[
https://issues.apache.org/jira/browse/HIVE-29525?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18073782#comment-18073782
]
Thomas Rebele commented on HIVE-29525:
--------------------------------------
In HIVE-29432 I also encountered the issue that partitioned Iceberg tables do
not provide info on the partitioning. At that time I just excluded Iceberg
tables. In case HIVE-29525 gets implemented, it might be possible to improve
HIVE-29432 as well.
> Iceberg partitioned columns are not reflected by RelOptHiveTable
> ----------------------------------------------------------------
>
> Key: HIVE-29525
> URL: https://issues.apache.org/jira/browse/HIVE-29525
> Project: Hive
> Issue Type: Bug
> Components: CBO
> Reporter: Stamatis Zampetakis
> Priority: Major
>
> Iceberg partitioned columns are not reflected by RelOptHiveTable and the
> respective data-structures (hivePartitionCols and hivePartitionColsMap) are
> always empty. RelOptHiveTable is used in various places during the CBO
> planning phase so any logic that depends on partitioning information is not
> kicking in and all Iceberg tables are treated as unpartitioned.
> Depending on the context where the partitioning information is used different
> issues may arise. For instance, selectivity estimation is computed
> differently if the predicate is over a partitioning column (see
> [FilterSelectivityEstimator|https://github.com/apache/hive/blob/931d4bb62b26de699240c816df439e00644e3dcb/ql/src/java/org/apache/hadoop/hive/ql/optimizer/calcite/stats/FilterSelectivityEstimator.java#L105])
> so the absence of partitioning information can have a direct impact on
> cardinality (row count) estimations and affect join order.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
