[
https://issues.apache.org/jira/browse/TEZ-4689?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18061103#comment-18061103
]
Raghav Aggarwal commented on TEZ-4689:
--------------------------------------
[~abstractdog] , yes you are correct. They should be optional. If we want to
use S3 then they are not even required.
> Introduce Node abstraction for DAGAppMaster instead of separate
> NodeManager-related fields
> ------------------------------------------------------------------------------------------
>
> Key: TEZ-4689
> URL: https://issues.apache.org/jira/browse/TEZ-4689
> Project: Apache Tez
> Issue Type: Sub-task
> Reporter: László Bodor
> Priority: Major
>
> In DAGAppMaster.main I can see:
> {code}
> String nodeHostString =
> System.getenv(ApplicationConstants.Environment.NM_HOST.name());
> String nodePortString =
> System.getenv(ApplicationConstants.Environment.NM_PORT.name());
> String nodeHttpPortString =
> System.getenv(ApplicationConstants.Environment.NM_HTTP_PORT.name());
> {code}
> this is problematic in different ways:
> 1. we're currently decoupling Tez from Yarn, we should abstract these and
> lazy-resolve them when they are really needed
> 2. they pollute the DAGAppMaster constructor: we should use a single
> `NodeContext`, which maybe can implemented by `YarnNameNodeContext` while
> running on Yarn
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
