airborne12 opened a new pull request, #70151:
URL: https://github.com/apache/doris/pull/70151

   ### What problem does this PR solve?
   
   Issue Number: None
   
   Related PR: #69790
   
   Problem Summary: inverted_index_p0/test_match_keyword_score, added by 
#69790, has
   failed in P0 since it merged, always on the VARIANT part of the first loop 
pass:
   
       Check tag 'V2_variant_sparse_path' failed, line 1, DOUBLE result 
mismatch.
       Expect cell is: 0.980829
       But real is   : 0
   
   (once in cloud_p0: V2_variant_text_path, 0.693147 vs 0).
   
   The suite expects every VARIANT path (n, s, p) to be a materialized subcolumn
   with its own inverted index. CI runs with use_fuzzy_session_variable, and
   SessionVariable.initFuzzyModeVariables sets 
default_variant_max_subcolumns_count
   to random.nextInt(10) on each new connection (0 when it picks doc mode). The 
suite
   already pinned default_variant_enable_typed_paths_to_sparse and
   default_variant_enable_doc_mode, but not this limit. With a limit of 1, the
   writer keeps only the top path of each segment (s, by the path tie-break) and
   moves n and p to the sparse column. The sparse column has no inverted index, 
so
   MATCH on those paths runs without the index and scores 0.
   
   The numeric path loses its score as well (row 4: 0.287682 -> 0), but
   OutputUtils accepts that mismatch: when the cells have different numbers of
   decimal places, it rounds 0.287682 to "0" places, which gives 0. So the suite
   fails only at the sparse path, whose score is 0.98.
   
   Reproduced on a local cluster: with SET GLOBAL 
default_variant_max_subcolumns_count = 1,
   the unchanged suite fails with exactly the CI message. Limits 0 and 2-9 pass.
   
   Fix: set default_variant_max_subcolumns_count = 0 (no limit) next to the 
other
   two pinned VARIANT defaults. The expected output does not change.
   
   ### Release note
   
   None
   
   ### Check List (For Author)
   
   - Test: Regression test
       - test_match_keyword_score on a local cluster: before the fix it fails 
with the
         CI message under a forced limit of 1; after the fix it passes under 
that
         limit and with the default.
   - Behavior changed: No
   - Does this need documentation: No
   
   
   🤖 Generated with [Claude Code](https://claude.com/claude-code)
   


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to