Vinod Kumar Vavilapalli commented on YARN-2836:

We had a lengthy discussion about token renewal vs token validation failures at 
YARN-3021. Linking tickets.

> RM behaviour on token renewal failures is broken
> ------------------------------------------------
>                 Key: YARN-2836
>                 URL: https://issues.apache.org/jira/browse/YARN-2836
>             Project: Hadoop YARN
>          Issue Type: Bug
>          Components: resourcemanager
>            Reporter: Vinod Kumar Vavilapalli
>            Assignee: Vinod Kumar Vavilapalli
>            Priority: Blocker
> Found this while reviewing YARN-2834.
> We now completely ignore token renewal failures. For things like Timeline 
> tokens which are automatically obtained whether the app needs it or not (we 
> should fix this to be user driven), we can ignore failures. But for HDFS 
> Tokens etc, ignoring failures is bad because it (1) wastes resources as AMs 
> will continue and eventually fail (2) app doesn't know what happened it fails 
> eventually.

This message was sent by Atlassian JIRA

Reply via email to