hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Marta Kuczora (JIRA)" <j...@apache.org>
Subject [jira] [Resolved] (HIVE-21632) Hive should not push partition columns to the Parquet predicate, even if the data file contains the partition column
Date Thu, 09 May 2019 07:57:00 GMT

     [ https://issues.apache.org/jira/browse/HIVE-21632?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Marta Kuczora resolved HIVE-21632.
----------------------------------
    Resolution: Duplicate

> Hive should not push partition columns to the Parquet predicate, even if the data file
contains the partition column
> --------------------------------------------------------------------------------------------------------------------
>
>                 Key: HIVE-21632
>                 URL: https://issues.apache.org/jira/browse/HIVE-21632
>             Project: Hive
>          Issue Type: Bug
>    Affects Versions: 4.0.0
>            Reporter: Marta Kuczora
>            Priority: Minor
>
> If there is a partitioned Parquet table in Hive, and the data file in one of the partitions
(not correctly) contains the partition column as well, filtering on the partition column will
return no rows if the Parquet predicate pushdown is enabled. If the PPD is disabled, the rows
will return correctly.
> The reason why it doesn't work is that, if the PPD is switched on, Hive will send the
predicate 'partition_column= ...' to parquet and a requested schema which doesn't contain
the partition column. When the data is read from parquet, this column will be skipped, because
the requested schema doesn't contain it, but it still tries to apply the filter predicate,
so it will return an empty result set.
> I think if the rows are returned correctly without PPD, they should be returned with
PPD as well. Hive should omit the partition column from the Parquet predicate.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Mime
View raw message