Package org.apache.druid.segment.index

Class DictionaryScanningBitmapIndex

  • All Implemented Interfaces:
    BitmapColumnIndex
    public abstract class DictionaryScanningBitmapIndex
extends SimpleImmutableBitmapIterableIndex
    Common BitmapColumnIndex implementation for indexes which are computed by scanning the entire value dictionary of the underlying column to check if the value index matches the filter. Contains logic to skip computing indexes with computeBitmapResult(BitmapResultFactory, int, int, boolean) if 'selectionRowCount' does not equal 'totalRowCount' and 'selectionRowCount' is smaller than dictionarySize multiplied by scaleThreshold. The default scaleThreshold value is 1.0, meaning that if dictionarySize is larger than 'selectionRowCount' we skip using indexes, the idea being we would either have to perform the check against the values in the dictionary or the values in the remaining rows, since remaining rows is smaller we should just do that instead of spending time to compute indexes to further shrink 'selectionRowCount'.

    • Constructor Detail

      • DictionaryScanningBitmapIndex

        public DictionaryScanningBitmapIndex​(int dictionarySize)

      • DictionaryScanningBitmapIndex

        public DictionaryScanningBitmapIndex​(int dictionarySize,
                                     double scaleThreshold)

    • Method Detail

      • computeBitmapResult

        @Nullable
public final <T> T computeBitmapResult​(BitmapResultFactory<T> bitmapResultFactory,
                                       int applyRowCount,
                                       int totalRowCount,
                                       boolean includeUnknown)
        Description copied from interface: BitmapColumnIndex
        Compute a bitmap result wrapped with the BitmapResultFactory representing the rows matched by this index, or null if the index cannot (or should not) be computed.
        Parameters:
        bitmapResultFactory - helper to format the ImmutableBitmap in a form ready for consumption by callers
        applyRowCount - upper bound on number of rows this filter would be applied to, after removing rows short-circuited by prior bundle operations. For example, given "x AND y", if "x" is resolved using an index, then "y" will receive the number of rows that matched the filter "x". As another example, given "x OR y", if "x" is resolved using an index, then "y" will receive the number of rows that did *not* match the filter "x".
        totalRowCount - total number of rows to be scanned if no indexes are used
        includeUnknown - mapping for Druid native two state logic system into SQL three-state logic system. If set to true, bitmaps returned by this method should include true bits for any rows where the matching result is 'unknown', such as from the input being null valued. See NullHandling.useThreeValueLogic().
        Returns:
        bitmap result representing rows matched by this index