[ https://issues.apache.org/jira/browse/HADOOP-10077?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13810427#comment-13810427 ]
Colin Patrick McCabe commented on HADOOP-10077: ----------------------------------------------- The {{TestMetricsSystemImpl}} test failure is https://issues.apache.org/jira/browse/HADOOP-9990, not related to this change. > o.a.h.s.Groups should refresh in the background > ----------------------------------------------- > > Key: HADOOP-10077 > URL: https://issues.apache.org/jira/browse/HADOOP-10077 > Project: Hadoop Common > Issue Type: Bug > Affects Versions: 2.2.1 > Reporter: Colin Patrick McCabe > Assignee: Colin Patrick McCabe > Attachments: HADOOP-10077.001.patch > > > {{org.apache.hadoop.security.Groups}} maintains a cache of mappings between > user names and sets of associated group names. Periodically, the entries in > this cache expire and must be refetched from the operating system. > Currently, this is done in the context of whatever thread happens to try to > access the group mapping information right after the time period expires. > However, this is problematic, since that thread may be holding the > {{FSNamesystem}} lock. This means that if the > {{GroupMappingServiceProvider}} is slow, the whole NameNode may grind to a > halt until it finishes. This can generate periodic load spikes or even > NameNode failovers. > Instead, we should allow the refreshing of the group mappings to be done > asynchronously in a background thread pool. -- This message was sent by Atlassian JIRA (v6.1#6144)