[ https://issues.apache.org/jira/browse/YARN-5634?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15671898#comment-15671898 ]
Carlo Curino commented on YARN-5634: ------------------------------------ [~subru] per our offline discussion we can't use ReentrantReadWriteLock as we need to escalate read to write locks, so I am retaining the caching a reference mechanism (so that getHomeSubcluster() operates on a snapshot of the ConcurrentHashMap). To further tighten this I am putting the two {{put}} within a synchronized block (and {{reset()}}) so that we don't risk ghost-updates (very unlikely, but not impossible). > Simplify initialization/use of RouterPolicy via a RouterPolicyFacade > --------------------------------------------------------------------- > > Key: YARN-5634 > URL: https://issues.apache.org/jira/browse/YARN-5634 > Project: Hadoop YARN > Issue Type: Sub-task > Components: nodemanager, resourcemanager > Affects Versions: YARN-2915 > Reporter: Carlo Curino > Assignee: Carlo Curino > Labels: oct16-medium > Attachments: YARN-5634-YARN-2915.01.patch, > YARN-5634-YARN-2915.02.patch, YARN-5634-YARN-2915.03.patch, > YARN-5634-YARN-2915.04.patch, YARN-5634-YARN-2915.05.patch, > YARN-5634-YARN-2915.06.patch, YARN-5634-YARN-2915.07.patch > > > The current set of policies require some machinery to (re)initialize based on > changes in the SubClusterPolicyConfiguration. This JIRA tracks the effort to > hide much of that behind a simple RouterPolicyFacade, making lifecycle and > usage of the policies easier to consumers. -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org