use of org.apache.helix.tools.ClusterStateVerifier.BestPossAndExtViewZkVerifier in project helix by apache.
the class TestAddStateModelFactoryAfterConnect method testBasic.
@Test
public void testBasic() throws Exception {
// Logger.getRootLogger().setLevel(Level.INFO);
String className = TestHelper.getTestClassName();
String methodName = TestHelper.getTestMethodName();
String clusterName = className + "_" + methodName;
final int n = 5;
System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
MockParticipantManager[] participants = new MockParticipantManager[n];
// participant port
TestHelper.setupCluster(// participant port
clusterName, // participant port
ZK_ADDR, // participant port
12918, // participant name prefix
"localhost", // resource name prefix
"TestDB", // resources
1, // partitions per resource
10, // number of nodes
n, // replicas
3, "MasterSlave", // do rebalance
true);
ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller_0");
controller.syncStart();
// start participants
for (int i = 0; i < n; i++) {
String instanceName = "localhost_" + (12918 + i);
participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
participants[i].syncStart();
}
boolean result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
// add a new idealState without registering message handling factory
ClusterSetup setupTool = new ClusterSetup(ZK_ADDR);
setupTool.addResourceToCluster(clusterName, "TestDB1", 16, "MasterSlave");
ZkBaseDataAccessor<ZNRecord> baseAccessor = new ZkBaseDataAccessor<ZNRecord>(_gZkClient);
ZKHelixDataAccessor accessor = new ZKHelixDataAccessor(clusterName, baseAccessor);
Builder keyBuilder = accessor.keyBuilder();
IdealState idealState = accessor.getProperty(keyBuilder.idealStates("TestDB1"));
idealState.setStateModelFactoryName("TestDB1_Factory");
accessor.setProperty(keyBuilder.idealStates("TestDB1"), idealState);
setupTool.rebalanceStorageCluster(clusterName, "TestDB1", 3);
// assert that we have received OFFLINE->SLAVE messages for all partitions
int totalMsgs = 0;
for (int retry = 0; retry < 5; retry++) {
Thread.sleep(100);
totalMsgs = 0;
for (int i = 0; i < n; i++) {
List<Message> msgs = accessor.getChildValues(keyBuilder.messages(participants[i].getInstanceName()));
totalMsgs += msgs.size();
}
if (// partition# x replicas
totalMsgs == 48)
break;
}
Assert.assertEquals(totalMsgs, 48, "Should accumulated 48 unprocessed messages (1 O->S per partition per replica) because TestDB1 is added without state-model-factory but was " + totalMsgs);
// Logger.getRootLogger().setLevel(Level.INFO);
for (int i = 0; i < n; i++) {
participants[i].getStateMachineEngine().registerStateModelFactory("MasterSlave", new MockMSModelFactory(), "TestDB1_Factory");
}
result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
// clean up
// wait for all zk callbacks done
controller.syncStop();
for (int i = 0; i < 5; i++) {
participants[i].syncStop();
}
System.out.println("END " + clusterName + " at " + new Date(System.currentTimeMillis()));
}
use of org.apache.helix.tools.ClusterStateVerifier.BestPossAndExtViewZkVerifier in project helix by apache.
the class TestCarryOverBadCurState method testCarryOverBadCurState.
@Test
public void testCarryOverBadCurState() throws Exception {
System.out.println("START testCarryOverBadCurState at " + new Date(System.currentTimeMillis()));
String clusterName = getShortClassName();
MockParticipantManager[] participants = new MockParticipantManager[5];
// participant port
TestHelper.setupCluster(// participant port
clusterName, // participant port
ZK_ADDR, // participant port
12918, // participant name prefix
"localhost", // resource name prefix
"TestDB", // resources
1, // partitions per resource
10, // number of nodes
5, // replicas
3, "MasterSlave", // do rebalance
true);
// add a bad current state
ZNRecord badCurState = new ZNRecord("TestDB0");
String path = PropertyPathBuilder.instanceCurrentState(clusterName, "localhost_12918", "session_0", "TestDB0");
_gZkClient.createPersistent(path, true);
_gZkClient.writeData(path, badCurState);
ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller_0");
controller.syncStart();
// start participants
for (int i = 0; i < 5; i++) {
String instanceName = "localhost_" + (12918 + i);
participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
participants[i].syncStart();
}
boolean result = ClusterStateVerifier.verifyByZkCallback(new MasterNbInExtViewVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
// clean up
controller.syncStop();
for (int i = 0; i < 5; i++) {
participants[i].syncStop();
}
System.out.println("END testCarryOverBadCurState at " + new Date(System.currentTimeMillis()));
}
use of org.apache.helix.tools.ClusterStateVerifier.BestPossAndExtViewZkVerifier in project helix by apache.
the class TestEntropyFreeNodeBounce method testBounceAll.
@Test
public void testBounceAll() throws Exception {
// pick numbers that don't divide evenly
final int NUM_PARTICIPANTS = 5;
final int NUM_PARTITIONS = 123;
final int NUM_REPLICAS = 1;
final String RESOURCE_PREFIX = "TestDB";
final String RESOURCE_NAME = RESOURCE_PREFIX + "0";
// create a cluster name based on this test name
String className = TestHelper.getTestClassName();
String methodName = TestHelper.getTestMethodName();
String clusterName = className + "_" + methodName;
System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
// Set up cluster
// participant port
TestHelper.setupCluster(// participant port
clusterName, // participant port
ZK_ADDR, // participant port
12918, // participant name prefix
"localhost", // resource name prefix
"TestDB", // resources
1, // partitions per resource
NUM_PARTITIONS, // number of nodes
NUM_PARTICIPANTS, // replicas
NUM_REPLICAS, // use FULL_AUTO mode to test node tagging
"OnlineOffline", // use FULL_AUTO mode to test node tagging
RebalanceMode.FULL_AUTO, // do rebalance
true);
// Start the participants
HelixManager[] participants = new HelixManager[NUM_PARTICIPANTS];
for (int i = 0; i < NUM_PARTICIPANTS; i++) {
final String instanceName = "localhost_" + (12918 + i);
participants[i] = createParticipant(clusterName, instanceName);
participants[i].connect();
}
// Start the controller
ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller");
controller.syncStart();
// get an admin and accessor
HelixAdmin helixAdmin = new ZKHelixAdmin(_gZkClient);
BaseDataAccessor<ZNRecord> baseAccessor = new ZkBaseDataAccessor<ZNRecord>(_gZkClient);
HelixDataAccessor accessor = new ZKHelixDataAccessor(clusterName, baseAccessor);
PropertyKey.Builder keyBuilder = accessor.keyBuilder();
// do the test
try {
Thread.sleep(1000);
// ensure that the external view coalesces
boolean result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
ExternalView stableExternalView = accessor.getProperty(keyBuilder.externalView(RESOURCE_NAME));
for (HelixManager participant : participants) {
// disable the controller, bounce the node, re-enable the controller, verify assignments
// remained the same
helixAdmin.enableCluster(clusterName, false);
participant.disconnect();
Thread.sleep(1000);
participant = createParticipant(clusterName, participant.getInstanceName());
participant.connect();
Thread.sleep(1000);
helixAdmin.enableCluster(clusterName, true);
Thread.sleep(1000);
result = ClusterStateVerifier.verifyByZkCallback(new MatchingExternalViewVerifier(stableExternalView, clusterName));
Assert.assertTrue(result);
}
} finally {
// clean up
controller.syncStop();
for (HelixManager participant : participants) {
participant.disconnect();
}
System.out.println("END " + clusterName + " at " + new Date(System.currentTimeMillis()));
}
}
use of org.apache.helix.tools.ClusterStateVerifier.BestPossAndExtViewZkVerifier in project helix by apache.
the class TestParticipantManager method testSessionExpiryInTransition.
@Test
public void testSessionExpiryInTransition() throws Exception {
String className = TestHelper.getTestClassName();
String methodName = TestHelper.getTestMethodName();
final String clusterName = className + "_" + methodName;
int n = 1;
CountDownLatch startCountdown = new CountDownLatch(1);
CountDownLatch endCountdown = new CountDownLatch(1);
System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
MockParticipantManager[] participants = new MockParticipantManager[n];
// participant port
TestHelper.setupCluster(// participant port
clusterName, // participant port
ZK_ADDR, // participant port
12918, // participant name prefix
"localhost", // resource name prefix
"TestDB", // resources
1, // partitions per resource
1, // number of nodes
n, // replicas
1, "MasterSlave", // do rebalance
true);
// start controller
ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller_0");
controller.syncStart();
// start participants
for (int i = 0; i < n; i++) {
String instanceName = "localhost_" + (12918 + i);
participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
participants[i].setTransition(new SessionExpiryTransition(startCountdown, endCountdown));
participants[i].syncStart();
}
// wait transition happens to trigger session expiry
startCountdown.await();
String oldSessionId = participants[0].getSessionId();
System.out.println("oldSessionId: " + oldSessionId);
ZkTestHelper.expireSession(participants[0].getZkClient());
boolean result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
String newSessionId = participants[0].getSessionId();
Assert.assertNotSame(newSessionId, oldSessionId);
// assert interrupt exception error in old session
String errPath = PropertyPathBuilder.instanceError(clusterName, "localhost_12918", oldSessionId, "TestDB0", "TestDB0_0");
ZNRecord error = _gZkClient.readData(errPath);
Assert.assertNotNull(error, "InterruptedException should happen in old session since task is being cancelled during handleNewSession");
String errString = new String(new ZNRecordSerializer().serialize(error));
Assert.assertTrue(errString.indexOf("InterruptedException") != -1);
// cleanup
controller.syncStop();
for (int i = 0; i < n; i++) {
participants[i].syncStop();
}
System.out.println("END " + clusterName + " at " + new Date(System.currentTimeMillis()));
}
use of org.apache.helix.tools.ClusterStateVerifier.BestPossAndExtViewZkVerifier in project helix by apache.
the class TestBatchMessage method testBasic.
@Test
public void testBasic() throws Exception {
// Logger.getRootLogger().setLevel(Level.INFO);
String className = TestHelper.getTestClassName();
String methodName = TestHelper.getTestMethodName();
String clusterName = className + "_" + methodName;
int n = 2;
System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
// participant port
TestHelper.setupCluster(// participant port
clusterName, // participant port
ZK_ADDR, // participant port
12918, // participant name prefix
"localhost", // resource name prefix
"TestDB", // resources
1, // partitions per resource
32, // number of nodes
n, // replicas
2, "MasterSlave", // do rebalance
true);
// enable batch message
ZKHelixDataAccessor accessor = new ZKHelixDataAccessor(clusterName, new ZkBaseDataAccessor<ZNRecord>(_gZkClient));
Builder keyBuilder = accessor.keyBuilder();
IdealState idealState = accessor.getProperty(keyBuilder.idealStates("TestDB0"));
idealState.setBatchMessageMode(true);
accessor.setProperty(keyBuilder.idealStates("TestDB0"), idealState);
// register a message listener so we know how many message generated
TestZkChildListener listener = new TestZkChildListener();
_gZkClient.subscribeChildChanges(keyBuilder.messages("localhost_12918").getPath(), listener);
ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller_0");
controller.syncStart();
// start participants
MockParticipantManager[] participants = new MockParticipantManager[n];
for (int i = 0; i < n; i++) {
String instanceName = "localhost_" + (12918 + i);
participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
participants[i].syncStart();
}
boolean result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
// Change to three is because there is an extra factory registered
// So one extra NO_OP message send
Assert.assertTrue(listener._maxNbOfChilds <= 3, "Should get no more than 2 messages (O->S and S->M)");
// clean up
// wait for all zk callbacks done
controller.syncStop();
for (int i = 0; i < n; i++) {
participants[i].syncStop();
}
System.out.println("END " + clusterName + " at " + new Date(System.currentTimeMillis()));
}
Aggregations