Package org.apache.cassandra.index.sai.plan

Class VectorTopKProcessor

java.lang.Object

org.apache.cassandra.index.sai.plan.VectorTopKProcessor

public class VectorTopKProcessor
extends Object

Processor that scans all rows from given partitions and selects rows with top-k scores based on vector indexes.

This processor performs the following steps: - collect rows with score into PriorityQueue that sorts rows based on score. If there are multiple vector indexes, the final score is the sum of all vector index scores. - remove rows with the lowest scores from PQ if PQ size exceeds limit - return rows from PQ in primary key order to client

Constructor Summary

Constructors

Constructor	Description
VectorTopKProcessor(ReadCommand command)

Method Summary

Modifier and Type	Method	Description
<U extends Unfiltered, R extends BaseRowIterator<U>, P extends BasePartitionIterator<R>> BasePartitionIterator<?>	consumeSortByScoreAndTakeTopK(P partitions)	Filter given partitions and keep the rows with the highest scores.
<U extends Unfiltered, R extends BaseRowIterator<U>, P extends BasePartitionIterator<R>> BasePartitionIterator<?>	takeTopKThenSortByPrimaryKey(P partitions)	Filter given partitions and keep the rows with the highest scores.

Methods inherited from class java.lang.Object

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

Constructor Details

VectorTopKProcessor

public VectorTopKProcessor(ReadCommand command)

Method Details

consumeSortByScoreAndTakeTopK

public <U extends Unfiltered,
R extends BaseRowIterator<U>,
P extends BasePartitionIterator<R>>
BasePartitionIterator<?> consumeSortByScoreAndTakeTopK(P partitions)

Filter given partitions and keep the rows with the highest scores. In case of UnfilteredPartitionIterator, all tombstones will be kept.

takeTopKThenSortByPrimaryKey

public <U extends Unfiltered,
R extends BaseRowIterator<U>,
P extends BasePartitionIterator<R>>
BasePartitionIterator<?> takeTopKThenSortByPrimaryKey(P partitions)

Filter given partitions and keep the rows with the highest scores. In case of UnfilteredPartitionIterator, all tombstones will be kept.