xudong963 commented on code in PR #16956:
URL: https://github.com/apache/datafusion/pull/16956#discussion_r2254228293
##########
datafusion/physical-plan/src/joins/hash_join.rs:
##########
@@ -892,21 +892,56 @@ impl ExecutionPlan for HashJoinExec {
}
fn partition_statistics(&self, partition: Option<usize>) ->
Result<Statistics> {
- if partition.is_some() {
- return Ok(Statistics::new_unknown(&self.schema()));
+ match (partition, self.mode) {
+ // For CollectLeft mode, the left side is collected into a single
partition,
+ // so all left partitions are available to each output partition.
+ // For the right side, we need the specific partition statistics.
+ (Some(partition), PartitionMode::CollectLeft) => {
+ let left_stats = self.left.partition_statistics(None)?;
+ let right_stats =
self.right.partition_statistics(Some(partition))?;
+
+ let stats = estimate_join_statistics(
+ left_stats,
+ right_stats,
+ self.on.clone(),
+ &self.join_type,
+ &self.join_schema,
+ )?;
+ Ok(stats.project(self.projection.as_ref()))
+ }
+
+ // For Partitioned mode, both sides are partitioned, so each
output partition
+ // only has access to the corresponding partition from both sides.
+ (Some(partition), PartitionMode::Partitioned) => {
+ let left_stats =
self.left.partition_statistics(Some(partition))?;
+ let right_stats =
self.right.partition_statistics(Some(partition))?;
+
+ let stats = estimate_join_statistics(
+ left_stats,
+ right_stats,
+ self.on.clone(),
+ &self.join_type,
+ &self.join_schema,
+ )?;
+ Ok(stats.project(self.projection.as_ref()))
+ }
+
+ // For Auto mode or when no specific partition is requested, fall
back to
Review Comment:
For auto mode, as the comment says:
```
/// DataFusion optimizer decides which PartitionMode
/// mode(Partitioned/CollectLeft) is optimal based on statistics. It will
/// also consider swapping the left and right inputs for the Join
Auto,
```
So if the method with partition is called after `JoinSelection` rule, it's
impossible to see Auto mode, however, the method may be called first, I suggest
evaluating the cost of related code about choosing partition mode in
`JoinSelection`, then decide if we can decide the PartitionMode here.
--
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
