morrySnow opened a new pull request, #69820:
URL: https://github.com/apache/doris/pull/69820

   ### What problem does this PR solve?
   
   Issue Number: None
   
   Related PR: None
   
   Problem Summary: Current BE profiles put operator counters under 
CommonCounters and CustomCounters, but the FE HBO converter reads only counters 
directly on the operator. HBO consequently learns zero output and join row 
counts. Read both groups using the existing conversion logic while retaining 
support for legacy flat profiles and excluding unrelated descendants. 
Runtime-filter timing counters do not contribute to row estimates; invalid 
counter checks remain in place.
   
   Update the skew regression to collect profiles, bypass SQL cache, use the 
actual slow-query threshold and await asynchronous statistics. Check that the 
result remains 21000200 before and after HBO selects a broadcast join.
   
   ### Release note
   
   Fix incorrect HBO cardinality estimates when runtime profile counters are 
grouped beneath operators.
   
   ### Check List (For Author)
   
   - Test: Unit Test / Regression test / Manual test
       - ASAN FE build through ./build.sh --fe -j32 passed, including 
Checkstyle.
       - ./run-fe-ut.sh --run RuntimeProfileTest,RuntimeProfileMergeTest: 9 
tests passed, covering flat/grouped counters, input/output rows, probe/build 
skew, runtime-filter rows and unrelated child profiles.
       - hbo_skew_shuffle_to_bc_test: output generated by the standard 
regression script, three subsequent verification runs passed, and an isolated 
run passed after restarting the final FE build.
   - Behavior changed: Yes (HBO consumes actual grouped row counters rather 
than zero estimates).
   - Does this need documentation: No
   


-- 
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