[ https://issues.apache.org/jira/browse/YARN-2721?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14179067#comment-14179067 ]
Jian He commented on YARN-2721: ------------------------------- Curator should handle the retry properly which is addressed in YARN-2716. As a temporary fix, we can simply ignore the potential NodeExist exception for now. Creating a patch. > Race condition: ZKRMStateStore retry logic may throw NodeExist exception > ------------------------------------------------------------------------- > > Key: YARN-2721 > URL: https://issues.apache.org/jira/browse/YARN-2721 > Project: Hadoop YARN > Issue Type: Bug > Reporter: Jian He > Assignee: Jian He > Fix For: 2.6.0 > > > Blindly retrying operations in zookeeper will not work for non-idempotent > operations (like create znode). The reason is that the client can do a create > znode, but the response may not be returned because the server can die or > timeout. In case of retrying the create znode, it will throw a NODE_EXISTS > exception from the earlier create from the same session. -- This message was sent by Atlassian JIRA (v6.3.4#6332)