viirya commented on code in PR #11218:
URL: https://github.com/apache/datafusion/pull/11218#discussion_r1682230886
##########
datafusion/physical-plan/src/joins/sort_merge_join.rs:
##########
@@ -858,6 +889,57 @@ impl SMJStream {
}
}
+ fn free_reservation(&mut self, buffered_batch: BufferedBatch) ->
Result<()> {
+ // Shrink memory usage for in memory batches only
+ if buffered_batch.spill_file.is_none() &&
buffered_batch.batch.is_some() {
+ self.reservation
+ .try_shrink(buffered_batch.size_estimation)?;
+ }
+
+ Ok(())
+ }
+
+ fn allocate_reservation(&mut self, mut buffered_batch: BufferedBatch) ->
Result<()> {
+ match self.reservation.try_grow(buffered_batch.size_estimation) {
+ Ok(_) => {
+ self.join_metrics
+ .peak_mem_used
+ .set_max(self.reservation.size());
+ Ok(())
+ }
+ Err(_) if self.runtime_env.disk_manager.tmp_files_enabled() => {
+ // spill buffered batch to disk
+ let spill_file = self
+ .runtime_env
+ .disk_manager
+ .create_tmp_file("SortMergeJoinBuffered")?;
+
+ if let Some(batch) = buffered_batch.batch {
+ spill_record_batches(
+ vec![batch],
+ spill_file.path().into(),
+ Arc::clone(&self.buffered_schema),
+ )?;
+ buffered_batch.spill_file = Some(spill_file);
+ buffered_batch.batch = None;
+
+ // update metrics to register spill
+ self.join_metrics.spill_count.add(1);
+ self.join_metrics
+ .spilled_bytes
+ .add(buffered_batch.size_estimation);
+
self.join_metrics.spilled_rows.add(buffered_batch.num_rows);
+ }
+
+ Ok(())
+ }
+ Err(e) => Err(e),
Review Comment:
For the returned error, it is better to let user know that
`tmp_files_enabled` is disabled so spilling is not enabled.
--
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]
