shatejas commented on PR #14076:
URL: https://github.com/apache/lucene/pull/14076#issuecomment-2695220393

   > For the exact case that was tested in 
https://github.com/apache/lucene/pull/13985 that might be a regression. I am 
challenging the result a bit here since I don't see how the copy time of the 
vectors could be comparable to the merge time when HNSW indexing is involved.
   
   @jimczi I am in line with the overall approach you are suggesting where the 
prefetch should only be done while fetching vectors and not the overall merge. 
    
   I do think some APIs like updateReadAdvice and finishMerge are helpful, I 
would want to see if we want to keep those and have a noop for this use case. 
this api was added in [Lucene 
10.1](https://github.com/apache/lucene/blob/branch_10_1/lucene/core/src/java/org/apache/lucene/store/IndexInput.java#L237)
   
   I do think prefetch solution will be a best effort to manage both merge and 
search - I can take a stab at it eventually 


-- 
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

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

Reply via email to