Interface BufferAggregator
-
- All Superinterfaces:
HotLoopCallee
- All Known Implementing Classes:
CardinalityBufferAggregator
,CountBufferAggregator
,DelegatingBufferAggregator
,DoubleAnyBufferAggregator
,DoubleFirstBufferAggregator
,DoubleLastBufferAggregator
,DoubleMaxBufferAggregator
,DoubleMeanBufferAggregator
,DoubleMinBufferAggregator
,DoubleSumBufferAggregator
,ExpressionLambdaBufferAggregator
,FilteredBufferAggregator
,FloatAnyBufferAggregator
,FloatFirstBufferAggregator
,FloatLastBufferAggregator
,FloatMaxBufferAggregator
,FloatMinBufferAggregator
,FloatSumBufferAggregator
,HistogramBufferAggregator
,HyperUniquesBufferAggregator
,JavaScriptBufferAggregator
,LongAnyBufferAggregator
,LongConstantBufferAggregator
,LongFirstBufferAggregator
,LongLastBufferAggregator
,LongMaxBufferAggregator
,LongMinBufferAggregator
,LongSumBufferAggregator
,NoopBufferAggregator
,NullableNumericBufferAggregator
,NumericAnyBufferAggregator
,NumericFirstBufferAggregator
,NumericLastBufferAggregator
,SimpleDoubleBufferAggregator
,SimpleFloatBufferAggregator
,SimpleLongBufferAggregator
,StringAnyBufferAggregator
,StringColumnDoubleBufferAggregatorWrapper
,StringColumnFloatBufferAggregatorWrapper
,StringColumnLongBufferAggregatorWrapper
,StringFirstBufferAggregator
,StringLastBufferAggregator
,SuppressedAggregatorFactory.SuppressedBufferAggregator
public interface BufferAggregator extends HotLoopCallee
A BufferAggregator is an object that can aggregate metrics into a ByteBuffer. Its aggregation-related methods (namely, aggregate(...) and get(...)) only take the ByteBuffer and position because it is assumed that the Aggregator was given something (one or more MetricSelector(s)) in its constructor that it can use to get at the next bit of data. Thus, an Aggregator can be thought of as a closure over some other thing that is stateful and changes between calls to aggregate(...). UnlikeAggregator
, BufferAggregators are never used by multiple threads at once. Implementations are not required to be thread safe.
-
-
Method Summary
All Methods Instance Methods Abstract Methods Default Methods Modifier and Type Method Description void
aggregate(ByteBuffer buf, int position)
Aggregates metric values into the given aggregate byte representation Implementations of this method must read in the aggregate value from the buffer at the given position, aggregate the next element of data and write the updated aggregate value back into the buffer.void
close()
Release any resources used by the aggregatorObject
get(ByteBuffer buf, int position)
Returns the intermediate object representation of the given aggregate.default double
getDouble(ByteBuffer buf, int position)
Returns the double representation of the given aggregate byte array Converts the given byte buffer representation into the intermediate aggregate value.float
getFloat(ByteBuffer buf, int position)
Returns the float representation of the given aggregate byte array Converts the given byte buffer representation into the intermediate aggregate value.long
getLong(ByteBuffer buf, int position)
Returns the long representation of the given aggregate byte array Converts the given byte buffer representation into the intermediate aggregate value.void
init(ByteBuffer buf, int position)
Initializes the buffer location Implementations of this method must initialize the byte buffer at the given position Implementations must not change the position, limit or mark of the given buffer This method must not exceed the number of bytes returned byAggregatorFactory.getMaxIntermediateSizeWithNulls()
in the correspondingAggregatorFactory
default void
inspectRuntimeShape(RuntimeShapeInspector inspector)
Implementations of this method should callinspector.visit()
with all fields of this class, which meet two conditions: 1.default boolean
isNull(ByteBuffer buf, int position)
returns true if aggregator's output type is primitive long/double/float and aggregated value is null, but when aggregated output type is Object, this method always returns false, and users are advised to check nullability for the object returned byget(ByteBuffer, int)
method.default void
relocate(int oldPosition, int newPosition, ByteBuffer oldBuffer, ByteBuffer newBuffer)
Relocates any cached objects.
-
-
-
Method Detail
-
init
@CalledFromHotLoop void init(ByteBuffer buf, int position)
Initializes the buffer location Implementations of this method must initialize the byte buffer at the given position Implementations must not change the position, limit or mark of the given buffer This method must not exceed the number of bytes returned byAggregatorFactory.getMaxIntermediateSizeWithNulls()
in the correspondingAggregatorFactory
- Parameters:
buf
- byte buffer to initializeposition
- offset within the byte buffer for initialization
-
aggregate
@CalledFromHotLoop void aggregate(ByteBuffer buf, int position)
Aggregates metric values into the given aggregate byte representation Implementations of this method must read in the aggregate value from the buffer at the given position, aggregate the next element of data and write the updated aggregate value back into the buffer. Implementations must not change the position, limit or mark of the given buffer- Parameters:
buf
- byte buffer storing the byte array representation of the aggregateposition
- offset within the byte buffer at which the current aggregate value is stored
-
get
@Nullable Object get(ByteBuffer buf, int position)
Returns the intermediate object representation of the given aggregate. Converts the given byte buffer representation into an intermediate aggregate Object Implementations must not change the position, limit or mark of the given buffer. The object returned must not have any references to the given buffer (i.e., make a copy), since the underlying buffer is a shared resource and may be given to another processing thread while the objects returned by this aggregator are still in use. If the correspondingAggregatorFactory.combine(Object, Object)
method for this aggregator expects its inputs to be mutable, then the object returned by this method must be mutable.- Parameters:
buf
- byte buffer storing the byte array representation of the aggregateposition
- offset within the byte buffer at which the aggregate value is stored- Returns:
- the Object representation of the aggregate
-
getFloat
float getFloat(ByteBuffer buf, int position)
Returns the float representation of the given aggregate byte array Converts the given byte buffer representation into the intermediate aggregate value. Implementations must not change the position, limit or mark of the given buffer Implementations are only required to support this method if they are aggregations which have anAggregatorFactory.getIntermediateType()
()} ofValueType.FLOAT
. If unimplemented, throwing anUnsupportedOperationException
is common and recommended.- Parameters:
buf
- byte buffer storing the byte array representation of the aggregateposition
- offset within the byte buffer at which the aggregate value is stored- Returns:
- the float representation of the aggregate
-
getLong
long getLong(ByteBuffer buf, int position)
Returns the long representation of the given aggregate byte array Converts the given byte buffer representation into the intermediate aggregate value. Implementations must not change the position, limit or mark of the given buffer Implementations are only required to support this method if they are aggregations which have anAggregatorFactory.getIntermediateType()
of ofValueType.LONG
. If unimplemented, throwing anUnsupportedOperationException
is common and recommended.- Parameters:
buf
- byte buffer storing the byte array representation of the aggregateposition
- offset within the byte buffer at which the aggregate value is stored- Returns:
- the long representation of the aggregate
-
getDouble
default double getDouble(ByteBuffer buf, int position)
Returns the double representation of the given aggregate byte array Converts the given byte buffer representation into the intermediate aggregate value. Implementations must not change the position, limit or mark of the given buffer Implementations are only required to support this method if they are aggregations which have anAggregatorFactory.getIntermediateType()
of ofValueType.DOUBLE
. If unimplemented, throwing anUnsupportedOperationException
is common and recommended. The default implementation castsgetFloat(ByteBuffer, int)
to double. This default method is added to enable smooth backward compatibility, please re-implement it if your aggregators work with numeric double columns.- Parameters:
buf
- byte buffer storing the byte array representation of the aggregateposition
- offset within the byte buffer at which the aggregate value is stored- Returns:
- the double representation of the aggregate
-
close
void close()
Release any resources used by the aggregator
-
inspectRuntimeShape
default void inspectRuntimeShape(RuntimeShapeInspector inspector)
Implementations of this method should callinspector.visit()
with all fields of this class, which meet two conditions: 1. They are used in methods of this class, annotated withCalledFromHotLoop
2. They are either: a. Nullable objects b. Instances of HotLoopCallee c. Objects, which don't always have a specific class in runtime. For example, a field of typeSet
could beHashSet
orTreeSet
in runtime, depending on how this instance (the instance on which inspectRuntimeShape() is called) is configured. d. ByteBuffer or similar objects, where byte order matters e. boolean flags, affecting branch taking f. Arrays of objects, meeting any of conditions a-e.The default implementation inspects nothing. Classes that implement
BufferAggregator
are encouraged to override this method, following the specification ofHotLoopCallee.inspectRuntimeShape(org.apache.druid.query.monomorphicprocessing.RuntimeShapeInspector)
.- Specified by:
inspectRuntimeShape
in interfaceHotLoopCallee
-
relocate
default void relocate(int oldPosition, int newPosition, ByteBuffer oldBuffer, ByteBuffer newBuffer)
Relocates any cached objects. If underlying ByteBuffer used for aggregation buffer relocates to a new ByteBuffer, positional caches(if any) built on top of old ByteBuffer can not be used for furtheraggregate(ByteBuffer, int)
calls. This method tells the BufferAggregator that the cached objects at a certain location has been relocated to a different location. Only used if there is any positional caches/objects in the BufferAggregator implementation. If relocate happens to be across multiple new ByteBuffers (say n ByteBuffers), this method should be called multiple times(n times) given all the new positions/old positions should exist in newBuffer/OldBuffer. Implementations must not change the position, limit or mark of the given buffer- Parameters:
oldPosition
- old position of a cached object before aggregation buffer relocates to a new ByteBuffer.newPosition
- new position of a cached object after aggregation buffer relocates to a new ByteBuffer.oldBuffer
- old aggregation buffer.newBuffer
- new aggregation buffer.
-
isNull
default boolean isNull(ByteBuffer buf, int position)
returns true if aggregator's output type is primitive long/double/float and aggregated value is null, but when aggregated output type is Object, this method always returns false, and users are advised to check nullability for the object returned byget(ByteBuffer, int)
method. The default implementation always return false to enable smooth backward compatibility, re-implement if your aggregator is nullable.- Parameters:
buf
- byte buffer storing the byte array representation of the aggregateposition
- offset within the byte buffer at which the aggregate value is stored- Returns:
- true if the aggregated value is primitive long/double/float and aggregated value is null otherwise false.
-
-