dungba88 commented on code in PR #13285:
URL: https://github.com/apache/lucene/pull/13285#discussion_r1689296027
##########
lucene/core/src/java/org/apache/lucene/search/AbstractVectorSimilarityQuery.java:
##########
@@ -103,16 +114,18 @@ public Explanation explain(LeafReaderContext context, int
doc) throws IOExceptio
public ScorerSupplier scorerSupplier(LeafReaderContext context) throws
IOException {
LeafReader leafReader = context.reader();
Bits liveDocs = leafReader.getLiveDocs();
- final Scorer vectorSimilarityScorer;
+
+ QueryTimeout queryTimeout = searcher.getTimeout();
+ TimeLimitingKnnCollectorManager timeLimitingKnnCollectorManager =
Review Comment:
Can we share this variable for all segments? Such as creating it at
top-level variable in `createWeight`?
##########
lucene/core/src/java/org/apache/lucene/search/AbstractVectorSimilarityQuery.java:
##########
@@ -143,27 +156,23 @@ protected boolean match(int doc) {
}
// Perform an approximate search
- TopDocs results = approximateSearch(context, acceptDocs,
cardinality);
+ TopDocs results =
+ approximateSearch(context, acceptDocs, cardinality,
timeLimitingKnnCollectorManager);
- // If the limit was exhausted
- if (results.totalHits.relation ==
TotalHits.Relation.GREATER_THAN_OR_EQUAL_TO) {
- // Return a lazy-loading iterator
- vectorSimilarityScorer =
- VectorSimilarityScorer.fromAcceptDocs(
- this,
- boost,
- createVectorScorer(context),
- new BitSetIterator(acceptDocs, cardinality),
- resultSimilarity);
- } else if (results.scoreDocs.length == 0) {
- return null;
- } else {
+ if (results.totalHits.relation == TotalHits.Relation.EQUAL_TO
+ // Return partial results only when timeout is met
+ || (queryTimeout != null && queryTimeout.shouldExit())) {
// Return an iterator over the collected results
- vectorSimilarityScorer =
- VectorSimilarityScorer.fromScoreDocs(this, boost,
results.scoreDocs);
+ return VectorSimilarityScorerSupplier.fromScoreDocs(boost,
results.scoreDocs);
+ } else {
+ // Return a lazy-loading iterator
+ return VectorSimilarityScorerSupplier.fromAcceptDocs(
Review Comment:
It seems to be a waste that we can't reuse the results from the approximate
search (I also saw similar behavior in top-k KnnVectorQuery).
Maybe we can pass the partial results to this method, and we don't need to
compute score for those?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
