[HIVE-20362] PTF: Invert sorting for CURRENT_ROW to UNBOUNDED FOLLOWING

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: CBO, PTF-Windowing
• Assignee: Unassigned
• Reporter: Gopal Vijayaraghavan
• Created: 10/Aug/18 16:52
• Updated: 10/Aug/18 16:53

Description

COALESCE(current_page, FIRST_VALUE(current_page,TRUE) OVER ( PARTITION BY user_id ORDER BY timestamp ROWS BETWEEN CURRENT ROW and UNBOUNDED FOLLOWING))

This is a spilling PTF because it needs an unseen row to generate the current row when executing this.

This can be inverted, so that the shuffle sorts it the other way, so the result can be generated from a previously seen row.

COALESCE(current_page, LAST_VALUE(current_page,TRUE) OVER ( PARTITION BY user_id ORDER BY timestamp desc ROWS BETWEEN UNBOUNDED PRECEDING and CURRENT ROW))

This query pattern is bad in general, but seems to exist as a workaround for - HIVE-18145