public void testErrorPartition() throws Exception {
String clusterName = TestUtil.getTestName();
MockParticipant[] participants = new MockParticipant[5];
System.out.println("START testErrorPartition() at " + new Date(System.currentTimeMillis()));
ZKHelixAdmin tool = new ZKHelixAdmin(_zkclient);
TestHelper.setupCluster(clusterName, _zkaddr, 12918, "localhost", "TestDB", 1, 10, 5, 3,
"MasterSlave", true);
MockController controller =
new MockController(_zkaddr, clusterName, "controller_0");
controller.syncStart();
for (int i = 0; i < 5; i++) {
String instanceName = "localhost_" + (12918 + i);
if (i == 0) {
Map<String, Set<String>> errPartitions = new HashMap<String, Set<String>>();
errPartitions.put("SLAVE-MASTER", TestHelper.setOf("TestDB0_4"));
participants[i] = new MockParticipant(_zkaddr, clusterName, instanceName);
participants[i].setTransition(new ErrTransition(errPartitions));
} else {
participants[i] = new MockParticipant(_zkaddr, clusterName, instanceName);
}
participants[i].syncStart();
}
Map<String, Map<String, String>> errStates = new HashMap<String, Map<String, String>>();
errStates.put("TestDB0", new HashMap<String, String>());
errStates.get("TestDB0").put("TestDB0_4", "localhost_12918");
boolean result =
ClusterStateVerifier.verifyByPolling(new ClusterStateVerifier.BestPossAndExtViewZkVerifier(
_zkaddr, clusterName, errStates));
Assert.assertTrue(result);
Map<String, Set<String>> errorStateMap = new HashMap<String, Set<String>>();
errorStateMap.put("TestDB0_4", TestHelper.setOf("localhost_12918"));
// verify "TestDB0_0", "localhost_12918" is in ERROR state
TestHelper.verifyState(clusterName, _zkaddr, errorStateMap, "ERROR");
// disable a partition on a node with error state
tool.enablePartition(false, clusterName, "localhost_12918", "TestDB0",
Arrays.asList("TestDB0_4"));
result =
ClusterStateVerifier.verifyByPolling(new ClusterStateVerifier.BestPossAndExtViewZkVerifier(
_zkaddr, clusterName, errStates));
Assert.assertTrue(result);
TestHelper.verifyState(clusterName, _zkaddr, errorStateMap, "ERROR");
// disable a node with error state
tool.enableInstance(clusterName, "localhost_12918", false);
result =
ClusterStateVerifier.verifyByPolling(new ClusterStateVerifier.BestPossAndExtViewZkVerifier(
_zkaddr, clusterName, errStates));
Assert.assertTrue(result);
// make sure after restart stale ERROR state is gone
tool.enablePartition(true, clusterName, "localhost_12918", "TestDB0",
Arrays.asList("TestDB0_4"));
tool.enableInstance(clusterName, "localhost_12918", true);
participants[0].syncStop();
result =
ClusterStateVerifier.verifyByPolling(new ClusterStateVerifier.BestPossAndExtViewZkVerifier(
_zkaddr, clusterName));