[
https://issues.apache.org/jira/browse/HIVE-5304?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sergey Shelukhin updated HIVE-5304:
-----------------------------------
Summary: Hive results can depend on metastore's underlying datastore (was:
Hive results can depend on metastore's underlying datastore, if autocreate is
used)
> Hive results can depend on metastore's underlying datastore
> -----------------------------------------------------------
>
> Key: HIVE-5304
> URL: https://issues.apache.org/jira/browse/HIVE-5304
> Project: Hive
> Issue Type: Bug
> Components: Metastore
> Reporter: Sergey Shelukhin
>
> [removed old description]
> Hive JDOQL filter pushdown and direct SQL may end up pushing StringCol op
> 'SomeString' to underlying SQL datastore. However, the datastore may handle
> these differently based on the encoding and collation used for the columns of
> the database.
> So, query results can change depending on the underlying store for the
> metastore, and OS used
> I am assuming that byte-order sort if the correct way to order things.
> Our MySQL script specifies _bin collation, and Postgres, as far as I see,
> defaults to "C"; both of those are byte-order collations.
> However, MySQL by default doesn't use _bin collation, so if database is
> auto-created, the order of things is going to change.
> Derby also uses the non-byte-order by
> drop_partitions_filter.q illustrates this problem. In byte order collation
> (proper way) USA is sorted before Uganda
--
This message was sent by Atlassian JIRA
(v6.1#6144)
