Package org.apache.druid.segment

Class AutoTypeColumnIndexer

java.lang.Object

org.apache.druid.segment.AutoTypeColumnIndexer

All Implemented Interfaces:

DimensionIndexer<StructuredData,StructuredData,StructuredData>

public class AutoTypeColumnIndexer
extends Object
implements DimensionIndexer<StructuredData,StructuredData,StructuredData>

Field Summary

Fields

Modifier and Type	Field	Description
protected final ExpressionType	castToExpressionType
protected final ColumnType	castToType
protected Object	constantValue
protected int	estimatedFieldKeySize
protected SortedMap<String,org.apache.druid.segment.AutoTypeColumnIndexer.FieldIndexer>	fieldIndexers
protected final ValueDictionary	globalDictionary
protected boolean	hasNestedData	Have we seen any objects? Empty objects in particular are sneaky, they don't have any nested paths, so we also broadly track if we have processed any objects StructuredDataProcessor.ProcessResults.hasObjects()
protected boolean	hasNulls	have we seen any null values?
protected final StructuredDataProcessor	indexerProcessor
protected boolean	isConstant

Constructor Summary

Constructors

Constructor	Description
AutoTypeColumnIndexer(String name, ColumnType castToType, NestedCommonFormatColumnFormatSpec columnFormatSpec)

Method Summary

Modifier and Type	Method	Description
boolean	checkUnsortedEncodedKeyComponentsEqual(StructuredData lhs, StructuredData rhs)	Check if two row value arrays from Row keys are equal.
int	compareUnsortedEncodedKeyComponents(StructuredData lhs, StructuredData rhs)	Compares the row values for this DimensionIndexer's dimension from a Row key.
Object	convertUnsortedEncodedKeyComponentToActualList(StructuredData key)	Given a row value array from a Row key, as described in the documentation for DimensionIndexer.compareUnsortedEncodedKeyComponents(EncodedKeyComponentType, EncodedKeyComponentType), convert the unsorted encoded values to a list of actual values.
ColumnValueSelector	convertUnsortedValuesToSorted(ColumnValueSelector selectorWithUnsortedValues)	Converts dictionary-encoded row values from unspecified (random) encoding order, to sorted encoding.
void	fillBitmapsFromUnsortedEncodedKeyComponent(StructuredData key, int rowNum, MutableBitmap[] bitmapIndexes, BitmapFactory factory)	Helper function for building bitmap indexes for integer-encoded dimensions.
int	getCardinality()	Get the cardinality of this dimension's values.
ColumnCapabilities	getColumnCapabilities()
Object	getConstantValue()
SortedMap<String,FieldTypeInfo.MutableTypeSet>	getFieldTypeInfo()
ColumnFormat	getFormat()
ColumnType	getLogicalType()
StructuredData	getMaxValue()	Get the maximum dimension value seen by this indexer.
StructuredData	getMinValue()	Get the minimum dimension value seen by this indexer.
CloseableIndexed<StructuredData>	getSortedIndexedValues()	Returns an indexed structure of this dimension's sorted actual values.
SortedValueDictionary	getSortedValueLookups()
int	getUnsortedEncodedKeyComponentHashCode(StructuredData key)	Given a row value array from a Row key, generate a hashcode.
StructuredData	getUnsortedEncodedValueFromSorted(StructuredData sortedIntermediateValue)	Given an encoded value that was ordered by associated actual value, return the equivalent encoded value ordered by time of ingestion.
boolean	isConstant()
ColumnValueSelector<?>	makeColumnValueSelector(IncrementalIndexRowHolder currEntry, IncrementalIndex.DimensionDesc desc)	Return an object used to read values from this indexer's column.
DimensionSelector	makeDimensionSelector(DimensionSpec spec, IncrementalIndexRowHolder currEntry, IncrementalIndex.DimensionDesc desc)	Return an object used to read values from this indexer's column as Strings.
EncodedKeyComponent<StructuredData>	processRowValsToUnsortedEncodedKeyComponent(Object dimValues, boolean reportParseExceptions)	Encodes the given row value(s) of the dimension to be used within a row key.
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.

Methods inherited from class java.lang.Object

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

Field Details

hasNulls

protected volatile boolean hasNulls

have we seen any null values?

hasNestedData

protected volatile boolean hasNestedData

Have we seen any objects? Empty objects in particular are sneaky, they don't have any nested paths, so we also broadly track if we have processed any objects StructuredDataProcessor.ProcessResults.hasObjects()

isConstant

protected volatile boolean isConstant

constantValue

@Nullable
protected volatile Object constantValue

fieldIndexers

protected SortedMap<String,org.apache.druid.segment.AutoTypeColumnIndexer.FieldIndexer> fieldIndexers

globalDictionary

protected final ValueDictionary globalDictionary

estimatedFieldKeySize

protected int estimatedFieldKeySize

castToType

@Nullable
protected final ColumnType castToType

castToExpressionType

@Nullable
protected final ExpressionType castToExpressionType

indexerProcessor

protected final StructuredDataProcessor indexerProcessor

Constructor Details

AutoTypeColumnIndexer

public AutoTypeColumnIndexer(String name,
 @Nullable
 ColumnType castToType,
 NestedCommonFormatColumnFormatSpec columnFormatSpec)

Method Details

processRowValsToUnsortedEncodedKeyComponent

public EncodedKeyComponent<StructuredData> processRowValsToUnsortedEncodedKeyComponent(@Nullable
 Object dimValues,
 boolean reportParseExceptions)

Description copied from interface: DimensionIndexer

Encodes the given row value(s) of the dimension to be used within a row key. It also updates the internal state of the DimensionIndexer, e.g. the dimLookup.

For example, the dictionary-encoded String-type column will return an int[] containing dictionary IDs.

Specified by:

processRowValsToUnsortedEncodedKeyComponent in interface DimensionIndexer<StructuredData,StructuredData,StructuredData>

Parameters:

dimValues - Value(s) of the dimension in a row. This can either be a single value or a list of values (for multi-valued dimensions)

reportParseExceptions - true if parse exceptions should be reported, false otherwise

Returns:

Encoded dimension value(s) to be used as a component for the row key. Contains an object of the DimensionIndexer and the effective size of the key component in bytes.

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<StructuredData,StructuredData,StructuredData>

getUnsortedEncodedValueFromSorted

public StructuredData getUnsortedEncodedValueFromSorted(StructuredData 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<StructuredData,StructuredData,StructuredData>

Parameters:

sortedIntermediateValue - value to convert

Returns:

converted value

getSortedIndexedValues

public CloseableIndexed<StructuredData> 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<StructuredData,StructuredData,StructuredData>

Returns:

Sorted index of actual values

getSortedValueLookups

public SortedValueDictionary getSortedValueLookups()

getFieldTypeInfo

public SortedMap<String,FieldTypeInfo.MutableTypeSet> getFieldTypeInfo()

getMinValue

public StructuredData 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<StructuredData,StructuredData,StructuredData>

Returns:

min value

getMaxValue

public StructuredData getMaxValue()

Description copied from interface: DimensionIndexer

Get the maximum dimension value seen by this indexer.

Specified by:

getMaxValue in interface DimensionIndexer<StructuredData,StructuredData,StructuredData>

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<StructuredData,StructuredData,StructuredData>

Returns:

value cardinality

makeDimensionSelector

public DimensionSelector makeDimensionSelector(DimensionSpec spec,
 IncrementalIndexRowHolder currEntry,
 IncrementalIndex.DimensionDesc desc)

Description copied from interface: DimensionIndexer

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

Specified by:

makeDimensionSelector in interface DimensionIndexer<StructuredData,StructuredData,StructuredData>

Parameters:

spec - Specifies the output name of a dimension and any extraction functions to be applied.

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

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<StructuredData,StructuredData,StructuredData>

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

getColumnCapabilities

public ColumnCapabilities getColumnCapabilities()

Specified by:

getColumnCapabilities in interface DimensionIndexer<StructuredData,StructuredData,StructuredData>

getLogicalType

public ColumnType getLogicalType()

isConstant

public boolean isConstant()

getConstantValue

@Nullable
public Object getConstantValue()

getFormat

public ColumnFormat getFormat()

Specified by:

getFormat in interface DimensionIndexer<StructuredData,StructuredData,StructuredData>

compareUnsortedEncodedKeyComponents

public int compareUnsortedEncodedKeyComponents(@Nullable
 StructuredData lhs,
 @Nullable
 StructuredData rhs)

Description copied from interface: DimensionIndexer

Compares the row values for this DimensionIndexer's dimension from a Row key. The dimension value arrays within a Row key always use the "unsorted" ordering for encoded values. The row values are passed to this function as an Object, the implementer should cast them to the type appropriate for this dimension. For example, a dictionary encoded String implementation would cast the Objects as int[] arrays. When comparing, if the two arrays have different lengths, the shorter array should be ordered first. Otherwise, the implementer of this function should iterate through the unsorted encoded values, converting them to their actual type (e.g., performing a dictionary lookup for a dict-encoded String dimension), and comparing the actual values until a difference is found. Refer to StringDimensionIndexer.compareUnsortedEncodedKeyComponents() for a reference implementation. The comparison rules used by this method should match the rules used by DimensionHandler.getEncodedValueSelectorComparator(), otherwise incorrect ordering/merging of rows can occur during ingestion, causing issues such as imperfect rollup.

Specified by:

compareUnsortedEncodedKeyComponents in interface DimensionIndexer<StructuredData,StructuredData,StructuredData>

Parameters:

lhs - dimension value array from a Row key

rhs - dimension value array from a Row key

Returns:

comparison of the two arrays

checkUnsortedEncodedKeyComponentsEqual

public boolean checkUnsortedEncodedKeyComponentsEqual(@Nullable
 StructuredData lhs,
 @Nullable
 StructuredData rhs)

Description copied from interface: DimensionIndexer

Check if two row value arrays from Row keys are equal.

Specified by:

checkUnsortedEncodedKeyComponentsEqual in interface DimensionIndexer<StructuredData,StructuredData,StructuredData>

Parameters:

lhs - dimension value array from a Row key

rhs - dimension value array from a Row key

Returns:

true if the two arrays are equal

getUnsortedEncodedKeyComponentHashCode

public int getUnsortedEncodedKeyComponentHashCode(@Nullable
 StructuredData key)

Description copied from interface: DimensionIndexer

Given a row value array from a Row key, generate a hashcode.

Specified by:

getUnsortedEncodedKeyComponentHashCode in interface DimensionIndexer<StructuredData,StructuredData,StructuredData>

Parameters:

key - dimension value array from a Row key

Returns:

hashcode of the array

convertUnsortedEncodedKeyComponentToActualList

public Object convertUnsortedEncodedKeyComponentToActualList(StructuredData key)

Description copied from interface: DimensionIndexer

Given a row value array from a Row key, as described in the documentation for DimensionIndexer.compareUnsortedEncodedKeyComponents(EncodedKeyComponentType, EncodedKeyComponentType), convert the unsorted encoded values to a list of actual values. If the key has one element, this method should return a single Object instead of a list.

Specified by:

convertUnsortedEncodedKeyComponentToActualList in interface DimensionIndexer<StructuredData,StructuredData,StructuredData>

Parameters:

key - dimension value array from a Row key

Returns:

single value or list containing the actual values corresponding to the encoded values in the input array

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<StructuredData,StructuredData,StructuredData>

fillBitmapsFromUnsortedEncodedKeyComponent

public void fillBitmapsFromUnsortedEncodedKeyComponent(StructuredData key,
 int rowNum,
 MutableBitmap[] bitmapIndexes,
 BitmapFactory factory)

Description copied from interface: DimensionIndexer

Helper function for building bitmap indexes for integer-encoded dimensions. Called by IncrementalIndexAdapter as it iterates through its sequence of rows. Given a row value array from a Row key, with the current row number indicated by "rowNum", set the index for "rowNum" in the bitmap index for each value that appears in the row value array. For example, if key is an int[] array with values [1,3,4] for a dictionary-encoded String dimension, and rowNum is 27, this function would set bit 27 in bitmapIndexes[1], bitmapIndexes[3], and bitmapIndexes[4] See StringDimensionIndexer.fillBitmapsFromUnsortedEncodedKeyComponent() for a reference implementation. If a dimension type does not support bitmap indexes, this function will not be called and can be left unimplemented.

Specified by:

fillBitmapsFromUnsortedEncodedKeyComponent in interface DimensionIndexer<StructuredData,StructuredData,StructuredData>

Parameters:

key - dimension value array from a Row key

rowNum - current row number

bitmapIndexes - array of bitmaps, indexed by integer dimension value

factory - bitmap factory