benwtrent opened a new issue, #14266:
URL: https://github.com/apache/lucene/issues/14266
### Description
I noticed a 10x failure for
`TestKnnByteVectorQueryMMap.testRandomWithFilter` and its an interestingly
weird edge case, and exceptionally improbable.
I initially thought this was due to the new filtering search logic that has
been merged, but it isn't. Apparently, for a particular filter & a particular
random byte vector, the graph search actually hits the lowest graph level, and
gets to nearest neighborhood almost immediately, succeeding the approximate
search and never hitting exact search.
here is an immediate reproduction hard-coding the values that succeed the
approximate search.
```
Query filter4 = IntPoint.newRangeQuery("tag", 168, 174);
System.out.println(filter4);
expectThrows(
UnsupportedOperationException.class,
() ->
searcher.search(
getThrowingKnnVectorQuery("field", new float[]{-127f,
30f, -35f, 100f, 0f}, 1, filter4),
numDocs));
```
```
./gradlew test --tests TestKnnByteVectorQueryMMap.testRandomWithFilter
-Dtests.seed=6680B74A421ED4F6
```
I don't know an immediate solution, other than trying more than one vector
until one eventually fails (e.g. search N times to ensure failure).
### Gradle command to reproduce
```
./gradlew test --tests TestKnnByteVectorQueryMMap.testRandomWithFilter
-Dtests.seed=6680B74A421ED4F6
```
--
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: issues-unsubscr...@lucene.apache.org.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
