Github user hvanhovell commented on the pull request:
https://github.com/apache/spark/pull/5604#issuecomment-97518465
Hi,
I have been experimenting with Window functions in Spark SQL as well. It
has been partially based on this. You can find my work
[here](https://github.com/hvanhovell/spark-window).
I have deviated from the original implementation in couple of ways:
- Implemented it as an extension to Spark SQL and not Hive. All aggregates
use the Spark SQL implementations (not the Hive UDAFs).
- Use of SPARK 1.4 child ordering requirements. Sorting is planned by the
engine; this will especially interesting as soon as exchange will start
supporting secondary sorting. I have tried a few sorting schemes but this one
is currently the fastest.
- Only a single window specification (grouping and ordering) is processed
at a time. The analyzer should take care of multiple window specifications.
- The current implementation is semi-blocking; it processes one group at a
time. This means only the rows for one group per partition are kept in memory.
In the future we should also accommodate the case in which all aggregates are
streaming (perhaps with some buffering).
Shall we try to join forces, and come up with one good PR?
Kind regards,
Herman
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
