[
https://issues.apache.org/jira/browse/SPARK-54203?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Max Gekk updated SPARK-54203:
-----------------------------
Description:
h2. What
Add a {{TimeType}} branch to {{RowToColumnConverter}} so rows containing TIME
columns can be
converted to columnar (Arrow/columnar batch) format.
h2. Why (confirmed needed)
{{RowToColumnConverter.getConverterForType}} in
{{sql/core/.../execution/Columnar.scala}} handles {{LongType | TimestampType |
TimestampNTZType | DayTimeIntervalType}} (all long-backed) but has no
{{TimeType}} case, so
a TIME column falls through to {{unsupportedDataTypeError}}. This blocks
row-to-column
conversion paths (e.g. {{ColumnarToRow}}/{{RowToColumnar}} transitions,
vectorized
execution) for TIME. (Supersedes the original "same change as TimestampNTZ in
commit
47276ab9902" note - the change is required, not speculative.)
h2. Scope
* Add the {{TimeType}} case to {{getConverterForType}} using the long-backed
({{LongConverter}}) path, since TIME is stored as nanos-of-day {{Long}}.
* Confirm the on/off-heap column vectors already round-trip TIME (they do -
TIME is in the
{{longData}} group).
h2. Acceptance criteria
* A DataFrame with a TIME column survives a row<->column round-trip (e.g.
cached columnar,
or a plan that inserts {{RowToColumnar}}).
* Test added to {{ColumnarBatchSuite}} / the Columnar conversion tests.
was:
Make the same change for the Time type that was made for TimestampNTZ (in
commit [47276ab9902|https://github.com/apache/spark/commit/47276ab9902]).
I can't say whether it is needed for the Time type or not, but it was made for
other types (including TimestampNTZ).
Summary: Support TimeType in RowToColumnConverter (was: Support Time
in RowToColumnConverter)
> Support TimeType in RowToColumnConverter
> ----------------------------------------
>
> Key: SPARK-54203
> URL: https://issues.apache.org/jira/browse/SPARK-54203
> Project: Spark
> Issue Type: Sub-task
> Components: SQL
> Affects Versions: 4.1.0
> Reporter: Bruce Robbins
> Priority: Major
>
> h2. What
> Add a {{TimeType}} branch to {{RowToColumnConverter}} so rows containing TIME
> columns can be
> converted to columnar (Arrow/columnar batch) format.
> h2. Why (confirmed needed)
> {{RowToColumnConverter.getConverterForType}} in
> {{sql/core/.../execution/Columnar.scala}} handles {{LongType | TimestampType |
> TimestampNTZType | DayTimeIntervalType}} (all long-backed) but has no
> {{TimeType}} case, so
> a TIME column falls through to {{unsupportedDataTypeError}}. This blocks
> row-to-column
> conversion paths (e.g. {{ColumnarToRow}}/{{RowToColumnar}} transitions,
> vectorized
> execution) for TIME. (Supersedes the original "same change as TimestampNTZ in
> commit
> 47276ab9902" note - the change is required, not speculative.)
> h2. Scope
> * Add the {{TimeType}} case to {{getConverterForType}} using the long-backed
> ({{LongConverter}}) path, since TIME is stored as nanos-of-day {{Long}}.
> * Confirm the on/off-heap column vectors already round-trip TIME (they do -
> TIME is in the
> {{longData}} group).
> h2. Acceptance criteria
> * A DataFrame with a TIME column survives a row<->column round-trip (e.g.
> cached columnar,
> or a plan that inserts {{RowToColumnar}}).
> * Test added to {{ColumnarBatchSuite}} / the Columnar conversion 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]
