Rich-T-kid commented on PR #26789:
URL: https://github.com/apache/datafusion/pull/26789#issuecomment-6857256535

   ### Current `main` vs PR head, with RLE dictionary reads disabled
   
   This is the requested merge-candidate comparison:
   
   - baseline: `main` at `4c7b30752c159a85c713453ad7041b9c29e4d647`
   - changed: this PR at `678d97de914fd0e55fbc1f2d5eea6ac67a916dff`
   - both sides: `DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY=false`
   - changed side only: 
`DATAFUSION_OPTIMIZER_ENABLE_DICTIONARY_AGGREGATION=true`
   
   | Suite | Minimum total | Mean total | Raw slower flags |
   |---|---:|---:|---|
   | 
[TPC-H](https://github.com/apache/datafusion/pull/26789#issuecomment-6857015448)
 | 716.26 -> 710.34 ms (-0.83%) | 732.37 -> 725.23 ms (-0.97%) | Q7 minimum 
only |
   | 
[TPC-DS](https://github.com/apache/datafusion/pull/26789#issuecomment-6857053894)
 | 8774.66 -> 8685.77 ms (-1.01%) | 8921.44 -> 8861.27 ms (-0.67%) | Q42, Q81, 
and Q86 mean only |
   | [partitioned 
ClickBench](https://github.com/apache/datafusion/pull/26789#issuecomment-6857130415)
 | 18951.62 -> 19006.22 ms (+0.29%) | 19387.67 -> 19510.02 ms (+0.63%) | 
Q19/Q36 minimum only; Q18/Q26/Q35 mean only |
   
   TPC-DS Q47 and Q57 are the clear repeatable wins: both improve by about 9% 
in the minimum table and 10% in the mean table. No slower query reproduces 
above 5% in both the minimum and mean tables. Q19 even changes direction: its 
minimum is 7% slower while its mean is 43% faster.
   
   The raw output is not literally free of red labels. Q7, Q19, and Q36 are 
minimum-only threshold crossings; Q81/Q86 and the ClickBench mean-only flags 
have large changed-side tail outliers; Q42 is a borderline 5% mean-only 
crossing. So this run is effectively flat outside the intended TPC-DS wins, but 
it does not satisfy a strict "zero bot flags" gate.
   
   There are three commits on current `main` that are not in the PR: 
documentation, an opt-in CSV header feature (these suites use Parquet), and a 
behavior-preserving `EnforceDistribution` refactor. None changes dictionary 
aggregation, Parquet dictionary reads, or the benchmark configuration.
   


-- 
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