[
https://issues.apache.org/jira/browse/TAJO-1919?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14950034#comment-14950034
]
Hyunsik Choi commented on TAJO-1919:
------------------------------------
According to my survey, there are roughly two modes about this. The important
thing is where the final client endpoint runs.
Spark has two modes for Yarn. They main differences are as follows:
* Client mode: The driver runs in the client process, and the application
master is only used for requesting resources from YARN.
* Cluster mode: the Spark driver runs inside an application master process
which is managed by YARN on the cluster, and the client can go away after
initiating the application.
Since in Spark, the driver program plays a role to start the job, they seem to
choose the two modes. In contrast, Tajo basically runs as server. We cannot
choose the client mode of Spark.
As you already know, Impala uses Llama. It provides a special application
master that reserves and releases resources in order to avoid the bit-by-bit
resource allocation overheads of Yarn.
> (Umbrella) Yarn support as a long-running application
> -----------------------------------------------------
>
> Key: TAJO-1919
> URL: https://issues.apache.org/jira/browse/TAJO-1919
> Project: Tajo
> Issue Type: New Feature
> Components: Yarn
> Reporter: Hyunsik Choi
> Assignee: Hyunsik Choi
> Fix For: 0.12.0
>
>
> Yarn is a widely used cluster management system in Hadoop. We need to support
> Yarn. Roughly, there may be two approaches to support Yarn. The first
> approach is running the system as a long-running application in Yarn. In this
> approach, Tajo runs as a long-running application, and Yarn client launches a
> Tajo cluster and request the cluster resources to RM. The another approach is
> let the scheduler of Tajo to directly communicate RM.
> In this issue, we will let Tajo to support the first approach. This is an
> umbrella issue to track all issues about it.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
