You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hive.apache.org by "Jason Dere (JIRA)" <ji...@apache.org> on 2017/09/26 21:08:00 UTC
[jira] [Updated] (HIVE-17611) Add new LazyBinary SerDe for faster
writes
[ https://issues.apache.org/jira/browse/HIVE-17611?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Jason Dere updated HIVE-17611:
------------------------------
Attachment: HIVE-17611.1.patch
Initial patch.
Avoid the ObjectInspector getCategory()/getPrimitiveCategory() calls for every row/column by making those calls during initialization and creating serialization objects for each column. Then serialization of the row is just a matter of calling serialize() on these pre-initialized objects.
> Add new LazyBinary SerDe for faster writes
> ------------------------------------------
>
> Key: HIVE-17611
> URL: https://issues.apache.org/jira/browse/HIVE-17611
> Project: Hive
> Issue Type: Improvement
> Components: Serializers/Deserializers
> Reporter: Jason Dere
> Assignee: Jason Dere
> Attachments: HIVE-17611.1.patch
>
>
> LazyBinarySerDe.serialize() ends up making getCategory()/getPrimitiveCategory() calls for every column of every row. Tried some simple tests to eliminate these calls for the non-vectorized version, this looks like it speeds up the writes by ~3x.
> Adding a LazyBinarySerDe2 class with this new implementation.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)