tzphh opened a new pull request, #10982:
URL: https://github.com/apache/paimon/pull/10982

   ### Purpose
   
   Primary-key vector queries in a batch can select overlapping rows from the 
same data file. This change groups their physical positions by partition, 
bucket, and file, reads the union once per file, and scatters the rows back 
into the original query slots. It preserves projections, empty results, 
deletion-vector filtering, and per-query limits.
   
   Both local and Ray `search_vectors(...).to_arrow()` use the shared lookup on 
the driver. Queries with post-read filters, query authorization, native reads, 
or unsupported position mappings use individual lookup.
   
   Fixes #10772.
   
   Depends on #10828. The branch includes that PR and current master. The 
additional lookup implementation and regression tests are isolated in 
[5080cc2aeb](https://github.com/apache/paimon/commit/5080cc2aebbc18d76817ebd4ed13061f54160e5a).
   
   ### Tests
   
   - Directed Python regression suite covering PK local/Ray vector search, 
batch lookup, Java golden files, snapshot consistency, search metadata, and Ray 
execution — 240 passed, 4 skipped.
   - Overlapping-result regression opens one data-file reader for three 
nonempty query results. Coverage includes repeated queries, disjoint positions, 
deletion vectors, projection variants, empty slots, per-query limits, post-read 
filters, and identical filenames in different buckets with distinct contents.
   - Flake8 with the repository configuration and `git diff --check` passed.
   


-- 
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]

Reply via email to