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 "Shane Kumpf (JIRA)" <ji...@apache.org> on 2018/04/17 15:44:00 UTC
[jira] [Comment Edited] (YARN-8160) Yarn Service Upgrade: Support
upgrade of service that use docker containers
[ https://issues.apache.org/jira/browse/YARN-8160?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16441045#comment-16441045 ]
Shane Kumpf edited comment on YARN-8160 at 4/17/18 3:43 PM:
------------------------------------------------------------
{quote}It might help to refine relaunch logic and reuse some of the existing code.
{quote}
I think there are parts of Relaunch that could be useful in the upgrade case, but I believe some semantics will clash. The idea behind Relaunch is to reuse the same NodeManager, the container id, previous localized resources, and the working directory. In the Docker case, we go a step further and run a {{docker start}} on the existing container. IIUC, the current upgrade design uses re-init which cleans up the existing container prior to launching with a new CLC. These are quite different in their approaches.
Where I see Relaunch features being beneficial is the reuse of the container id/NM, which may allow for reusing the IP address previously assigned. Reuse of the existing work dir could also be useful in some case, but I expect those cases are limited. However, I'm concerned about Relaunch skipping localization and {{docker start}} won't work. At this point, it seems like re-init might be more appropriate.
[~csingh] - based on your experience here, do you have a high level idea of how this will work in the Docker case?
was (Author: shanekumpf@gmail.com):
{quote}It might help to refine relaunch logic and reuse some of the existing code.
{quote}
I think there are parts of Relaunch that could be useful in the upgrade case, but I believe some semantics will clash. The idea behind Relaunch is tore use the same NodeManager, the container id, localization, and the working directory. In the Docker case, we go a step further and run a {{docker start}} on the existing container. IIUC, the current upgrade design uses re-init which cleans up the existing container prior to launching with a new CLC. These are quite different in their approaches.
Where I see Relaunch features being beneficial is the reuse of the container id/NM, which may allow for reusing the IP address previously assigned. Reuse of the existing work dir could also be useful in some case, but I expect those cases are limited. However, I'm concerned about Relaunch skipping localization and {{docker start}} won't work. At this point, it seems like re-init might be more appropriate.
[~csingh] - based on your experience here, do you have a high level idea of how this will work in the Docker case?
> Yarn Service Upgrade: Support upgrade of service that use docker containers
> ----------------------------------------------------------------------------
>
> Key: YARN-8160
> URL: https://issues.apache.org/jira/browse/YARN-8160
> Project: Hadoop YARN
> Issue Type: Sub-task
> Reporter: Chandni Singh
> Assignee: Chandni Singh
> Priority: Major
>
> Ability to upgrade dockerized yarn native services.
> Ref: YARN-5637
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org