[jira] [Commented] (YARN-5606) Support multi-label merge into one node label

["Bin Wang (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/YARN-5606?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16000385#comment-16000385 ] 

Bin Wang commented on YARN-5606:
--------------------------------

Hi Jialei, this one is different than Constrain Label. Constrain Label is implemented in parallel with the node label expression, this is a enhancement to the node label expression. Normally a node label expression is a resource pool, if user want's to use more resource pools for one application, there is no way at current. Introducing node label wildcard matching is a way to solve this situation. In order to make matching more reasonable, wildcard pattern can only match the full substrings separated by "&".

> Support multi-label merge into one node label
> ---------------------------------------------
>
>                 Key: YARN-5606
>                 URL: https://issues.apache.org/jira/browse/YARN-5606
>             Project: Hadoop YARN
>          Issue Type: New Feature
>          Components: resourcemanager
>            Reporter: jialei weng
>              Labels: oct16-hard
>         Attachments: YARN-5606.001.patch, YARN-5606.002.patch
>
>
> Support multi-label merge into one node label
> 1. we want to support multo-label like SSD,GPU,FPGA label merged into single machine, joined by &. like SSD,GPU,FPGA-> SSD&GPU&FPGA
> 2. we add wildcard match to extend the job request. we define the wildcard like *GPU*, it will math all the node labels with GPU as part of multi-label merged label. For example, *GPU* will match  SSD&GPU&FPGA, GPU&FPGA. We define  SSD&GPU&FPGA={SSD,GPU,FPGA}, and GPU is one of {SSD,GPU,FPGA}, so the job can run on  SSD&GPU&FPGA node.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org