[jira] [Commented] (YARN-1193) ResourceManger.clusterTimeStamp should be reset when RM transitions to active
[ https://issues.apache.org/jira/browse/YARN-1193?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13766645#comment-13766645 ] Bikas Saha commented on YARN-1193: -- isnt this already fixed in YARN-1027. Did you mean to open this jira to create a getClusterTimestamp() method? ResourceManger.clusterTimeStamp should be reset when RM transitions to active - Key: YARN-1193 URL: https://issues.apache.org/jira/browse/YARN-1193 Project: Hadoop YARN Issue Type: Sub-task Components: resourcemanager Reporter: Karthik Kambatla Assignee: Karthik Kambatla Attachments: yarn-1193-1.patch ResourceManager.clusterTimeStamp is used to generate application-ids. Currently, when the RM transitions to active-standby-active back and forth, the clusterTimeStamp stays the same leading to apps getting the same ids as jobs from before. This leads to other races in staging directory etc. To avoid this, it is better to set it on every transition to Active. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (YARN-1193) ResourceManger.clusterTimeStamp should be reset when RM transitions to active
[ https://issues.apache.org/jira/browse/YARN-1193?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13766684#comment-13766684 ] Karthik Kambatla commented on YARN-1193: [~bikassaha], looks like I misunderstood your comments on YARN-1027. The latest patch there (yarn-1027-7.patch) doesn't have this. clusterTimeStamp is a public static final variable in ResourceManager. To set it when transitioning to Active, we need to make it non-final, which would expose a public static variable. This change, I thought, should go with making it private and adding a public get method. Do you suggest I merge this back into YARN-1027? Or, is it okay to handle it separately. I think it might be cleaner to handle separately so it is easier for people to understand why a particular change has been made. ResourceManger.clusterTimeStamp should be reset when RM transitions to active - Key: YARN-1193 URL: https://issues.apache.org/jira/browse/YARN-1193 Project: Hadoop YARN Issue Type: Sub-task Components: resourcemanager Reporter: Karthik Kambatla Assignee: Karthik Kambatla Attachments: yarn-1193-1.patch ResourceManager.clusterTimeStamp is used to generate application-ids. Currently, when the RM transitions to active-standby-active back and forth, the clusterTimeStamp stays the same leading to apps getting the same ids as jobs from before. This leads to other races in staging directory etc. To avoid this, it is better to set it on every transition to Active. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (YARN-1193) ResourceManger.clusterTimeStamp should be reset when RM transitions to active
[ https://issues.apache.org/jira/browse/YARN-1193?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13766701#comment-13766701 ] Bikas Saha commented on YARN-1193: -- The patch in YARN-1027 is incorrect without the clustertimestamp modification. However, we dont need to add a getClusterTimeStamp method in that patch itself. That can be done in a separate jira. So I meant that we can do the method refactoring separately. Sorry for not being clear. However, in interest of getting YARN-1027 done I will take back those comments. So lets keep those changes in YARN-1027. I am closing this jira. ResourceManger.clusterTimeStamp should be reset when RM transitions to active - Key: YARN-1193 URL: https://issues.apache.org/jira/browse/YARN-1193 Project: Hadoop YARN Issue Type: Sub-task Components: resourcemanager Reporter: Karthik Kambatla Assignee: Karthik Kambatla Attachments: yarn-1193-1.patch ResourceManager.clusterTimeStamp is used to generate application-ids. Currently, when the RM transitions to active-standby-active back and forth, the clusterTimeStamp stays the same leading to apps getting the same ids as jobs from before. This leads to other races in staging directory etc. To avoid this, it is better to set it on every transition to Active. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira