alamb commented on code in PR #12471:
URL: https://github.com/apache/datafusion/pull/12471#discussion_r1771831141
##########
datafusion/core/src/datasource/physical_plan/parquet/mod.rs:
##########
@@ -741,7 +741,18 @@ impl ExecutionPlan for ParquetExec {
}
fn statistics(&self) -> Result<Statistics> {
- Ok(self.projected_statistics.clone())
+ // When filters are pushed down, we have no way of knowing the exact
statistics.
+ // Note that pruning predicate is also a kind of filter pushdown.
+ // (bloom filters use `pruning_predicate` too)
+ let stats = if self.pruning_predicate.is_some()
+ || self.page_pruning_predicate.is_some()
+ || (self.predicate.is_some() && self.pushdown_filters())
+ {
+ self.projected_statistics.clone().to_inexact()
Review Comment:
I thought that inexact was the least difference than we had
In terms of actually doing cardinality estimation of filters, I think that
is its own very hard challenge -- there is some vestigal stuff in DataFusion,
but if we wanted to make it better I think we should add some very simple
implementation in DataFusion with an API to extend if/when needed
--
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]
