XiaoHongbo-Hope commented on PR #672:
URL: https://github.com/apache/paimon-rust/pull/672#issuecomment-5178038960

   Addressed the mixed-format P1 in a4098c8.
   
   Exactness no longer depends on the current `file.format` writer option. 
Mosaic now widens its physical projection with predicate columns, applies the 
shared exact residual evaluator per batch, and projects back to the requested 
output. Added an end-to-end regression with current `file.format=parquet`, an 
old `.mosaic` split, an unprojected LIKE column, and content assertions (`id = 
[1, 3]`).
   
   Validation: Mosaic 33/33; Paimon table 994 passed / 1 ignored; DataFusion 
329 passed / same 7 missing-fixture tests; clippy passed.
   
   Also reran the production-scale Shanghai query on this exact head: EXPLAIN 
5.625 s, 1 split / 3 files, 1.41 GiB max RSS; SELECT 6.605 s, 1000 rows, all 
1000 LIKE-validated, 1.42 GiB max RSS.
   
   On the earlier P2 suggestions: the limited scan remains sequential to avoid 
speculative shard I/O and multiple decoded bitmaps. Preferred-probe results are 
cleared before capped full fallback because rediscovered duplicate row IDs 
could otherwise consume the per-shard remaining cap before enough new unique 
matches are found; comments now document both constraints.


-- 
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