[ 
https://issues.apache.org/jira/browse/HIVE-640?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12737380#action_12737380
 ] 

Zheng Shao commented on HIVE-640:
---------------------------------

@HIVE-640.2.patch:
LazyBinaryByte etc: The copy constructors should copy the value of the data as 
well. LazyBinaryString/LazyBinaryBoolean is already doing that but not others.
It's also OK to remove these copy constructors at all, since they are never 
used.

LazyBinaryArray.java:150: We should get 
((ListObjectInspector)oi).getListElementObjectInspector() and save it to a 
local variable, so we don't need to call the function again and again.

LazyBinaryUtils.java:295: "Returns the lazy binary object inspector that can be 
used to translate an object of that typeInfo to a standard object type. " -> 
"Returns the lazy binary object inspector that can be used to inspect an lazy 
binary object of that typeInfo"

LazyBinaryString.java: The length of the string can be stored in a VInt instead 
of fixed size 4 bytes. This will help us save some space (especially for short 
strings).

LazyBinarySerDe.java: Remove the commented code (Line 464). It seems the only 
difference between serializing the row and serializing a struct is that we 
don't need the total number of bytes at the beginning. Can we add a parameter 
to serialize(Output byteStream,Object obj, ObjectInspector objInspector), and 
let serialize(Object o, OI oi) to call that directly?



> Add LazyBinarySerDe to Hive
> ---------------------------
>
>                 Key: HIVE-640
>                 URL: https://issues.apache.org/jira/browse/HIVE-640
>             Project: Hadoop Hive
>          Issue Type: New Feature
>            Reporter: Zheng Shao
>            Assignee: Yuntao Jia
>         Attachments: HIVE-640.1.patch, HIVE-640.2.patch
>
>
> LazyBinarySerDe will serialize the data in binary format while supporting 
> LazyDeserialization.
> This will be used as the SerDe for value between map and reduce, and also 
> between different map-reduce jobs.
> This will help improve the performance of Hive a lot.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to