[ 
https://issues.apache.org/jira/browse/HIVE-3833?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13559572#comment-13559572
 ] 

Namit Jain commented on HIVE-3833:
----------------------------------

bq. In case of identity converter, there is no conversion cost, but in case of 
non-identity this will be worse than current impl, since converter will examine 
every single column value, which wasn't the case earlier. However, it's not 
clear how expensive this would be?

For the above, it is fairly difficult to address. In a follow-up, I can add a 
serde level property, which indicates that the serde can handle different 
datatypes (for eg.
lazySimpleSerde) - if all the partitions of the table have serde's with this 
property, then we can use identityConverter. This is kind of hacky, and am not 
sure if it is
useful, since it should not be a common case. Usually, the partition schema 
should match the table schema.
                
> object inspectors should be initialized based on partition metadata
> -------------------------------------------------------------------
>
>                 Key: HIVE-3833
>                 URL: https://issues.apache.org/jira/browse/HIVE-3833
>             Project: Hive
>          Issue Type: Improvement
>          Components: Query Processor
>            Reporter: Namit Jain
>            Assignee: Namit Jain
>         Attachments: hive.3833.10.patch, hive.3833.11.patch, 
> hive.3833.12.patch, hive.3833.13.patch, hive.3833.14.patch, 
> hive.3833.16.path, hive.3833.17.patch, hive.3833.18.patch, 
> hive.3833.19.patch, hive.3833.1.patch, hive.3833.20.patch, hive.3833.2.patch, 
> hive.3833.3.patch, hive.3833.4.patch, hive.3833.5.patch, hive.3833.6.patch, 
> hive.3833.7.patch, hive.3833.8.patch, hive.3833.9.patch
>
>
> Currently, different partitions can be picked up for the same input split 
> based on the
> serdes' etc. And, we dont allow to change the schema for 
> LazyColumnarBinarySerDe.
> Instead of that, different partitions should be part of the same split, only 
> if the
> partition schemas exactly match. The operator tree object inspectors should 
> be based
> on the partition schema. That would give greater flexibility and also help 
> using binary serde with rcfile

--
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

Reply via email to