[ https://issues.apache.org/jira/browse/YARN-8808?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16626864#comment-16626864 ]
Arun Suresh commented on YARN-8808: ----------------------------------- {quote}Not sure if I follow you correctly. Sounds like you are referring to aggregateUtilization as the aggregate resource ALLOCATION of all containers, right? {quote} Hmmm.. nope, I meant {{aggregateContainerUtilization}}. Ah.. Apologize if my comment was not clear: I did not mean {{aggregateUtilization _DIVIDED BY_ nodeUtilization}}, I meant {{aggregateUtilization _OR_ nodeUtilization}}. I was just saying.. we need an additional check to see if either one of them (you are proposing to use the former in this JIRA) is {{0}}.. in which case, we should return {{Resources.none()}} since we do not want to over-allocate on a node which hasn't allocated anything yet right ? I hope, now this makes more sense: bq. Also, I am thinking a combination of containerUtilization + nodeUtilization should be used though. Consider the situation where the container utilization is high but the node utilization is low - Node has capacity for 4 1GB containers, but is currently running 2 containers each using more than 1.9GB - in this case, overallocation should be allowed. Again, in the below comment - I was just saying if either one is null, there is a case where we get an NPE, since {{ResourceUtilization.newInstance(..)}} cant handle null arguments. bq. Also, it looks like you need to null check getNodeUtilization() / getAggregatedContainersUtilization() - there seems to be case where you can get an NPE if you don't > Use aggregate container utilization instead of node utilization to determine > resources available for oversubscription > --------------------------------------------------------------------------------------------------------------------- > > Key: YARN-8808 > URL: https://issues.apache.org/jira/browse/YARN-8808 > Project: Hadoop YARN > Issue Type: Sub-task > Affects Versions: YARN-1011 > Reporter: Haibo Chen > Assignee: Haibo Chen > Priority: Major > Attachments: YARN-8088-YARN-1011.01.patch, > YARN-8808-YARN-1011.00.patch > > > Resource oversubscription should be bound to the amount of the resources that > can be allocated to containers, hence the allocation threshold should be with > respect to aggregate container utilization. -- This message was sent by Atlassian JIRA (v7.6.3#76005) --------------------------------------------------------------------- To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org