[
https://issues.apache.org/jira/browse/SPARK-57823?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Max Gekk updated SPARK-57823:
-----------------------------
Shepherd: Max Gekk
> Support ORC predicate pushdown for nanosecond-precision timestamps
> ------------------------------------------------------------------
>
> Key: SPARK-57823
> URL: https://issues.apache.org/jira/browse/SPARK-57823
> Project: Spark
> Issue Type: Sub-task
> Components: SQL
> Affects Versions: 4.3.0
> Reporter: Max Gekk
> Priority: Major
>
> This sub-task is part of the umbrella SPARK-56822 (timestamps with nanosecond
> precision).
> h2. Problem
> {{OrcFilters.getPredicateLeafType}} / {{castLiteralValue}}
> (datasources/orc/OrcFilters.scala ~L142-179) have no {{AnyTimestampNanoType}}
> arm and fall through to {{throw
> QueryExecutionErrors.unsupportedOperationForDataTypeError(dataType)}}.
> Because {{convertibleFilters}} probes convertibility by building the search
> argument, a filter on a nanosecond column throws when ORC pushdown runs.
> h2. Goal
> Convert predicates on nanosecond timestamp columns to ORC search arguments
> (or safely skip them) instead of throwing.
> h2. Scope
> Add nanosecond arms to {{getPredicateLeafType}} and {{castLiteralValue}},
> using the ORC timestamp representation consistent with the ORC nanosecond
> read/write (SPARK-57455).
> h2. Non-goals
> Footer MIN/MAX aggregate pushdown is NOT included -
> {{AggregatePushDownUtils}} (~L69-87) excludes all timestamp types
> (microsecond and nanosecond alike), so it is not a nanosecond-specific parity
> gap.
> h2. Acceptance criteria
> * Filters on nanosecond ORC columns push down (or skip) without error;
> results correct.
> h2. Testing
> {{OrcFilterSuite}}; ORC nanosecond query tests.
> h2. Dependencies
> None - independent (ORC datasource support resolved in SPARK-57455).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
