[GitHub] [lucene] jpountz commented on pull request #12050: Reuse HNSW graph for intialization during merge

["jpountz (via GitHub)" <gi...@apache.org>]

jpountz commented on PR #12050:
URL: https://github.com/apache/lucene/pull/12050#issuecomment-1425810218

   Nightlies have failed for the last couple days, complaining that KNN searches now return different hits. Is it expected that given the exact same indexing conditions (flushing on doc count and serial merge scheduler), KNN searches may return different hits for the same query with this change?
   
   Here's the error I'm seeing in the log for reference (can be retrieved via `curl -r -10000 http://people.apache.org/~mikemccand/lucenebench/nightly.log`):
   
   ```
   RuntimeError: search result differences: ["query=KnnFloatVectorQuery:vector[0.024077624,...][100] filter=None sort=None groupField=None hitCount=100: hit 6 has wrong field/score value ([19995955], '0.82841617') vs ([19404640], '0.8304943')", "query=KnnFloatVectorQuery:vector[0.028473025,...][100] filter=None sort=None groupField=None hitCount=100: hit 1 has wrong field/score value ([2139705], '0.9640273') vs ([20795785], '0.9655802')", "query=KnnFloatVectorQuery:vector[0.02227773,...][100] filter=None sort=None groupField=None hitCount=100: hit 19 has wrong field/score value ([20249582], '0.9433427') vs ([8538823], '0.94324553')", "query=KnnFloatVectorQuery:vector[-0.047548626,...][100] filter=None sort=None groupField=None hitCount=100: hit 0 has wrong field/score value ([24831434], '0.84341675') vs ([20712471], '0.8335463')", "query=KnnFloatVectorQuery:vector[0.02625591,...][100] filter=None sort=None groupField=None hitCount=100: hit 6 has wrong field/score value ([25459412], '
 0.8309758') vs ([15548210], '0.8312737')"]
   ```


-- 
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-unsubscribe@lucene.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@lucene.apache.org
For additional commands, e-mail: issues-help@lucene.apache.org