Package org.apache.druid.segment

Class DictionaryEncodedColumnIndexer<KeyType,ActualType extends Comparable<ActualType>>

java.lang.Object

org.apache.druid.segment.DictionaryEncodedColumnIndexer<KeyType,ActualType>

All Implemented Interfaces:

DimensionIndexer<Integer,KeyType,ActualType>

Direct Known Subclasses:

StringDimensionIndexer

public abstract class DictionaryEncodedColumnIndexer<KeyType,ActualType extends Comparable<ActualType>>
extends Object
implements DimensionIndexer<Integer,KeyType,ActualType>

Basic structure for indexing dictionary encoded columns

Field Summary

Fields

Modifier and Type	Field	Description
protected final DimensionDictionary<ActualType>	dimLookup
protected boolean	isSparse
protected SortedDimensionDictionary<ActualType>	sortedLookup

Constructor Summary

Constructors

Constructor	Description
DictionaryEncodedColumnIndexer(@NotNull DimensionDictionary<ActualType> dimLookup)	Creates a new DictionaryEncodedColumnIndexer.

Method Summary

Modifier and Type	Method	Description
ColumnValueSelector	convertUnsortedValuesToSorted(ColumnValueSelector selectorWithUnsortedValues)	Converts dictionary-encoded row values from unspecified (random) encoding order, to sorted encoding.
protected ActualType	getActualValue(int intermediateValue, boolean idSorted)
int	getCardinality()	Get the cardinality of this dimension's values.
protected int	getEncodedValue(ActualType fullValue, boolean idSorted)
ActualType	getMaxValue()	Get the maximum dimension value seen by this indexer.
ActualType	getMinValue()	Get the minimum dimension value seen by this indexer.
int	getSortedEncodedValueFromUnsorted(Integer unsortedIntermediateValue)
CloseableIndexed<ActualType>	getSortedIndexedValues()	Returns an indexed structure of this dimension's sorted actual values.
Integer	getUnsortedEncodedValueFromSorted(Integer sortedIntermediateValue)	Given an encoded value that was ordered by associated actual value, return the equivalent encoded value ordered by time of ingestion.
ColumnValueSelector<?>	makeColumnValueSelector(IncrementalIndexRowHolder currEntry, IncrementalIndex.DimensionDesc desc)	Return an object used to read values from this indexer's column.
void	setSparseIndexed()	This method will be called while building an IncrementalIndex whenever a known dimension column (either through an explicit schema on the ingestion spec, or auto-discovered while processing rows) is absent in any row that is processed, to allow an indexer to account for any missing rows if necessary.
protected SortedDimensionDictionary<ActualType>	sortedLookup()

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.apache.druid.segment.DimensionIndexer

checkUnsortedEncodedKeyComponentsEqual, compareUnsortedEncodedKeyComponents, convertUnsortedEncodedKeyComponentToActualList, fillBitmapsFromUnsortedEncodedKeyComponent, getColumnCapabilities, getFormat, getUnsortedEncodedKeyComponentHashCode, makeDimensionSelector, processRowValsToUnsortedEncodedKeyComponent

Field Details

dimLookup

protected final DimensionDictionary<ActualType extends Comparable<ActualType>> dimLookup

isSparse

protected volatile boolean isSparse

sortedLookup

@Nullable
protected SortedDimensionDictionary<ActualType extends Comparable<ActualType>> sortedLookup

Constructor Details

DictionaryEncodedColumnIndexer

public DictionaryEncodedColumnIndexer(@NotNull
 @NotNull DimensionDictionary<ActualType> dimLookup)

Creates a new DictionaryEncodedColumnIndexer.

Parameters:

dimLookup - Dimension Dictionary to lookup dimension values.

Method Details

setSparseIndexed

public void setSparseIndexed()

Description copied from interface: DimensionIndexer

This method will be called while building an IncrementalIndex whenever a known dimension column (either through an explicit schema on the ingestion spec, or auto-discovered while processing rows) is absent in any row that is processed, to allow an indexer to account for any missing rows if necessary. Useful so that a string DimensionSelector built on top of an IncrementalIndex may accurately report DimensionDictionarySelector.nameLookupPossibleInAdvance() by allowing it to track if it has any implicit null valued rows. At index persist/merge time all missing columns for a row will be explicitly replaced with the value appropriate null or default value.

Specified by:

setSparseIndexed in interface DimensionIndexer<Integer,KeyType,ActualType extends Comparable<ActualType>>

getSortedEncodedValueFromUnsorted

public int getSortedEncodedValueFromUnsorted(Integer unsortedIntermediateValue)

getUnsortedEncodedValueFromSorted

public Integer getUnsortedEncodedValueFromSorted(Integer sortedIntermediateValue)

Description copied from interface: DimensionIndexer

Given an encoded value that was ordered by associated actual value, return the equivalent encoded value ordered by time of ingestion. Using the example in the class description: getUnsortedEncodedValueFromSorted(2) would return 0

Specified by:

getUnsortedEncodedValueFromSorted in interface DimensionIndexer<Integer,KeyType,ActualType extends Comparable<ActualType>>

Parameters:

sortedIntermediateValue - value to convert

Returns:

converted value

getSortedIndexedValues

public CloseableIndexed<ActualType> getSortedIndexedValues()

Description copied from interface: DimensionIndexer

Returns an indexed structure of this dimension's sorted actual values. The integer IDs represent the ordering of the sorted values. Using the example in the class description: "Apple"=0, "Hello"=1, "World"=2

Specified by:

getSortedIndexedValues in interface DimensionIndexer<Integer,KeyType,ActualType extends Comparable<ActualType>>

Returns:

Sorted index of actual values

getMinValue

public ActualType getMinValue()

Description copied from interface: DimensionIndexer

Get the minimum dimension value seen by this indexer. NOTE: On an in-memory segment (IncrementalIndex), we can determine min/max values by looking at the stream of row values seen in calls to processSingleRowValToIndexKey(). However, on a disk-backed segment (QueryableIndex), the numeric dimensions do not currently have any supporting index structures that can be used to efficiently determine min/max values. When numeric dimension support is added, the segment format should be changed to store min/max values, to avoid performing a full-column scan to determine these values for numeric dims.

Specified by:

getMinValue in interface DimensionIndexer<Integer,KeyType,ActualType extends Comparable<ActualType>>

Returns:

min value

getMaxValue

public ActualType getMaxValue()

Description copied from interface: DimensionIndexer

Get the maximum dimension value seen by this indexer.

Specified by:

getMaxValue in interface DimensionIndexer<Integer,KeyType,ActualType extends Comparable<ActualType>>

Returns:

max value

getCardinality

public int getCardinality()

Description copied from interface: DimensionIndexer

Get the cardinality of this dimension's values.

Specified by:

getCardinality in interface DimensionIndexer<Integer,KeyType,ActualType extends Comparable<ActualType>>

Returns:

value cardinality

makeColumnValueSelector

public ColumnValueSelector<?> makeColumnValueSelector(IncrementalIndexRowHolder currEntry,
 IncrementalIndex.DimensionDesc desc)

Description copied from interface: DimensionIndexer

Return an object used to read values from this indexer's column.

Specified by:

makeColumnValueSelector in interface DimensionIndexer<Integer,KeyType,ActualType extends Comparable<ActualType>>

Parameters:

currEntry - Provides access to the current Row object in the Cursor

desc - Descriptor object for this dimension within an IncrementalIndex

Returns:

A new object that reads rows from currEntry

convertUnsortedValuesToSorted

public ColumnValueSelector convertUnsortedValuesToSorted(ColumnValueSelector selectorWithUnsortedValues)

Description copied from interface: DimensionIndexer

Converts dictionary-encoded row values from unspecified (random) encoding order, to sorted encoding. This step is needed to be able to correctly map per-segment encoded values to global values on the next conversion step, DimensionMerger.convertSortedSegmentRowValuesToMergedRowValues(int, org.apache.druid.segment.ColumnValueSelector). The latter method requires sorted encoding values on the input, because DimensionMerger.writeMergedValueDictionary(java.util.List<org.apache.druid.segment.IndexableAdapter>) takes sorted lookups as it's input. For columns which do not use the DimensionMerger to merge dictionary encoded values, this method should provide a selector which is compatible with the expectations of DimensionMerger.processMergedRow(ColumnValueSelector), which might simply be to pass-through the 'unsorted' selector.

Specified by:

convertUnsortedValuesToSorted in interface DimensionIndexer<Integer,KeyType,ActualType extends Comparable<ActualType>>

sortedLookup

protected SortedDimensionDictionary<ActualType> sortedLookup()

getActualValue

@Nullable
protected ActualType getActualValue(int intermediateValue,
 boolean idSorted)

getEncodedValue

protected int getEncodedValue(@Nullable
 ActualType fullValue,
 boolean idSorted)