FelixYBW commented on issue #11397:
URL:
https://github.com/apache/incubator-gluten/issues/11397#issuecomment-3791921574
So if we follow Spark's way, we need to remove rule
RemoveNativeWriteFilesSortAndProject, then in valox we write partition one by
one. Close previous writer when a new writer is open.
There are 3 solutions here:
- Follow Spark's way
- Manually repartition data before write like
`df.repartition(tbl_filenum[tbl]['part_cnt'],
tbl_filenum[tbl]['part']).write.mode("overwrite").format("parquet").option("compression",
"zstd").partitionBy(tbl_filenum[tbl]['part']).option('fs.s3a.committer.name',
'magic').save(f"s3a://presto-workload/{databasename}/{tbl}")`. Each partition
write a parquet file. It's the ideal case but if data is biased, we wll get a
large partition as below:
<img width="546" height="116" alt="Image"
src="https://github.com/user-attachments/assets/cb48b8c0-3e7c-4035-9a81-bd902875db03";
/>
- Add a `RebalanceByColumn(part columns)` before writer as @wForget shared
in https://github.com/apache/incubator-gluten/issues/7656. It's something
similar to add a `repartition(partitionnum, tbl_filenum[tbl]['part'])`
automatically
--
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]
