Max Gekk created SPARK-57785:
--------------------------------
Summary: Support window RANGE frames with offset bounds over a
TIME ORDER BY column
Key: SPARK-57785
URL: https://issues.apache.org/jira/browse/SPARK-57785
Project: Spark
Issue Type: Sub-task
Components: SQL
Affects Versions: 4.3.0
Reporter: Max Gekk
h2. Background
Umbrella: SPARK-57550 (Extend support for the TIME data type), SPIP SPARK-51162.
ANSI SQL permits window frames of the form
{{RANGE BETWEEN <offset> PRECEDING AND <offset> FOLLOWING}} where the {{ORDER
BY}} column
is a datetime type and the offset is an interval. For DATE/TIMESTAMP, Spark
supports
value-based RANGE frames with day-time / year-month interval offsets.
h2. Problem
It is not verified whether a {{TimeType}} {{ORDER BY}} column is supported in
value-based
RANGE frames (where the offset would be a day-time interval, e.g.
{{RANGE BETWEEN INTERVAL '1' HOUR PRECEDING AND CURRENT ROW}}). The expected
behavior
and supported offset type for TIME are undefined, so this may throw an unclear
error or
behave inconsistently with DATE/TIMESTAMP.
h2. Scope
* Determine the expected semantics for value-based RANGE frames with a TIME
order column
(offset type, wrap-around behavior at the 00:00 / 23:59:59.999999999
boundaries).
* Support TIME as the RANGE-frame ordering column with a day-time interval
offset, matching
DATE/TIMESTAMP behavior; or, if intentionally unsupported, raise a clear,
typed error
and document the limitation.
* Add window-function tests covering TIME RANGE frames (positive and/or
clear-error cases).
h2. Acceptance
* {{RANGE BETWEEN <interval> PRECEDING/FOLLOWING}} over a TIME {{ORDER BY}}
column either
works on par with DATE/TIMESTAMP or fails with a clear, documented error --
verified by
tests.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
