[ https://issues.apache.org/jira/browse/HDFS-9410?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15000120#comment-15000120 ]
Walter Su commented on HDFS-9410: --------------------------------- Fortunately we fork a JVM for each test class. I think we still need to fix this to avoid System.out pollution in other test cases in same class. It's just I think can we backup {{originOut}} at class level? {code} private static final PrintStream originOut = System.out; private static final PrintStream originErr = System.err; {code} And I really don't trust {{System.err.flush();}} won't throw any unexpected runtime exception. So I think can we put the restore code in a "finally\{ //here \}"? Just some thought, if you don't agree. Anyway, the patch looks good to me. > TestDFSAdminWithHA#tearDown may throw NPE > ----------------------------------------- > > Key: HDFS-9410 > URL: https://issues.apache.org/jira/browse/HDFS-9410 > Project: Hadoop HDFS > Issue Type: Test > Components: HDFS > Reporter: Xiao Chen > Assignee: Xiao Chen > Priority: Minor > Attachments: HDFS-9410.001.patch > > > If for any reason {{setUpHaCluster}} fails before storing {{System.out}} and > {{System.err}} as member variables {{originOut}} and {{originErr}}, then in > {{tearDown}} {{System.out}} and {{System.err}} are set to null. This could > cause all following tests to fail when calling {{flush}}. -- This message was sent by Atlassian JIRA (v6.3.4#6332)