NGA-TRAN commented on PR #18521:
URL: https://github.com/apache/datafusion/pull/18521#issuecomment-3506608750

   >>>> @2010YOUY01 : The second one is tpch-q4, I saw a 20% speedup in 
benchmark result, so I checked the query plan, and the difference is it's 
removing a round robin repartition above parquet reader:
   
   >>> @gene-bordegaray : Can you calrify the data you used to create the 
tables, thank you
   
   >> @NGA-TRAN : @2010YOUY01 : The reason Gene asked you to verify your test 
data is 
https://github.com/apache/datafusion/issues/18341#issuecomment-3505130845. It 
seems in the build side your lineitem table is too small to repartition. Thus 
the plan after the fix does not repartition it. That is the reason for the 
speedup
   
   > @gene-bordegaray: Yes, I am suspecting that i is due to the size of the 
tables because the logic for repartitioning at the file level lives above and 
independent of the repartitioning logic changes I have made
   
   That’s likely the case—there’s a predicate on the lineitem table, and we’ve 
pushed down a kind-of bloom filter to prune data during the scan. As a result, 
the dataset could be quite small. I wouldn’t worry about this scenario; it’s a 
reasonable outcome and one of the motivations behind the fix. Plus, it clearly 
improves query performance.


-- 
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