Sergey Shelukhin created HIVE-5304:
--------------------------------------
Summary: JDO and SQL filters can both return different results for
string compares depending on underlying datastore
Key: HIVE-5304
URL: https://issues.apache.org/jira/browse/HIVE-5304
Project: Hive
Issue Type: Bug
Components: Metastore
Reporter: Sergey Shelukhin
Hive uses JDOQL filters to optimize partition retrieval; recently direct SQL
was added to optimize it further. Both of these methods may end up pushing
StringCol op 'SomeString' to underlying SQL datastore. Many paths also pushes
order by-s, although these are not as problematic.
The problem is that different datastores handle string compares differently.
While testing on Postgres, I see that results in different things, from
innocent like order changes in "show partitions", to more serious like
{code}
alter table ptestfilter drop partition (c>='US', d<='2')
{code}
in drop_partitions_filter.q - in Derby, with which the .q.out file was
generated, it drops "c=Uganda/d=2"; this also passes on MySQL (I ran tests with
autocreated db); on Postgres with a db from the script it doesn't.
Looks like we need to enforce collation in partition names and
part_key_values-es; both in the create scripts, as well as during autocreate
(via package.jdo?)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
