org.apache.druid.query.groupby.epinephelinae.column

Interface GroupByColumnSelectorStrategy

All Superinterfaces:

ColumnSelectorStrategy

All Known Implementing Classes:

ArrayDoubleGroupByColumnSelectorStrategy, ArrayLongGroupByColumnSelectorStrategy, ArrayNumericGroupByColumnSelectorStrategy, ArrayStringGroupByColumnSelectorStrategy, DictionaryBuildingStringGroupByColumnSelectorStrategy, DoubleGroupByColumnSelectorStrategy, FloatGroupByColumnSelectorStrategy, LongGroupByColumnSelectorStrategy, NullableNumericGroupByColumnSelectorStrategy, StringGroupByColumnSelectorStrategy

public interface GroupByColumnSelectorStrategy
extends ColumnSelectorStrategy

Contains a collection of query processing methods for type-specific operations used exclusively by GroupByQueryEngineV2. Each GroupByColumnSelectorStrategy is associated with a single dimension. Strategies may have internal state, such as the dictionary maintained by DictionaryBuildingStringGroupByColumnSelectorStrategy. Callers should assume that the internal state footprint starts out empty (zero bytes) and is also reset to zero on each call to reset(). Each call to initColumnValues(org.apache.druid.segment.ColumnValueSelector, int, java.lang.Object[]) or writeToKeyBuffer(int, ColumnValueSelector, ByteBuffer) returns the incremental increase in internal state footprint that happened as a result of that particular call.

See Also:

the vectorized version

Field Summary

Fields

Modifier and Type	Field and Description
static int	GROUP_BY_MISSING_VALUE

Method Summary

Modifier and Type	Method and Description
Grouper.BufferComparator	bufferComparator(int keyBufferPosition, StringComparator stringComparator) Return BufferComparator for values written using this strategy when limit is pushed down to segment scan.
boolean	checkRowIndexAndAddValueToGroupingKey(int keyBufferPosition, Object rowObj, int rowValIdx, ByteBuffer keyBuffer) If rowValIdx is less than the size of rowObj (haven't handled all of the row values): First, read the value at rowValIdx from a rowObj and write that value to the keyBuffer at keyBufferPosition.
int	getGroupingKeySize() Return the size, in bytes, of this dimension's values in the grouping key.
int	initColumnValues(ColumnValueSelector selector, int columnIndex, Object[] valuess) Retrieve a row object from the ColumnValueSelector and put it in valuess at columnIndex.
void	initGroupingKeyColumnValue(int keyBufferPosition, int dimensionIndex, Object rowObj, ByteBuffer keyBuffer, int[] stack) Read the first value within a row values object (e.
void	processValueFromGroupingKey(GroupByColumnSelectorPlus selectorPlus, ByteBuffer key, ResultRow resultRow, int keyBufferPosition) Read a value from a grouping key and add it to the group by query result row, using the output name specified in a DimensionSpec.
void	reset() Reset any internal state held by this selector.
int	writeToKeyBuffer(int keyBufferPosition, ColumnValueSelector selector, ByteBuffer keyBuffer) Write a single object from the given selector to the keyBuffer at keyBufferPosition.

Field Detail

GROUP_BY_MISSING_VALUE

static final int GROUP_BY_MISSING_VALUE

See Also:

Constant Field Values

Method Detail

getGroupingKeySize

int getGroupingKeySize()

Return the size, in bytes, of this dimension's values in the grouping key. For example, a String implementation would return 4, the size of an int.

Returns:

size, in bytes, of this dimension's values in the grouping key.

processValueFromGroupingKey

void processValueFromGroupingKey(GroupByColumnSelectorPlus selectorPlus,
                                 ByteBuffer key,
                                 ResultRow resultRow,
                                 int keyBufferPosition)

Read a value from a grouping key and add it to the group by query result row, using the output name specified in a DimensionSpec. An implementation may choose to not add anything to the result row (e.g., as the String implementation does for empty rows) selectorPlus provides access to: - the keyBufferPosition offset from which to read the value - the dimension value selector - the DimensionSpec for this dimension from the query

Parameters:

selectorPlus - dimension info containing the key offset, value selector, and dimension spec

resultRow - result row for the group by query being served

key - grouping key

keyBufferPosition - buffer position for the grouping key, added to support chaining multiple ColumnSelectorStrategy

initColumnValues

int initColumnValues(ColumnValueSelector selector,
                     int columnIndex,
                     Object[] valuess)

Retrieve a row object from the ColumnValueSelector and put it in valuess at columnIndex.

Parameters:

selector - Value selector for a column.

columnIndex - Index of the column within the row values array

valuess - Row values array, one index per column

Returns:

estimated increase in internal state footprint, in bytes, as a result of this operation. May be zero if memory did not increase as a result of this operation. Will not be negative.

initGroupingKeyColumnValue

void initGroupingKeyColumnValue(int keyBufferPosition,
                                int dimensionIndex,
                                Object rowObj,
                                ByteBuffer keyBuffer,
                                int[] stack)

Read the first value within a row values object (e. g. IndexedInts, as the value in a dictionary-encoded string column) and write that value to the keyBuffer at keyBufferPosition. If the row size is 0 (e. g. IndexedInts.size()), write GROUP_BY_MISSING_VALUE instead. If the size of the row is > 0, write 1 to stack[] at columnIndex, otherwise write 0.

Parameters:

keyBufferPosition - Starting offset for this column's value within the grouping key.

dimensionIndex - Index of this dimension within the stack array

rowObj - Row value object for this column

keyBuffer - grouping key

stack - array containing the current within-row value index for each column

checkRowIndexAndAddValueToGroupingKey

boolean checkRowIndexAndAddValueToGroupingKey(int keyBufferPosition,
                                              Object rowObj,
                                              int rowValIdx,
                                              ByteBuffer keyBuffer)

If rowValIdx is less than the size of rowObj (haven't handled all of the row values): First, read the value at rowValIdx from a rowObj and write that value to the keyBuffer at keyBufferPosition. Then return true Otherwise, return false.

Parameters:

keyBufferPosition - Starting offset for this column's value within the grouping key.

rowObj - Row value object for this column (e.g., IndexedInts)

rowValIdx - Index of the current value being grouped on within the row

keyBuffer - grouping key

Returns:

true if rowValIdx < size of rowObj, false otherwise

writeToKeyBuffer

int writeToKeyBuffer(int keyBufferPosition,
                     ColumnValueSelector selector,
                     ByteBuffer keyBuffer)

Write a single object from the given selector to the keyBuffer at keyBufferPosition. The reading column must have a single value. The position of the keyBuffer may be modified.

Parameters:

keyBufferPosition - starting offset for this column's value within the grouping key

selector - selector to retrieve row value object from

keyBuffer - grouping key

Returns:

estimated increase in internal state footprint, in bytes, as a result of this operation. May be zero if memory did not increase as a result of this operation. Will not be negative.

bufferComparator

Grouper.BufferComparator bufferComparator(int keyBufferPosition,
                                          @Nullable
                                          StringComparator stringComparator)

Return BufferComparator for values written using this strategy when limit is pushed down to segment scan.

Parameters:

keyBufferPosition - starting offset for this column's value within the grouping key

stringComparator - stringComparator from LimitSpec for this column. If this is null, implementations will use the StringComparators.LEXICOGRAPHIC comparator.

Returns:

BufferComparator for comparing values written

reset

void reset()

Reset any internal state held by this selector. After this method is called, any row objects or key objects generated by any methods of this class must be considered unreadable. Calling processValueFromGroupingKey(org.apache.druid.query.groupby.epinephelinae.column.GroupByColumnSelectorPlus, java.nio.ByteBuffer, org.apache.druid.query.groupby.ResultRow, int) on that memory has undefined behavior.