cetra3 commented on issue #20715: URL: https://github.com/apache/datafusion/issues/20715#issuecomment-4001475471
> For point 1, what is the difference between the proposed change and the existing mechanism in the fair pool? The fair pool does weird things with partitions. Essentially it uses `num_spill` to sub-divide spillable consumers into their own buckets, but in my testing this wasn't sufficient. The reason I think being that each consumer has a portion of memory that is spillable and non-spillable, but when registering you have to pick one or the other. > For point 2, currently we only have a single watermark. What are the benefits of introducing two watermarks? I can see that this may make operators less likely to wait for memory as mentioned in point 3, but I’m wondering if there are other considerations as well. That's not entirely true. There is the total amount within a pool, but there is also memory reservation bytes which essentially acts as two watermarks, albeit in a roundabout way. -- 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]
