Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
Description
The current implementation does not redistribute, which would tend to result in a lot of extra files. Specifically, the number of files will be larger by a factor equal to the number of fragments in the final stage of the query. On even a moderately sized cluster, this number could easily be in the thousands, so a table with a 100 different partitions would end up with hundreds of thousands of files.
To allow a workaround for this situation, we should add an option to include an extra distribution, so that all the rows for any given partition are written from the same writer.