Package org.apache.spark.sql.vectorized

Class ColumnVector

Object
org.apache.spark.sql.vectorized.ColumnVector
All Implemented Interfaces:
AutoCloseable
Direct Known Subclasses:
ArrowColumnVector
@Evolving public abstract class ColumnVector extends Object implements AutoCloseable
An interface representing in-memory columnar data in Spark. This interface defines the main APIs to access the data, as well as their batched versions. The batched versions are considered to be faster and preferable whenever possible.

Most of the APIs take the rowId as a parameter. This is the batch local 0-based row id for values in this ColumnVector.

Spark only calls specific get method according to the data type of this ColumnVector, e.g. if it's int type, Spark is guaranteed to only call getInt(int) or getInts(int, int).

ColumnVector supports all the data types including nested types. To handle nested types, ColumnVector can have children and is a tree structure. Please refer to getStruct(int), getArray(int) and getMap(int) for the details about how to implement nested types.

ColumnVector is expected to be reused during the entire data loading process, to avoid allocating memory again and again.

ColumnVector is meant to maximize CPU efficiency but not to minimize storage footprint. Implementations should prefer computing efficiency over storage efficiency when design the format. Since it is expected to reuse the ColumnVector instance while loading data, the storage footprint is negligible.

  • Method Summary

    Modifier and Type
    Method
    Description
    abstract void
    close()
    Cleans up memory for this column vector.
    final DataType
    dataType()
    Returns the data type of this column vector.
    abstract ColumnarArray
    getArray(int rowId)
    Returns the array type value for rowId.
    abstract byte[]
    getBinary(int rowId)
    Returns the binary type value for rowId.
    abstract boolean
    getBoolean(int rowId)
    Returns the boolean type value for rowId.
    boolean[]
    getBooleans(int rowId, int count)
    Gets boolean type values from [rowId, rowId + count).
    abstract byte
    getByte(int rowId)
    Returns the byte type value for rowId.
    byte[]
    getBytes(int rowId, int count)
    Gets byte type values from [rowId, rowId + count).
    abstract ColumnVector
    getChild(int ordinal)
     
    abstract Decimal
    getDecimal(int rowId, int precision, int scale)
    Returns the decimal type value for rowId.
    abstract double
    getDouble(int rowId)
    Returns the double type value for rowId.
    double[]
    getDoubles(int rowId, int count)
    Gets double type values from [rowId, rowId + count).
    abstract float
    getFloat(int rowId)
    Returns the float type value for rowId.
    float[]
    getFloats(int rowId, int count)
    Gets float type values from [rowId, rowId + count).
    abstract int
    getInt(int rowId)
    Returns the int type value for rowId.
    CalendarInterval
    getInterval(int rowId)
    Returns the calendar interval type value for rowId.
    int[]
    getInts(int rowId, int count)
    Gets int type values from [rowId, rowId + count).
    abstract long
    getLong(int rowId)
    Returns the long type value for rowId.
    long[]
    getLongs(int rowId, int count)
    Gets long type values from [rowId, rowId + count).
    abstract ColumnarMap
    getMap(int ordinal)
    Returns the map type value for rowId.
    abstract short
    getShort(int rowId)
    Returns the short type value for rowId.
    short[]
    getShorts(int rowId, int count)
    Gets short type values from [rowId, rowId + count).
    final ColumnarRow
    getStruct(int rowId)
    Returns the struct type value for rowId.
    abstract org.apache.spark.unsafe.types.UTF8String
    getUTF8String(int rowId)
    Returns the string type value for rowId.
    final org.apache.spark.unsafe.types.VariantVal
    getVariant(int rowId)
    Returns the Variant value for rowId.
    abstract boolean
    hasNull()
    Returns true if this column vector contains any null values.
    abstract boolean
    isNullAt(int rowId)
    Returns whether the value at rowId is NULL.
    abstract int
    numNulls()
    Returns the number of nulls in this column vector.

    Methods inherited from class java.lang.Object

    equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

  • Method Details

    • dataType

      public final DataType dataType()
      Returns the data type of this column vector.

    • close

      public abstract void close()
      Cleans up memory for this column vector. The column vector is not usable after this.

      This overwrites AutoCloseable.close() to remove the throws clause, as column vector is in-memory and we don't expect any exception to happen during closing.

      Specified by:
      close in interface AutoCloseable

    • hasNull

      public abstract boolean hasNull()
      Returns true if this column vector contains any null values.

    • numNulls

      public abstract int numNulls()
      Returns the number of nulls in this column vector.

    • isNullAt

      public abstract boolean isNullAt(int rowId)
      Returns whether the value at rowId is NULL.

    • getBoolean

      public abstract boolean getBoolean(int rowId)
      Returns the boolean type value for rowId. The return value is undefined and can be anything, if the slot for rowId is null.

    • getBooleans

      public boolean[] getBooleans(int rowId, int count)
      Gets boolean type values from [rowId, rowId + count). The return values for the null slots are undefined and can be anything.

    • getByte

      public abstract byte getByte(int rowId)
      Returns the byte type value for rowId. The return value is undefined and can be anything, if the slot for rowId is null.

    • getBytes

      public byte[] getBytes(int rowId, int count)
      Gets byte type values from [rowId, rowId + count). The return values for the null slots are undefined and can be anything.

    • getShort

      public abstract short getShort(int rowId)
      Returns the short type value for rowId. The return value is undefined and can be anything, if the slot for rowId is null.

    • getShorts

      public short[] getShorts(int rowId, int count)
      Gets short type values from [rowId, rowId + count). The return values for the null slots are undefined and can be anything.

    • getInt

      public abstract int getInt(int rowId)
      Returns the int type value for rowId. The return value is undefined and can be anything, if the slot for rowId is null.

    • getInts

      public int[] getInts(int rowId, int count)
      Gets int type values from [rowId, rowId + count). The return values for the null slots are undefined and can be anything.

    • getLong

      public abstract long getLong(int rowId)
      Returns the long type value for rowId. The return value is undefined and can be anything, if the slot for rowId is null.

    • getLongs

      public long[] getLongs(int rowId, int count)
      Gets long type values from [rowId, rowId + count). The return values for the null slots are undefined and can be anything.

    • getFloat

      public abstract float getFloat(int rowId)
      Returns the float type value for rowId. The return value is undefined and can be anything, if the slot for rowId is null.

    • getFloats

      public float[] getFloats(int rowId, int count)
      Gets float type values from [rowId, rowId + count). The return values for the null slots are undefined and can be anything.

    • getDouble

      public abstract double getDouble(int rowId)
      Returns the double type value for rowId. The return value is undefined and can be anything, if the slot for rowId is null.

    • getDoubles

      public double[] getDoubles(int rowId, int count)
      Gets double type values from [rowId, rowId + count). The return values for the null slots are undefined and can be anything.

    • getStruct

      public final ColumnarRow getStruct(int rowId)
      Returns the struct type value for rowId. If the slot for rowId is null, it should return null.

      To support struct type, implementations must implement getChild(int) and make this vector a tree structure. The number of child vectors must be same as the number of fields of the struct type, and each child vector is responsible to store the data for its corresponding struct field.

    • getArray

      public abstract ColumnarArray getArray(int rowId)
      Returns the array type value for rowId. If the slot for rowId is null, it should return null.

      To support array type, implementations must construct an ColumnarArray and return it in this method. ColumnarArray requires a ColumnVector that stores the data of all the elements of all the arrays in this vector, and an offset and length which points to a range in that ColumnVector, and the range represents the array for rowId. Implementations are free to decide where to put the data vector and offsets and lengths. For example, we can use the first child vector as the data vector, and store offsets and lengths in 2 int arrays in this vector.

    • getMap

      public abstract ColumnarMap getMap(int ordinal)
      Returns the map type value for rowId. If the slot for rowId is null, it should return null.

      In Spark, map type value is basically a key data array and a value data array. A key from the key array with a index and a value from the value array with the same index contribute to an entry of this map type value.

      To support map type, implementations must construct a ColumnarMap and return it in this method. ColumnarMap requires a ColumnVector that stores the data of all the keys of all the maps in this vector, and another ColumnVector that stores the data of all the values of all the maps in this vector, and a pair of offset and length which specify the range of the key/value array that belongs to the map type value at rowId.

    • getDecimal

      public abstract Decimal getDecimal(int rowId, int precision, int scale)
      Returns the decimal type value for rowId. If the slot for rowId is null, it should return null.

    • getUTF8String

      public abstract org.apache.spark.unsafe.types.UTF8String getUTF8String(int rowId)
      Returns the string type value for rowId. If the slot for rowId is null, it should return null.

      Note that the returned UTF8String may point to the data of this column vector, please copy it if you want to keep it after this column vector is freed.

    • getBinary

      public abstract byte[] getBinary(int rowId)
      Returns the binary type value for rowId. If the slot for rowId is null, it should return null.

    • getInterval

      public CalendarInterval getInterval(int rowId)
      Returns the calendar interval type value for rowId. If the slot for rowId is null, it should return null.

      In Spark, calendar interval type value is basically two integer values representing the number of months and days in this interval, and a long value representing the number of microseconds in this interval. An interval type vector is the same as a struct type vector with 3 fields: months, days and microseconds.

      To support interval type, implementations must implement getChild(int) and define 3 child vectors: the first child vector is an int type vector, containing all the month values of all the interval values in this vector. The second child vector is an int type vector, containing all the day values of all the interval values in this vector. The third child vector is a long type vector, containing all the microsecond values of all the interval values in this vector. Note that the ArrowColumnVector leverages its built-in IntervalMonthDayNanoVector instead of above-mentioned protocol.

    • getVariant

      public final org.apache.spark.unsafe.types.VariantVal getVariant(int rowId)
      Returns the Variant value for rowId. Similar to getInterval(int), the implementation must implement getChild(int) and define 2 child vectors of binary type for the Variant value and metadata.

    • getChild

      public abstract ColumnVector getChild(int ordinal)
      Returns:
      child ColumnVector at the given ordinal.