avantgardnerio opened a new issue, #7191:
URL: https://github.com/apache/arrow-datafusion/issues/7191

   ### Is your feature request related to a problem or challenge?
   
   Currently, there is only one Aggregation: `GroupedHashAggregateStream`. It 
does a lovely job, but it allocates memory for every unique `group by` value. 
   
   For large datasets, this can cause OOM errors, even if the very next 
operation is a `sort by max(x) limit y`.
   
   ### Describe the solution you'd like
   
   I would like to add a `GroupedAggregateStream` based on a `PriorityQueue` of 
grouped values that can be used instead of `GroupedHashAggregateStream` under 
the specific conditions above, so that Top K queries work even on datasets with 
cardinality larger than available memory.
   
   ### Describe alternatives you've considered
   
   A more generalized implementation where we:
   
   1. sort by group_val
   2. aggregate by group_val `emit`ing rows in a stream as the aggregate for 
each group is computed
   3. feed that into a (new) generalized `TopKExec` node that is _only_ 
responsible for doing the top K operation
   
   Unfortunately, despite being more general, I'm told that this approach will 
still OOM in our case.
   
   ### Additional context
   
   Please see the following similar (but not same) tickets for related top K 
issues:
   
   1. https://github.com/apache/arrow-datafusion/issues/7149
   2. https://github.com/apache/arrow-datafusion/issues/6937
   3. https://github.com/apache/arrow-datafusion/issues/7064
   4. https://github.com/apache/arrow-datafusion/issues/6899
   


-- 
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]

Reply via email to