)=5), Row(MIN(age)=2)]
+        [Row(MIN(age)=2), Row(MIN(age)=5)]
         """
         assert exprs, "exprs should not be empty"
         if len(exprs) == 1 and isinstance(exprs[0], dict):

sql/catalyst/src/main/scala/org/apache/spark/sql/Row.scala

@@ -20,7 +20,7 @@ package org.apache.spark.sql
 import scala.util.hashing.MurmurHash3
 
 import org.apache.spark.sql.catalyst.expressions.GenericRow
-import org.apache.spark.sql.types.{StructType, DateUtils}
+import org.apache.spark.sql.types.StructType
 
 object Row {
   /**
@@ -257,6 +257,7 @@ trait Row extends Serializable {
    *
    * @throws ClassCastException when data type does not match.
    */
+  // TODO(davies): This is not the right default implementation, we use Int as Date internally
   def getDate(i: Int): java.sql.Date = apply(i).asInstanceOf[java.sql.Date]
 
   /**

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/CatalystTypeConverters.scala

@@ -77,6 +77,9 @@ object CatalystTypeConverters {
       }
       new GenericRowWithSchema(ar, structType)
 
+    case (d: String, _) =>
+      UTF8String(d)
+
     case (d: BigDecimal, _) =>
       Decimal(d)
 
@@ -175,6 +178,11 @@ object CatalystTypeConverters {
         case other => other
       }
 
+      case dataType: StringType => (item: Any) => extractOption(item) match {
+        case s: String => UTF8String(s)
+        case other => other
+      }
+
       case _ =>
         (item: Any) => extractOption(item) match {
           case d: BigDecimal => Decimal(d)
@@ -184,6 +192,26 @@ object CatalystTypeConverters {
     }
   }
 
+  /**
+   *  Converts Scala objects to catalyst rows / types.
+   *
+   *  Note: This should be called before do evaluation on Row
+   *        (It does not support UDT)
+   *  This is used to create an RDD or test results with correct types for Catalyst.
+   */
+  def convertToCatalyst(a: Any): Any = a match {
+    case s: String => UTF8String(s)
+    case d: java.sql.Date => DateUtils.fromJavaDate(d)
+    case d: BigDecimal => Decimal(d)
+    case d: java.math.BigDecimal => Decimal(d)
+    case seq: Seq[Any] => seq.map(convertToCatalyst)
+    case r: Row => Row(r.toSeq.map(convertToCatalyst): _*)
+    case arr: Array[Any] => arr.toSeq.map(convertToCatalyst).toArray
+    case m: Map[Any, Any] =>
+      m.map { case (k, v) => (convertToCatalyst(k), convertToCatalyst(v)) }.toMap
+    case other => other
+  }
+
   /** 
    * Converts Catalyst types used internally in rows to standard Scala types
    * This method is slow, and for batch conversion you should be using converter
@@ -211,6 +239,9 @@ object CatalystTypeConverters {
     case (i: Int, DateType) =>
       DateUtils.toJavaDate(i)
 
+    case (s: UTF8String, StringType) =>
+      s.toString()
+
     case (other, _) =>
       other
   }
@@ -262,6 +293,12 @@ object CatalystTypeConverters {
         case other => other
       }
 
+    case StringType =>
+      (item: Any) => item match {
+        case s: UTF8String => s.toString()
+        case other => other
+      }
+
     case other =>
       (item: Any) => item
   }
@@ -284,9 +321,9 @@ object CatalystTypeConverters {
       row: Row,
       schema: StructType,
       converters: Array[Any => Any]): Row = {
-    val ar = new Array[Any](row.size)
+    val ar = new Array[Any](converters.size)
     var idx = 0
-    while (idx < row.size) {
+    while (idx < converters.size && idx < row.size) {
       ar(idx) = converters(idx)(row(idx))
       idx += 1
     }

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/ScalaReflection.scala

@@ -138,6 +138,7 @@ trait ScalaReflection {
     // The data type can be determined without ambiguity.
     case obj: BooleanType.JvmType => BooleanType
     case obj: BinaryType.JvmType => BinaryType
+    case obj: String => StringType
     case obj: StringType.JvmType => StringType
     case obj: ByteType.JvmType => ByteType
     case obj: ShortType.JvmType => ShortType

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/HiveTypeCoercion.scala

@@ -115,7 +115,7 @@ trait HiveTypeCoercion {
    * the appropriate numeric equivalent.
    */
   object ConvertNaNs extends Rule[LogicalPlan] {
-    val stringNaN = Literal.create("NaN", StringType)
+    val stringNaN = Literal("NaN")
 
     def apply(plan: LogicalPlan): LogicalPlan = plan transform {
       case q: LogicalPlan => q transformExpressions {
@@ -563,6 +563,10 @@ trait HiveTypeCoercion {
       case Sum(e @ TimestampType()) => Sum(Cast(e, DoubleType))
       case Average(e @ TimestampType()) => Average(Cast(e, DoubleType))
 
+      // Compatible with Hive
+      case Substring(e, start, len) if e.dataType != StringType =>
+        Substring(Cast(e, StringType), start, len)
+
       // Coalesce should return the first non-null value, which could be any column
       // from the list. So we need to make sure the return type is deterministic and
       // compatible with every child column.

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Cast.scala

@@ -21,7 +21,6 @@ import java.sql.{Date, Timestamp}
 import java.text.{DateFormat, SimpleDateFormat}
 
 import org.apache.spark.Logging
-import org.apache.spark.sql.catalyst.errors.TreeNodeException
 import org.apache.spark.sql.types._
 
 /** Cast the child expression to the target data type. */
@@ -112,21 +111,21 @@ case class Cast(child: Expression, dataType: DataType) extends UnaryExpression w
 
   // UDFToString
   private[this] def castToString(from: DataType): Any => Any = from match {
-    case BinaryType => buildCast[Array[Byte]](_, new String(_, "UTF-8"))
-    case DateType => buildCast[Int](_, d => DateUtils.toString(d))
-    case TimestampType => buildCast[Timestamp](_, timestampToString)
-    case _ => buildCast[Any](_, _.toString)
+    case BinaryType => buildCast[Array[Byte]](_, UTF8String(_))
+    case DateType => buildCast[Int](_, d => UTF8String(DateUtils.toString(d)))
+    case TimestampType => buildCast[Timestamp](_, t => UTF8String(timestampToString(t)))
+    case _ => buildCast[Any](_, o => UTF8String(o.toString))
   }
 
   // BinaryConverter
   private[this] def castToBinary(from: DataType): Any => Any = from match {
-    case StringType => buildCast[String](_, _.getBytes("UTF-8"))
+    case StringType => buildCast[UTF8String](_, _.getBytes)
   }
 
   // UDFToBoolean
   private[this] def castToBoolean(from: DataType): Any => Any = from match {
     case StringType =>
-      buildCast[String](_, _.length() != 0)
+      buildCast[UTF8String](_, _.length() != 0)
     case TimestampType =>
       buildCast[Timestamp](_, t => t.getTime() != 0 || t.getNanos() != 0)
     case DateType =>
@@ -151,8 +150,9 @@ case class Cast(child: Expression, dataType: DataType) extends UnaryExpression w
   // TimestampConverter
   private[this] def castToTimestamp(from: DataType): Any => Any = from match {
     case StringType =>
-      buildCast[String](_, s => {
+      buildCast[UTF8String](_, utfs => {
         // Throw away extra if more than 9 decimal places
+        val s = utfs.toString
         val periodIdx = s.indexOf(".")
         var n = s
         if (periodIdx != -1 && n.length() - periodIdx > 9) {
@@ -227,8 +227,8 @@ case class Cast(child: Expression, dataType: DataType) extends UnaryExpression w
   // DateConverter
   private[this] def castToDate(from: DataType): Any => Any = from match {
     case StringType =>
-      buildCast[String](_, s =>
-        try DateUtils.fromJavaDate(Date.valueOf(s))
+      buildCast[UTF8String](_, s =>
+        try DateUtils.fromJavaDate(Date.valueOf(s.toString))
         catch { case _: java.lang.IllegalArgumentException => null }
       )
     case TimestampType =>
@@ -245,7 +245,7 @@ case class Cast(child: Expression, dataType: DataType) extends UnaryExpression w
   // LongConverter
   private[this] def castToLong(from: DataType): Any => Any = from match {
     case StringType =>
-      buildCast[String](_, s => try s.toLong catch {
+      buildCast[UTF8String](_, s => try s.toString.toLong catch {
         case _: NumberFormatException => null
       })
     case BooleanType =>
@@ -261,7 +261,7 @@ case class Cast(child: Expression, dataType: DataType) extends UnaryExpression w
   // IntConverter
   private[this] def castToInt(from: DataType): Any => Any = from match {
     case StringType =>
-      buildCast[String](_, s => try s.toInt catch {
+      buildCast[UTF8String](_, s => try s.toString.toInt catch {
         case _: NumberFormatException => null
       })
     case BooleanType =>
@@ -277,7 +277,7 @@ case class Cast(child: Expression, dataType: DataType) extends UnaryExpression w
   // ShortConverter
   private[this] def castToShort(from: DataType): Any => Any = from match {
     case StringType =>
-      buildCast[String](_, s => try s.toShort catch {
+      buildCast[UTF8String](_, s => try s.toString.toShort catch {
         case _: NumberFormatException => null
       })
     case BooleanType =>
@@ -293,7 +293,7 @@ case class Cast(child: Expression, dataType: DataType) extends UnaryExpression w
   // ByteConverter
   private[this] def castToByte(from: DataType): Any => Any = from match {
     case StringType =>
-      buildCast[String](_, s => try s.toByte catch {
+      buildCast[UTF8String](_, s => try s.toString.toByte catch {
         case _: NumberFormatException => null
       })
     case BooleanType =>
@@ -323,7 +323,9 @@ case class Cast(child: Expression, dataType: DataType) extends UnaryExpression w
 
   private[this] def castToDecimal(from: DataType, target: DecimalType): Any => Any = from match {
     case StringType =>
-      buildCast[String](_, s => try changePrecision(Decimal(s.toDouble), target) catch {
+      buildCast[UTF8String](_, s => try {
+        changePrecision(Decimal(s.toString.toDouble), target)
+      } catch {
         case _: NumberFormatException => null
       })
     case BooleanType =>
@@ -348,7 +350,7 @@ case class Cast(child: Expression, dataType: DataType) extends UnaryExpression w
   // DoubleConverter
   private[this] def castToDouble(from: DataType): Any => Any = from match {
     case StringType =>
-      buildCast[String](_, s => try s.toDouble catch {
+      buildCast[UTF8String](_, s => try s.toString.toDouble catch {
         case _: NumberFormatException => null
       })
     case BooleanType =>
@@ -364,7 +366,7 @@ case class Cast(child: Expression, dataType: DataType) extends UnaryExpression w
   // FloatConverter
   private[this] def castToFloat(from: DataType): Any => Any = from match {
     case StringType =>
-      buildCast[String](_, s => try s.toFloat catch {
+      buildCast[UTF8String](_, s => try s.toString.toFloat catch {
         case _: NumberFormatException => null
       })
     case BooleanType =>

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/SpecificMutableRow.scala

@@ -230,13 +230,17 @@ final class SpecificMutableRow(val values: Array[MutableValue]) extends MutableR
     new GenericRow(newValues)
   }
 
-  override def update(ordinal: Int, value: Any): Unit = {
-    if (value == null) setNullAt(ordinal) else values(ordinal).update(value)
+  override def update(ordinal: Int, value: Any) {
+    if (value == null) {
+      setNullAt(ordinal)
+    } else {
+      values(ordinal).update(value)
+    }
   }
 
-  override def setString(ordinal: Int, value: String): Unit = update(ordinal, value)
+  override def setString(ordinal: Int, value: String): Unit = update(ordinal, UTF8String(value))
 
-  override def getString(ordinal: Int): String = apply(ordinal).asInstanceOf[String]
+  override def getString(ordinal: Int): String = apply(ordinal).toString
 
   override def setInt(ordinal: Int, value: Int): Unit = {
     val currentValue = values(ordinal).asInstanceOf[MutableInt]