2010YOUY01 commented on code in PR #15700:
URL: https://github.com/apache/datafusion/pull/15700#discussion_r2046041614
##########
datafusion/physical-plan/src/sorts/sort.rs:
##########
@@ -431,12 +422,16 @@ impl ExternalSorter {
let batches_to_spill = std::mem::take(globally_sorted_batches);
self.reservation.free();
- let in_progress_file =
self.in_progress_spill_file.as_mut().ok_or_else(|| {
- internal_datafusion_err!("In-progress spill file should be
initialized")
- })?;
+ let (in_progress_file, max_record_batch_size) =
+ self.in_progress_spill_file.as_mut().ok_or_else(|| {
+ internal_datafusion_err!("In-progress spill file should be
initialized")
+ })?;
for batch in batches_to_spill {
in_progress_file.append_batch(&batch)?;
+
+ *max_record_batch_size =
+ (*max_record_batch_size).max(batch.get_actually_used_size());
Review Comment:
I think we should not estimate, even if it's correct 99% of the time, IMO
it's impossible to make sure it's always accurate for nested type's reader
implementation. If the estimate is way off for edge cases, the bug would be
hard to investigate.
If we want to follow this optimistic approach, the only required memory
accounting I think is during buffering batches inside `SortExec`, and all the
remaining memory-tracking code can be deleted to make the implementation much
more simpler, the potential problem is unexpected behavior for non-primitive
types (e.g. dictionary array's row format size can explode)
--
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: github-unsubscr...@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org
For additional commands, e-mail: github-h...@datafusion.apache.org
