Github user scwf commented on the issue:
https://github.com/apache/spark/pull/16633
Not get you, but let me explain more,
If we use map output statistics to decide each global limit should take how
many element.
1. local limit shuffle with the maillist partitioner and return the map
output statistics
2. global limit each partition take or drop some rows(just like what you
do in this pr) based on the statistics
Then,
1. the shuffle cost is almost the same as now
2. global limit without single partition issue when a big limit number
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
