suremarc commented on code in PR #15473:
URL: https://github.com/apache/datafusion/pull/15473#discussion_r2020364348
##########
datafusion/datasource/src/file_scan_config.rs:
##########
@@ -575,6 +575,95 @@ impl FileScanConfig {
})
}
+ /// Splits file groups into new groups based on statistics to enable
efficient parallel processing.
+ ///
+ /// The method distributes files across a target number of partitions
while ensuring
+ /// files within each partition maintain sort order based on their min/max
statistics.
+ ///
+ /// The algorithm works by:
+ /// 1. Sorting all files by their minimum values
+ /// 2. Trying to place each file into an existing group where it can
maintain sort order
+ /// 3. Creating new groups when necessary if a file cannot fit into
existing groups
+ /// 4. Prioritizing smaller groups when multiple suitable groups exist
(for load balancing)
+ ///
+ /// # Parameters
+ /// * `table_schema`: Schema containing information about the columns
+ /// * `file_groups`: The original file groups to split
+ /// * `sort_order`: The lexicographical ordering to maintain within each
group
+ /// * `target_partitions`: The desired number of output partitions
+ ///
+ /// # Returns
+ /// A new set of file groups, where files within each group are
non-overlapping with respect to
+ /// their min/max statistics and maintain the specified sort order.
+ pub fn split_groups_by_statistics_v2(
Review Comment:
Perhaps we could call it `split_groups_by_statistics_with_target_partitions`?
TBH I am not sure if anyone is using the old code, so I would wager it is
safe to replace with the new implementation. But I agree the old one is
probably more useful in certain scenarios, e.g. if you are doing a sort merge
above it.
If we were to keep it, I would rather unify the implementations, the only
thing that differs is the policy for selecting the group to insert. I think we
could probably abstract that out into an enum or generic parameter. (Not really
sure how common generics are in datafusion though)
--
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
