datafusion-comet
datafusion-comet copied to clipboard
feat: Use fair-spill pool when `spark.memory.offHeap.enabled=false`
Which issue does this PR close?
Closes https://github.com/apache/datafusion-comet/issues/996
Possibly depends on https://github.com/apache/datafusion-comet/pull/988 being resolved first
Rationale for this change
This is an alternative to the approach in https://github.com/apache/datafusion-comet/pull/1002
What changes are included in this PR?
- Take task concurrency into account when allocating memory per native plan
- Use fair spool pool instead of greedy pool