[
https://issues.apache.org/jira/browse/HIVE-15366?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15799472#comment-15799472
]
Sushanth Sowmyan commented on HIVE-15366:
-----------------------------------------
I left some review comments on review board. Basically, it comes down to 2
things:
a) We should not expose JSONMessageFactory to ReplSemanticAnalyzer, and we
shouldn't depend on knowledge of field names. Instead, each message should be
self-serving, and provide a usable api for users such as ReplSemanticAnalyzer,
and in turn, the concrete implementations of the JSON\*Messages can refer to
JSONMessageFactory's generic methods to procide accessors. This means that
JSONMessageFactory should not know or care that InsertMessage has a "table"
field or a "files" field - that knowledge should be limited to
JSONInsertMessage. See the ALTER patch on how CREATE_TABLE/ADD_PTN was
refactored for a code example.
b) I don't think exposing getChecksums() from InsertMessage is a good idea - I
would prefer to see that be self-contained inside the getFiles().
> REPL LOAD & DUMP support for incremental INSERT events
> ------------------------------------------------------
>
> Key: HIVE-15366
> URL: https://issues.apache.org/jira/browse/HIVE-15366
> Project: Hive
> Issue Type: Sub-task
> Components: repl
> Reporter: Vaibhav Gumashta
> Assignee: Vaibhav Gumashta
> Attachments: HIVE-15366.1.patch, HIVE-15366.2.patch
>
>
> Follow up of HIVE-15294. Once we capture the metadata, we should implement
> the LOAD/DUMP part
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
