[ https://issues.apache.org/jira/browse/HADOOP-12474?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14955544#comment-14955544 ]
Haohui Mai commented on HADOOP-12474: ------------------------------------- +1. I'll commit it shortly. > Fix data race when allocating server port in MiniKMS > ----------------------------------------------------- > > Key: HADOOP-12474 > URL: https://issues.apache.org/jira/browse/HADOOP-12474 > Project: Hadoop Common > Issue Type: Bug > Reporter: Mingliang Liu > Assignee: Mingliang Liu > Attachments: HADOOP-12474.000.patch, HADOOP-12474.001.patch > > > In {{MiniKMS}}, there is potential data race when create jetty server in > {{createJettyServer}}. It looks like the code searches for a free port and > then starts jetty, but maybe there's enough of a race condition between port > location and jetty.start to cause intermittent failures. > {code} > ServerSocket ss = new ServerSocket((inPort < 0) ? 0 : inPort, 50, > localhost); > int port = ss.getLocalPort(); > Server server = new Server(0); > if (!ssl) { > server.getConnectors()[0].setHost(host); > server.getConnectors()[0].setPort(port); > } else { > ... > c.setPort(port); > {code} > We've seen test failures saying {{java.net.BindException: Address already in > use}}, e.g. > [https://builds.apache.org/job/PreCommit-HDFS-Build/12942/testReport/] > As in [HADOOP-12417], we should always bind port 0 which gives us an > ephemeral port, instead of searching a free port before starting jetty. -- This message was sent by Atlassian JIRA (v6.3.4#6332)