[
https://issues.apache.org/jira/browse/ZOOKEEPER-2137?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15317319#comment-15317319
]
Michael Han commented on ZOOKEEPER-2137:
----------------------------------------
Add a canonical test case that fails pretty often for me locally to make life
easier for those who want a easy reproduce. Also verified this is the only
place in entire ReConfig test that will ever fail; other test cases in ReConfig
test are solid.
{code}
@Test
public void testPortChange() throws Exception {
qu = new QuorumUtil(1); // create 3 servers
qu.disableJMXTest = true;
qu.startAll();
ZooKeeper[] zkArr = createHandles(qu);
List<String> joiningServers = new ArrayList<String>();
int leaderIndex = getLeaderId(qu);
int followerIndex = leaderIndex == 1 ? 2 : 1;
testNormalOperation(zkArr[followerIndex], zkArr[leaderIndex]);
// change leader's leading port - should renounce leadership
int newQuorumPort = PortAssignment.unique();
joiningServers.add("server." + leaderIndex + "=localhost:"
+ newQuorumPort
+ ":"
+ qu.getPeer(leaderIndex).peer.getElectionAddress().getPort()
+ ":participant;localhost:"
+ qu.getPeer(leaderIndex).peer.getClientPort());
reconfig(zkArr[leaderIndex], joiningServers, null, null, -1);
// Fails here.
testNormalOperation(zkArr[followerIndex], zkArr[leaderIndex]);
Assert.assertTrue(qu.getPeer(leaderIndex).peer.getQuorumAddress()
.getPort() == newQuorumPort);
Assert.assertTrue(getLeaderId(qu) != leaderIndex); // the leader changed
joiningServers.clear();
closeAllHandles(zkArr);
}
{code}
> Make testPortChange() less flaky
> --------------------------------
>
> Key: ZOOKEEPER-2137
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-2137
> Project: ZooKeeper
> Issue Type: Sub-task
> Reporter: Hongchao Deng
> Assignee: Michael Han
> Fix For: 3.5.2, 3.6.0
>
> Attachments: ZOOKEEPER-2137.patch
>
>
> The cause of flaky failure of testPortChange() is a race in sync().
> I figured out it could take some time to fix sync(). Meanwhile, we can make
> testPortChange() less flaky by doing reconfig on the leader. We can change
> this back in the fix of ZOOKEEPER-2136.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
