[
https://issues.apache.org/jira/browse/SPARK-16628?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15385184#comment-15385184
]
Tejas Patil commented on SPARK-16628:
-------------------------------------
Thanks for notifying [~yhuai]. Is this specific to ORC only ? I remember for
the change I made, the codepath was similar to what Parquet used (and there was
spark.sql.hive.convertMetastoreParquet as well).
> OrcConversions should not convert an ORC table represented by
> MetastoreRelation to HadoopFsRelation if metastore schema does not match
> schema stored in ORC files
> -----------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: SPARK-16628
> URL: https://issues.apache.org/jira/browse/SPARK-16628
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Reporter: Yin Huai
>
> When {{spark.sql.hive.convertMetastoreOrc}} is enabled, we will convert a ORC
> table represented by a MetastoreRelation to HadoopFsRelation that uses
> Spark's OrcFileFormat internally. This conversion aims to make table scanning
> have a better performance since at runtime, the code path to scan
> HadoopFsRelation's performance is better. However, OrcFileFormat's
> implementation is based on the assumption that ORC files store their schema
> with correct column names. However, before Hive 2.0, an ORC table created by
> Hive does not store column name correctly in the ORC files (HIVE-4243). So,
> for this kind of ORC datasets, we cannot really convert the code path.
> Right now, if ORC tables are created by Hive 1.x or 0.x, enabling
> {{spark.sql.hive.convertMetastoreOrc}} will introduce a runtime exception for
> non-partitioned ORC tables and drop the metastore schema for partitioned ORC
> tables.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
