Enis Soztutar created HBASE-6822:
------------------------------------
Summary: [WINDOWS] MiniZookeeperCluster multiple daemons bind to
the same port
Key: HBASE-6822
URL: https://issues.apache.org/jira/browse/HBASE-6822
Project: HBase
Issue Type: Bug
Affects Versions: 0.96.0, 0.94.3
Reporter: Enis Soztutar
Assignee: Enis Soztutar
TestHBaseTestingUtility.testMiniZooKeeper() tests whether the mini zk cluster
is working by launching 5 threads corresponding to zk servers.
NIOServerCnxnFactory.configure() configures the socket as:
{code}
this.ss = ServerSocketChannel.open();
ss.socket().setReuseAddress(true);
{code}
setReuseAddress() is set, because it allows the server to come back up and bind
to the same port before the socket is timed-out by the kernel.
Under windows, the behavior on ServerSocket.setReuseAddress() is different than
on linux, in which it allows any process to bind to an already-bound port. This
causes ZK nodes starting on the same node, to be able to bind to the same port.
The following part of the patch at
https://issues.apache.org/jira/browse/HADOOP-8223 deals with this case for
Hadoop:
{code}
if(Shell.WINDOWS) {
+ // result of setting the SO_REUSEADDR flag is different on Windows
+ // http://msdn.microsoft.com/en-us/library/ms740621(v=vs.85).aspx
+ // without this 2 NN's can start on the same machine and listen on
+ // the same port with indeterminate routing of incoming requests to them
+ ret.setReuseAddress(false);
+ }
{code}
We should do the same in Zookeeper (I'll open a ZOOK issue). But in the
meantime, we can fix hbase tests to not rely on BindException to resolve for
bind errors. Especially, in MiniZKCluster.startup() when starting more than 1
servers, we already know that we have to increment the port number.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
