You are viewing a plain text version of this content. The canonical link for it is here.
Posted to yarn-issues@hadoop.apache.org by "Karthik Kambatla (JIRA)" <ji...@apache.org> on 2014/07/03 07:35:54 UTC

[jira] [Updated] (YARN-2227) Move containerMgrProxy from RM's AMLaunch to get rid of issues that new client talking with old server

     [ https://issues.apache.org/jira/browse/YARN-2227?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Karthik Kambatla updated YARN-2227:
-----------------------------------

    Target Version/s: 2.6.0  (was: 2.5.0)

> Move containerMgrProxy from RM's AMLaunch to get rid of issues that new client talking with old server
> ------------------------------------------------------------------------------------------------------
>
>                 Key: YARN-2227
>                 URL: https://issues.apache.org/jira/browse/YARN-2227
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: resourcemanager
>            Reporter: Junping Du
>            Assignee: Junping Du
>
> In rolling upgrade semantics, we should handle cases that old client should talk with new servers if only compatible changes happen in RPC protocol. In this semantics, there is no guarantee that new client should able to talk with old server which need us to pay specially attention on upgrading sequence. Even this, we will find that it is still hard to deal with NM talk with RM as there are both client and server at both side: in regular heartbeat, NM is client and RM is server; when RM launch AM client, it go through containerMgrProxy and RM is client while NM is server in this case. We should get rid of this situation, i.e. by removing containerMgrProxy in RM and use other way to launch container.



--
This message was sent by Atlassian JIRA
(v6.2#6252)