@Test
public void testClusterRestart() throws Exception {
cluster.stop();
// Give the cluster time to stop the whole cluster.
AbstractDaemonClient cli = cluster.getJTClient();
int i = 1;
while (i < 40) {
try {
cli.ping();
Thread.sleep(1000);
i++;
} catch (Exception e) {
break;
}
}
if (i >= 40) {
Assert.fail("JT on " + cli.getHostName() + " Should have been down.");
}
i = 1;
for (AbstractDaemonClient tcli : cluster.getTTClients()) {
i = 1;
while (i < 40) {
try {
tcli.ping();
Thread.sleep(1000);
i++;
} catch (Exception e) {
break;
}
}
if (i >= 40) {
Assert.fail("TT on " + tcli.getHostName() + " Should have been down.");
}
}
cluster.start();
cli = cluster.getJTClient();
i = 1;
while (i < 40) {
try {
cli.ping();
break;
} catch (Exception e) {
i++;
Thread.sleep(1000);
LOG.info("Waiting for Jobtracker on host : " + cli.getHostName() + " to come up.");
}
}
if (i >= 40) {
Assert.fail("JT on " + cli.getHostName() + " Should have been up.");
}
for (AbstractDaemonClient tcli : cluster.getTTClients()) {
i = 1;
while (i < 40) {
try {
tcli.ping();
break;
} catch (Exception e) {
i++;
Thread.sleep(1000);
LOG.info("Waiting for Tasktracker on host : " + tcli.getHostName() + " to come up.");
}
}
if (i >= 40) {
Assert.fail("TT on " + tcli.getHostName() + " Should have been Up.");
}
}
}
