hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Karthik Kambatla (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (YARN-986) YARN should use cluster-id as token service address
Date Thu, 27 Feb 2014 00:53:28 GMT

    [ https://issues.apache.org/jira/browse/YARN-986?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13913822#comment-13913822

Karthik Kambatla commented on YARN-986:

Thanks for the review, Vinod. Uploaded a patch addressing most of your comments. 

bq. getRMDelegationTokenService() API doesn't belong in this class
Moved it to YarnClient - that seemed the most reasonable place for this.

bq. In the same API, you are only building the service against one address RM_ADDRESS. That
should be qualified by RM-ID?
If you are referring to the non-HA case, we don't need to set RM-ID as it is used. In the
HA case, we do set the RM-ID.

bq. Essentially the new API is what everyone should use to be able to work with RM fail-over.
Why not just deprecate the older API? And change all the callers?
As we discussed offline, the old API is for all other kinds of tokens. IIUC, only the RMDelegationTokens
are affected by the failover. Do you think any other tokens require augmenting their service
names to capture both RM's service addresses? 

bq. We need to be sure distributed-shell and other apps also work fine in this new world,
beyond just MR.
No one else seems to be using RMDelegationTokens, at least in the Hadoop code base. Apps outside
of Hadoop can continue doing what they do to work against non-HA RM. If they want to run against
HA RM, the onus is on them to update. Should we release note this in 2.4? 

> YARN should use cluster-id as token service address
> ---------------------------------------------------
>                 Key: YARN-986
>                 URL: https://issues.apache.org/jira/browse/YARN-986
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>            Reporter: Vinod Kumar Vavilapalli
>            Assignee: Karthik Kambatla
>            Priority: Blocker
>         Attachments: yarn-986-1.patch, yarn-986-2.patch, yarn-986-prelim-0.patch
> This needs to be done to support non-ip based fail over of RM. Once the server sets the
token service address to be this generic ClusterId/ServiceId, clients can translate it to
appropriate final IP and then be able to select tokens via TokenSelectors.
> Some workarounds for other related issues were put in place at YARN-945.

This message was sent by Atlassian JIRA

View raw message