Kontinuation commented on PR #1021:
URL: 
https://github.com/apache/datafusion-comet/pull/1021#issuecomment-2419719313

   > Thanks for filing this @Kontinuation. I will close my PRs.
   > 
   > Also, there is a suggestion in #1017 for always using unified memory.
   
   I agree that using the unified memory manager is a better approach. Vanilla 
Spark operators and comet operators are governed by the same memory manager and 
they are all using offheap memory. Vanilla Spark operators can free some memory 
when comet operators are under pressure. I'll also put more work into improving 
the unified memory management.
   
   I think the native memory management approach may still be relevant when 
users don't want vanilla Spark to use off-heap memory. We can set the default 
value of `spark.comet.exec.memoryPool` as `greedy_task_shared` to make comet 
memory overhead configuration more intuitive. If users want memory-intensive 
operators to spill properly, they can try out `fair_spill_task_shared`.


-- 
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