[
https://issues.apache.org/jira/browse/HIVE-15061?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15609998#comment-15609998
]
Chaoyu Tang edited comment on HIVE-15061 at 10/31/16 1:37 PM:
--------------------------------------------------------------
This patch is to fix the following issues:
1. alterTableUpdateTableColumnStats and updatePartColumnStatsForAlterColumns
did not ignore the case when they compare the columns to determine if their
stats should be updated or deleted
2. update_table_column_statistics/update_partition_column_statistics allow the
uppercase column type to be populated to HMS backend DB
was (Author: ctang.ma):
This patch is to fix the following issues:
1. alterTableUpdateTableColumnStats and updatePartColumnStatsForAlterColumns
did not ignore the case when they compare the columns to determine if their
stats should be updated or deleted
2. HMS alterTable allows the uppercase column name and type to be populated to
HMS backend DB
3. update_table_column_statistics/update_partition_column_statistics allow the
uppercase column type to be populated to HMS backend DB
> Metastore types are sometimes case sensitive
> --------------------------------------------
>
> Key: HIVE-15061
> URL: https://issues.apache.org/jira/browse/HIVE-15061
> Project: Hive
> Issue Type: Bug
> Components: API
> Affects Versions: 1.1.0
> Reporter: Thomas Tauber-Marshall
> Assignee: Chaoyu Tang
> Attachments: HIVE-15061.1.patch, HIVE-15061.1.patch, HIVE-15061.patch
>
>
> Impala recently encountered an issue with the metastore
> ([IMPALA-4260|https://issues.cloudera.org/browse/IMPALA-4260] ) where column
> stats would get dropped when adding a column to a table.
> The reason seems to be that Hive does a case sensitive check on the column
> stats types during an "alter table" and expects the types to be all lower
> case. This case sensitive check doesn't appear to happen when the stats are
> set in the first place.
> We're solving this on the Impala end by storing types in the metastore as all
> lower case, but Hive's behavior here is very confusing. It should either
> always be case sensitive, so that you can't create column stats with types
> that Hive considers invalid, or it should never be case sensitive.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
