[
https://issues.apache.org/jira/browse/KYLIN-875?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14627798#comment-14627798
]
liyang commented on KYLIN-875:
------------------------------
The initial design is to think the Engine as a mainboard, to define the Input
and Output interface it wants. Then Table Source and Storage will adapt to the
Input/Output spec of the mainboard and be able to plug into the Engine.
For example MapReduce as an Engine defines MRInput and MROutput interfaces.
Hive as a Table Source will implement MRInput in order to plug into MREngine.
Similarly HBase will implement MROutput.
Later a SparkEngine may define SparkInput and SparkOutput interfaces. To work
with Spark, Hive and HBase will implement the SparkInput and SparkOutput as
well.
> Decouple with Hadoop to allow alternative Input / Build Engine / Storage
> ------------------------------------------------------------------------
>
> Key: KYLIN-875
> URL: https://issues.apache.org/jira/browse/KYLIN-875
> Project: Kylin
> Issue Type: Improvement
> Affects Versions: v0.8.1
> Reporter: liyang
> Assignee: liyang
>
> We've got many requests about how to use an input source other than hive; or
> how to use Spark instead of Hadoop MR; or how to use Cassandra in place of
> HBase.
> In order to decouple with Hadoop, interfaces must be defined for Input Source
> / Build Engine / Storage respectively. And let hive / MR / hbase become a
> piece of plugin of implementation.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
