zjw1111 opened a new issue, #361:
URL: https://github.com/apache/paimon-cpp/issues/361

   ### Search before asking
   
   - [x] I searched existing issues and found no issue specifically tracking 
vector and full-text search in File Index. #173 tracks general File Index 
writing.
   
   ### Motivation
   
   Paimon C++ has File Index infrastructure for per-data-file indexes, but does 
not yet provide an end-to-end File Index search path for vector similarity or 
full-text queries. Applications should be able to build these indexes with 
their data files and use them to select matching file-local rows without 
requiring a Global Index.
   
   ### Solution
   
   Add Vector Search and Full Text Search support to File Index:
   
   - Build and serialize vector and full-text index artifacts through the 
existing File Index write path and container format, and open them through File 
Index readers.
   - Add file-local search visits to `FileIndexReader` and route search 
requests from the read context to the appropriate index reader.
   - Apply returned file-local row positions to the data-file reader. Vector 
Search should return a TopK result independently for each data file and keep 
scores aligned with selected rows; expose scores through an explicitly 
projected `_INDEX_SCORE` column.
   - Support Full Text Search row selection. Score output for Full Text Search 
can be handled separately.
   - Add focused tests for write/read round trips, null or empty indexes, 
multiple data files and batches, score alignment, and columns with more than 
one File Index type.
   
   The initial implementation can target append-only/raw-file reads. Predicate 
combination, deletion-vector and row-range pre-filtering, Merge-on-Read/Data 
Evolution, and cross-file global TopK should be handled in follow-up work so 
that filtering does not silently change TopK semantics.
   
   ### Anything else?
   
   This builds on the File Index writing work in #173; it is specifically about 
vector/full-text index implementations and their search read path.
   


-- 
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