[
https://issues.apache.org/jira/browse/HIVE-30110?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated HIVE-30110:
----------------------------------
Labels: correctness pull-request-available (was: correctness)
> Iceberg: ORC filters rows inside a vectorized batch, misaligning row
> positions
> -------------------------------------------------------------------------------
>
> Key: HIVE-30110
> URL: https://issues.apache.org/jira/browse/HIVE-30110
> Project: Hive
> Issue Type: Bug
> Components: Iceberg integration
> Reporter: Denys Kuzmenko
> Priority: Major
> Labels: correctness, pull-request-available
>
> Trigger: orc.sarg.to.filter=true
> ORC drops rows inside a batch that fail the pushed-down filter and packs the
> remaining rows together. Batch row i is then no longer file row batchStart +
> i.
> Everything that assumes it is then points at the wrong rows: position
> deletes, ROW__POSITION and row lineage.
>
> Turning off orc.filter.use.selected isn't enough; sarg.to.filter alone shifts
> positions.
> *Fix* is one line in HiveVectorizedReader: force orc.sarg.to.filter off on
> the Iceberg vectorized ORC path, next to the existing ORC overrides
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
