[ https://issues.apache.org/jira/browse/YARN-2571?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14269548#comment-14269548 ]
Steve Loughran commented on YARN-2571: -------------------------------------- Vinod, having the RM create the user paths allows the registry to be set up with the correct permissions as YARN jobs are created. Without that, if there is no path for that user set up, the application is likely to fail post-launch with some error. For cleanup, automatic purging of records keeps the registry data somewhat under control, without applications having to go to the effort of writing these not-yet-implemented cleanup containers. It's not a particularly complex piece of code; there's tests for the distributed shell that verify it works in YARN-2646 > RM to support YARN registry > ---------------------------- > > Key: YARN-2571 > URL: https://issues.apache.org/jira/browse/YARN-2571 > Project: Hadoop YARN > Issue Type: Sub-task > Components: resourcemanager > Reporter: Steve Loughran > Assignee: Steve Loughran > Attachments: YARN-2571-001.patch, YARN-2571-002.patch, > YARN-2571-003.patch, YARN-2571-005.patch, YARN-2571-007.patch, > YARN-2571-008.patch, YARN-2571-009.patch > > > The RM needs to (optionally) integrate with the YARN registry: > # startup: create the /services and /users paths with system ACLs (yarn, hdfs > principals) > # app-launch: create the user directory /users/$username with the relevant > permissions (CRD) for them to create subnodes. > # attempt, container, app completion: remove service records with the > matching persistence and ID -- This message was sent by Atlassian JIRA (v6.3.4#6332)