[
https://issues.apache.org/jira/browse/HIVE-22641?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chiran Ravani updated HIVE-22641:
---------------------------------
Description:
In Hive 1.2.1 and 2.0 while displaying columns for a table, it used to return
in same order as it was created. for example
{code}
create table col_order_test(server_name string, task_name string,
partition_name string, start_time string, end_time string, table_owner string,
table_name string) stored as orc;
show columns in col_order_test;
+-----------------+--+
| field |
+-----------------+--+
| server_name |
| task_name |
| partition_name |
| start_time |
| end_time |
| table_owner |
| table_name |
+-----------------+--+
{code}
For Hive 3 columns are returned in sorted order for the same query, below is
output.
{code}
create table col_order_test(server_name string, task_name string,
partition_name string, start_time string, end_time string, table_owner string,
table_name string) stored as orc;
show columns in col_order_test;
+-----------------+
| field |
+-----------------+
| end_time |
| partition_name |
| server_name |
| start_time |
| table_name |
| table_owner |
| task_name |
+-----------------+
{code}
Above behavior looks to be changed with the introduction of search column
feature as part of Jira [HIVE-18373
|https://issues.apache.org/jira/browse/HIVE-18373]
This behavior change can cause an existing process to fail in a few
environments, for example, code to generate the INSERT OVERWRITE in a different
manner, which may result in query failure.
I would like to request a community if we can improve the Jira [HIVE-18373
|https://issues.apache.org/jira/browse/HIVE-18373] by returning column order
same as it was created if the search pattern provided by the user is null.
Attaching patch with the change.
was:
In Hive 1.2.1 and 2.0 while displaying columns for a table, it used to return
in same order as it was created. for example
{code}
create table col_order_test(server_name string, task_name string,
partition_name string, start_time string, end_time string, table_owner string,
table_name string) stored as orc;
show columns in col_order_test;
+-----------------+--+
| field |
+-----------------+--+
| server_name |
| task_name |
| partition_name |
| start_time |
| end_time |
| table_owner |
| table_name |
+-----------------+--+
{code}
For Hive 3 columns are returned in sorted order for the same query, below is
output.
{code}
create table col_order_test(server_name string, task_name string,
partition_name string, start_time string, end_time string, table_owner string,
table_name string) stored as orc;
show columns in col_order_test;
+-----------------+
| field |
+-----------------+
| end_time |
| partition_name |
| server_name |
| start_time |
| table_name |
| table_owner |
| task_name |
+-----------------+
{code}
Above behavior looks to be changed with the introduction of search column
feature as part of Jira [HIVE-18373
|https://issues.apache.org/jira/browse/HIVE-18373]
This behavior change can cause existing process to fail setup in a few
environments, for example: code to generate the INSERT OVERWRITE in a different
manner, which may result in query failure.
I would like to request a community if we can improve the Jira [HIVE-18373
|https://issues.apache.org/jira/browse/HIVE-18373] by returning column order
same as it was created if the search pattern provided by the user is null.
Attaching patch with the change.
> Columns returned in sorted order when show columns query is run with no
> search pattern.
> ---------------------------------------------------------------------------------------
>
> Key: HIVE-22641
> URL: https://issues.apache.org/jira/browse/HIVE-22641
> Project: Hive
> Issue Type: Improvement
> Components: Hive, HiveServer2
> Affects Versions: 3.1.0
> Reporter: Chiran Ravani
> Assignee: Chiran Ravani
> Priority: Major
> Attachments: HIVE-22641.patch
>
>
> In Hive 1.2.1 and 2.0 while displaying columns for a table, it used to return
> in same order as it was created. for example
> {code}
> create table col_order_test(server_name string, task_name string,
> partition_name string, start_time string, end_time string, table_owner
> string, table_name string) stored as orc;
> show columns in col_order_test;
> +-----------------+--+
> | field |
> +-----------------+--+
> | server_name |
> | task_name |
> | partition_name |
> | start_time |
> | end_time |
> | table_owner |
> | table_name |
> +-----------------+--+
> {code}
> For Hive 3 columns are returned in sorted order for the same query, below is
> output.
> {code}
> create table col_order_test(server_name string, task_name string,
> partition_name string, start_time string, end_time string, table_owner
> string, table_name string) stored as orc;
> show columns in col_order_test;
> +-----------------+
> | field |
> +-----------------+
> | end_time |
> | partition_name |
> | server_name |
> | start_time |
> | table_name |
> | table_owner |
> | task_name |
> +-----------------+
> {code}
> Above behavior looks to be changed with the introduction of search column
> feature as part of Jira [HIVE-18373
> |https://issues.apache.org/jira/browse/HIVE-18373]
> This behavior change can cause an existing process to fail in a few
> environments, for example, code to generate the INSERT OVERWRITE in a
> different manner, which may result in query failure.
> I would like to request a community if we can improve the Jira [HIVE-18373
> |https://issues.apache.org/jira/browse/HIVE-18373] by returning column order
> same as it was created if the search pattern provided by the user is null.
> Attaching patch with the change.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
