[HDFS-891] DataNode no longer needs to check for dfs.network.script

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 0.23.0
• Fix Version/s: 2.0.0-alpha
• Component/s: datanode
• Hadoop Flags: Reviewed
• Assignee: Harsh J
• Reporter: Steve Loughran
• Created: 09/Jan/10 14:55
• Updated: 28/Sep/15 20:58
• Resolved: 19/Apr/12 01:10

Description

Looking at the code for DataNode.instantiateDataNode())} , I see that it calls {{system.exit(-1) if it is not happy with the configuration

    if (conf.get("dfs.network.script") != null) {
      LOG.error("This configuration for rack identification is not supported" +
          " anymore. RackID resolution is handled by the NameNode.");
      System.exit(-1);
    }

This is excessive. It should throw an exception and let whoever called the method decide how to handle it. The DataNode.main() method will log the exception and exit with a -1 value, but other callers (such as anything using MiniDFSCluster will now see a meaningful message rather than some Junit "tests exited without completing" warning.

Easy to write a test for the correct behaviour: start a MiniDFSCluster with this configuration set, see what happens.

Issue Links

• relates to: HADOOP-1985 Abstract node to switch mapping into a topology service class used by namenode and jobtracker