[HIVE-10555] Improve windowing spec of range based windowing to support additional range formats

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.3.0
• Fix Version/s: 1.3.0, 2.0.0
• Component/s: PTF-Windowing
• Assignee: Aihua Xu
• Reporter: Aihua Xu
• Created: 30/Apr/15 15:27
• Updated: 16/Feb/16 23:50
• Resolved: 02/Jun/15 19:05

Description

Currently windowing function only supports the formats of x preceding and current, x preceding and y following, current and y following.

Windowing of x preceding and y preceding and x following and y following doesn't work properly.

The following functions should be supported.
First_value(), last_value(), sum(), avg() , count(), min(), max()

Issue Links

• is duplicated by: HIVE-9412 Window clause ROW BETWEEN for PRECEDING does not work
• requires: HIVE-10140 Window boundary is not compared correctly

Sub-Tasks

• Refactoring Windowing for sum() to pass WindowFrameDef instead of two numbers (1 for number of preceding and 1 for number of following): Closed
• Improve sum() function over windowing to support additional range formats: Resolved
• COUNT(*) over windowing 'x preceding and y preceding' doesn't work properly: Closed
• Support min()/max() functions over x preceding and y preceding windowing: Closed
• Support First_value()/last_value() over x preceding and y preceding windowing: Closed
• Add tests to cover avg() function for 'x preceding and y preceding' windowing spec.: Closed
• windowing function returns incorrect result when the window size is larger than the partition size: Closed
• Similar to HIVE-12574, collect_set and count() give incorrect result when partition size is smaller than window size: Resolved