[
https://issues.apache.org/jira/browse/HIVE-8719?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14197791#comment-14197791
]
Sushanth Sowmyan commented on HIVE-8719:
----------------------------------------
If it matters, this can be tested using the HCat E2E test framework, with the
following test conf addition
{noformat}
{
+ 'name' => 'LoadData',
+ 'tests' => [ {
+ 'num' => 1,
+ 'sql' => "drop table if exists load_8719;
+ dfs -mkdir /:HDFSTMP:/load_8719;
+ create external table load_8719 (name string) partitioned
by (date string) location '/:HDFSTMP:/load_8719';
+ insert overwrite local directory ':OUTPATH:' select name
from studenttab10k;
+ alter table load_8719 add partition(date='2014-09-16')
location ':HDFSTMP:/load_8719/20140916';
+ load data local inpath ':OUTPATH:' into table load_8719
partition(date='2014-09-16');
+ describe formatted load_8719
partition(date='2014-09-16');",
+ 'cmd_fail' => 1, # Isn't really expected to fail, but we want to
prevent benchmark generation because we aren't interested in it.
+ 'expected_out_regex' => "Location:.*/load_8719/20140916",
+ }
{noformat}
I will take on a task on another bug to try and extend the current .q test
framework to be able to test this, and file a jira for that.
> LoadSemanticAnalyzer ignores previous partition location if inserting into
> partition that already exists
> --------------------------------------------------------------------------------------------------------
>
> Key: HIVE-8719
> URL: https://issues.apache.org/jira/browse/HIVE-8719
> Project: Hive
> Issue Type: Bug
> Components: Query Processor
> Affects Versions: 0.14.0
> Reporter: Sushanth Sowmyan
> Assignee: Sushanth Sowmyan
> Attachments: HIVE-8719.patch
>
>
> LOAD DATA INSERT INTO seems to be broken currently for partitions that do not
> use hive's native directory structure naming scheme, thus ignoring any
> location previously set by an ALTER TABLE ADD PARTITION ... LOCATION ...
> command.
> Here is a simple reproducer:
> {noformat}
> echo 1 > /tmp/data1.txt
> hive -e "create external table testpart(id int) partitioned by (date string)
> location '/tmp/testpart';"
> hive -e "alter table testpart add partition(date='2014-09-16') location
> '/tmp/testpart/20140916';"
> hive -e "describe formatted testpart partition(date='2014-09-16') ;" | egrep
> '/tmp/testpart/(date=.?)?2014-?09-?16' > /tmp/a
> hive -e "load data local inpath '/tmp/data1.txt' into table testpart
> partition(date='2014-09-16');"
> hive -e "describe formatted testpart partition(date='2014-09-16') ;" | egrep
> '/tmp/testpart/(date=.?)?2014-?09-?16' > /tmp/b
> diff /tmp/a /tmp/b
> hadoop fs -ls /tmp/testpart/
> {noformat}
> Basically, what happens is that after the ALTER TABLE ADD PARTITION ...
> LOCATION, the location is "/tmp/testpart/20140916". After the LOAD DATA has
> run, the partition location becomes "/tmp/testpart/date=2014-09-16/". Any
> data previously present in the other location will then be ignored as well.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
