kosiew commented on PR #24857:
URL: https://github.com/apache/datafusion/pull/24857#issuecomment-5568920329

   re ungrouped_utf8_view_accumulator_is_never_worse_than_a_pre_allocated_set, 
   could we split this test matrix based on the hash mode?
   
   The 100,000 and 500,000 cardinality cases also run in the `hash collisions 
(amd64)` CI job, which enables `datafusion-common/force_hash_collisions`. In 
that configuration every distinct value hashes to the same bucket, so each 
insertion has to search an increasingly long collision chain. Since both the 
lazy and pre-allocated sets receive every value, these larger cases become 
quadratic and can make the job run for hours or hit its 360-minute timeout, as 
seen in #25011.
   
   I suggest keeping a small, bounded case under forced collisions to verify 
collision correctness, while leaving the large cardinalities for the normal 
hash configuration where they are useful for checking the allocation and 
capacity-convergence invariant. It would also be good to run the bounded 
collision case with the same CI profile and forwarded feature before merging.


-- 
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