@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
TestHelper.setupCluster(
clusterName,
ZK_ADDR,
12918, // participant port
"localhost", // participant name prefix
"TestDB", // resource name prefix
1, // resources
NUM_PARTITIONS, // partitions per resource
NUM_PARTICIPANTS, // number of nodes
NUM_REPLICAS, // replicas
"OnlineOffline",
RebalanceMode.FULL_AUTO, // use FULL_AUTO mode to test node tagging
true); // do rebalance
// 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()));
}
}
@Test
public void testRemoveUserCbHandlerOnPathRemoval() throws Exception {
String className = TestHelper.getTestClassName();
String methodName = TestHelper.getTestMethodName();
String clusterName = className + "_" + methodName;
final int n = 3;
final String zkAddr = ZK_ADDR;
System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
TestHelper.setupCluster(
clusterName,
zkAddr,
12918,
"localhost",
"TestDB",
1, // resource
32, // partitions
n, // nodes
2, // replicas
"MasterSlave",
true);
final ClusterControllerManager controller =
new ClusterControllerManager(zkAddr, clusterName, "controller_0");
controller.syncStart();
MockParticipantManager[] participants = new MockParticipantManager[n];
for (int i = 0; i < n; i++) {
String instanceName = "localhost_" + (12918 + i);
participants[i] = new MockParticipantManager(zkAddr, clusterName, instanceName);
participants[i].syncStart();
// register a controller listener on participant_0
if (i == 0) {
// ZkHelixTestManager manager = participants[0].getManager();
MockParticipantManager manager = participants[0];
manager.addCurrentStateChangeListener(
new CurrentStateChangeListener() {
@Override
public void onStateChange(
String instanceName,
List<CurrentState> statesInfo,
NotificationContext changeContext) {
// To change body of implemented methods use File | Settings | File Templates.
// System.out.println(instanceName + " on current-state change, type: " +
// changeContext.getType());
}
},
manager.getInstanceName(),
manager.getSessionId());
}
}
Boolean result =
ClusterStateVerifier.verifyByZkCallback(
new ClusterStateVerifier.BestPossAndExtViewZkVerifier(zkAddr, clusterName));
Assert.assertTrue(result);
MockParticipantManager participantToExpire = participants[0];
String oldSessionId = participantToExpire.getSessionId();
PropertyKeyBuilder keyBuilder = new PropertyKeyBuilder(clusterName);
// check manager#hanlders
Assert.assertEquals(
participantToExpire.getHandlers().size(),
3,
"Should have 3 handlers: CURRENTSTATE/{sessionId}, CONTROLLER, and MESSAGES");
// check zkclient#listeners
Map<String, Set<IZkDataListener>> dataListeners =
ZkTestHelper.getZkDataListener(participantToExpire.getZkClient());
Map<String, Set<IZkChildListener>> childListeners =
ZkTestHelper.getZkChildListener(participantToExpire.getZkClient());
// printZkListeners(participantToExpire.getZkClient());
Assert.assertEquals(
dataListeners.size(),
1,
"Should have 1 path (CURRENTSTATE/{sessionId}/TestDB0) which has 1 data-listeners");
String path =
keyBuilder
.currentState(participantToExpire.getInstanceName(), oldSessionId, "TestDB0")
.getPath();
Assert.assertEquals(
dataListeners.get(path).size(), 1, "Should have 1 data-listeners on path: " + path);
Assert.assertEquals(
childListeners.size(),
3,
"Should have 3 paths (CURRENTSTATE/{sessionId}, CONTROLLER, and MESSAGES) each of which has 1 child-listener");
path = keyBuilder.currentStates(participantToExpire.getInstanceName(), oldSessionId).getPath();
Assert.assertEquals(
childListeners.get(path).size(), 1, "Should have 1 child-listener on path: " + path);
path = keyBuilder.messages(participantToExpire.getInstanceName()).getPath();
Assert.assertEquals(
childListeners.get(path).size(), 1, "Should have 1 child-listener on path: " + path);
path = keyBuilder.controller().getPath();
Assert.assertEquals(
childListeners.get(path).size(), 1, "Should have 1 child-listener on path: " + path);
// check zookeeper#watches on client side
Map<String, List<String>> watchPaths =
ZkTestHelper.getZkWatch(participantToExpire.getZkClient());
// System.out.println("localhost_12918 zk-client side watchPaths: " + watchPaths + "\n");
Assert.assertEquals(
watchPaths.get("dataWatches").size(),
4,
"Should have 4 data-watches: CURRENTSTATE/{sessionId}, CURRENTSTATE/{sessionId}/TestDB, CONTROLLER, MESSAGES");
Assert.assertEquals(
watchPaths.get("childWatches").size(),
3,
"Should have 3 child-watches: CONTROLLER, MESSAGES, and CURRENTSTATE/{sessionId}");
// expire localhost_12918
System.out.println(
"Expire participant: "
+ participantToExpire.getInstanceName()
+ ", session: "
+ participantToExpire.getSessionId());
ZkTestHelper.expireSession(participantToExpire.getZkClient());
String newSessionId = participantToExpire.getSessionId();
System.out.println(
participantToExpire.getInstanceName()
+ " oldSessionId: "
+ oldSessionId
+ ", newSessionId: "
+ newSessionId);
result =
ClusterStateVerifier.verifyByZkCallback(
new ClusterStateVerifier.BestPossAndExtViewZkVerifier(zkAddr, clusterName));
Assert.assertTrue(result);
// check manager#hanlders
Assert.assertEquals(
participantToExpire.getHandlers().size(),
2,
"Should have 2 handlers: CONTROLLER and MESSAGES. CURRENTSTATE/{sessionId} handler should be removed by CallbackHandler#handleChildChange()");
// check zkclient#listeners
dataListeners = ZkTestHelper.getZkDataListener(participantToExpire.getZkClient());
childListeners = ZkTestHelper.getZkChildListener(participantToExpire.getZkClient());
// printZkListeners(participantToExpire.getZkClient());
Assert.assertTrue(dataListeners.isEmpty(), "Should have no data-listeners");
Assert.assertEquals(
childListeners.size(),
3,
"Should have 3 paths (CURRENTSTATE/{oldSessionId}, CONTROLLER, and MESSAGES). "
+ "CONTROLLER and MESSAGE has 1 child-listener each. CURRENTSTATE/{oldSessionId} doesn't have listener (ZkClient doesn't remove empty childListener set. probably a ZkClient bug. see ZkClient#unsubscribeChildChange())");
path = keyBuilder.currentStates(participantToExpire.getInstanceName(), oldSessionId).getPath();
Assert.assertEquals(
childListeners.get(path).size(), 0, "Should have no child-listener on path: " + path);
path = keyBuilder.messages(participantToExpire.getInstanceName()).getPath();
Assert.assertEquals(
childListeners.get(path).size(), 1, "Should have 1 child-listener on path: " + path);
path = keyBuilder.controller().getPath();
Assert.assertEquals(
childListeners.get(path).size(), 1, "Should have 1 child-listener on path: " + path);
// check zookeeper#watches on client side
watchPaths = ZkTestHelper.getZkWatch(participantToExpire.getZkClient());
// System.out.println("localhost_12918 zk-client side watchPaths: " + watchPaths + "\n");
Assert.assertEquals(
watchPaths.get("dataWatches").size(),
2,
"Should have 2 data-watches: CONTROLLER and MESSAGES");
Assert.assertEquals(
watchPaths.get("childWatches").size(),
2,
"Should have 2 child-watches: CONTROLLER and MESSAGES");
Assert.assertEquals(
watchPaths.get("existWatches").size(),
2,
"Should have 2 exist-watches: CURRENTSTATE/{oldSessionId} and CURRENTSTATE/{oldSessionId}/TestDB0");
// another session expiry on localhost_12918 should clear the two exist-watches on
// CURRENTSTATE/{oldSessionId}
System.out.println(
"Expire participant: "
+ participantToExpire.getInstanceName()
+ ", session: "
+ participantToExpire.getSessionId());
ZkTestHelper.expireSession(participantToExpire.getZkClient());
result =
ClusterStateVerifier.verifyByZkCallback(
new ClusterStateVerifier.BestPossAndExtViewZkVerifier(zkAddr, clusterName));
Assert.assertTrue(result);
// check zookeeper#watches on client side
watchPaths = ZkTestHelper.getZkWatch(participantToExpire.getZkClient());
// System.out.println("localhost_12918 zk-client side watchPaths: " + watchPaths + "\n");
Assert.assertEquals(
watchPaths.get("dataWatches").size(),
2,
"Should have 2 data-watches: CONTROLLER and MESSAGES");
Assert.assertEquals(
watchPaths.get("childWatches").size(),
2,
"Should have 2 child-watches: CONTROLLER and MESSAGES");
Assert.assertEquals(
watchPaths.get("existWatches").size(),
0,
"Should have no exist-watches. exist-watches on CURRENTSTATE/{oldSessionId} and CURRENTSTATE/{oldSessionId}/TestDB0 should be cleared during handleNewSession");
// Thread.sleep(1000);
// clean up
controller.syncStop();
for (int i = 0; i < n; i++) {
participants[i].syncStop();
}
System.out.println("END " + clusterName + " at " + new Date(System.currentTimeMillis()));
}
@Test
public void testCbHandlerLeakOnParticipantSessionExpiry() throws Exception {
// Logger.getRootLogger().setLevel(Level.INFO);
String className = TestHelper.getTestClassName();
String methodName = TestHelper.getTestMethodName();
String clusterName = className + "_" + methodName;
final int n = 2;
System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
TestHelper.setupCluster(
clusterName,
ZK_ADDR,
12918, // participant port
"localhost", // participant name prefix
"TestDB", // resource name prefix
1, // resources
32, // partitions per resource
n, // number of nodes
2, // replicas
"MasterSlave",
true); // do rebalance
final 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 ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
final MockParticipantManager participantManagerToExpire = participants[1];
// check controller zk-watchers
result =
TestHelper.verify(
new TestHelper.Verifier() {
@Override
public boolean verify() throws Exception {
Map<String, Set<String>> watchers = ZkTestHelper.getListenersBySession(ZK_ADDR);
// Set<String> watchPaths = watchers.get("0x" + controllerManager.getSessionId());
Set<String> watchPaths = watchers.get("0x" + controller.getSessionId());
// System.out.println("controller watch paths: " + watchPaths);
// controller should have 5 + 2n + m + (m+2)n zk-watchers
// where n is number of nodes and m is number of resources
return watchPaths.size() == (6 + 5 * n);
}
},
500);
Assert.assertTrue(result, "Controller should have 6 + 5*n zk-watchers.");
// check participant zk-watchers
result =
TestHelper.verify(
new TestHelper.Verifier() {
@Override
public boolean verify() throws Exception {
Map<String, Set<String>> watchers = ZkTestHelper.getListenersBySession(ZK_ADDR);
Set<String> watchPaths =
watchers.get("0x" + participantManagerToExpire.getSessionId());
// System.out.println("participant watch paths: " + watchPaths);
// participant should have 2 zk-watchers: 1 for MESSAGE and 1 for CONTROLLER
return watchPaths.size() == 2;
}
},
500);
Assert.assertTrue(result, "Participant should have 2 zk-watchers.");
// check HelixManager#_handlers
// printHandlers(controllerManager);
// printHandlers(participantManagerToExpire);
int controllerHandlerNb = controller.getHandlers().size();
int particHandlerNb = participantManagerToExpire.getHandlers().size();
Assert.assertEquals(
controllerHandlerNb,
9,
"HelixController should have 9 (5+2n) callback handlers for 2 (n) participant");
Assert.assertEquals(
particHandlerNb, 2, "HelixParticipant should have 2 (msg+cur-state) callback handlers");
// expire the session of participant
System.out.println("Expiring participant session...");
String oldSessionId = participantManagerToExpire.getSessionId();
ZkTestHelper.expireSession(participantManagerToExpire.getZkClient());
String newSessionId = participantManagerToExpire.getSessionId();
System.out.println(
"Expried participant session. oldSessionId: "
+ oldSessionId
+ ", newSessionId: "
+ newSessionId);
result =
ClusterStateVerifier.verifyByPolling(
new ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
// check controller zk-watchers
result =
TestHelper.verify(
new TestHelper.Verifier() {
@Override
public boolean verify() throws Exception {
Map<String, Set<String>> watchers = ZkTestHelper.getListenersBySession(ZK_ADDR);
Set<String> watchPaths = watchers.get("0x" + controller.getSessionId());
// System.out.println("controller watch paths after session expiry: " + watchPaths);
// controller should have 5 + 2n + m + (m+2)n zk-watchers
// where n is number of nodes and m is number of resources
return watchPaths.size() == (6 + 5 * n);
}
},
500);
Assert.assertTrue(result, "Controller should have 6 + 5*n zk-watchers after session expiry.");
// check participant zk-watchers
result =
TestHelper.verify(
new TestHelper.Verifier() {
@Override
public boolean verify() throws Exception {
Map<String, Set<String>> watchers = ZkTestHelper.getListenersBySession(ZK_ADDR);
Set<String> watchPaths =
watchers.get("0x" + participantManagerToExpire.getSessionId());
// System.out.println("participant watch paths after session expiry: " +
// watchPaths);
// participant should have 2 zk-watchers: 1 for MESSAGE and 1 for CONTROLLER
return watchPaths.size() == 2;
}
},
500);
Assert.assertTrue(result, "Participant should have 2 zk-watchers after session expiry.");
// check handlers
// printHandlers(controllerManager);
// printHandlers(participantManagerToExpire);
int handlerNb = controller.getHandlers().size();
Assert.assertEquals(
handlerNb,
controllerHandlerNb,
"controller callback handlers should not increase after participant session expiry");
handlerNb = participantManagerToExpire.getHandlers().size();
Assert.assertEquals(
handlerNb,
particHandlerNb,
"participant callback handlers should not increase after participant session expiry");
// clean up
controller.syncStop();
for (int i = 0; i < n; i++) {
participants[i].syncStop();
}
System.out.println("END " + clusterName + " at " + new Date(System.currentTimeMillis()));
}
@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];
TestHelper.setupCluster(
clusterName,
ZK_ADDR,
12918, // participant port
"localhost", // participant name prefix
"TestDB", // resource name prefix
1, // resources
10, // partitions per resource
n, // number of nodes
3, // replicas
"MasterSlave",
true); // do rebalance
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 (totalMsgs == 48) // partition# x replicas
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);
// register "TestDB1_Factory" state model factory
// 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()));
}