@Test
public void simpleIntegrationTest() throws Exception {
// Logger.getRootLogger().setLevel(Level.INFO);
String className = TestHelper.getTestClassName();
String methodName = TestHelper.getTestMethodName();
String clusterName = className + "_" + methodName;
int n = 1;
System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
TestHelper.setupCluster(
clusterName,
_zkaddr,
12918, // participant port
"localhost", // participant name prefix
"TestDB", // resource name prefix
1, // resources
4, // partitions per resource
n, // number of nodes
1, // replicas
"MasterSlave",
true); // do rebalance
HelixManager participant =
new ZKHelixManager(clusterName, "localhost_12918", InstanceType.PARTICIPANT, _zkaddr);
participant
.getStateMachineEngine()
.registerStateModelFactory(StateModelDefId.MasterSlave, new MockMSModelFactory());
participant.connect();
HelixManager controller =
new ZKHelixManager(clusterName, "controller_0", InstanceType.CONTROLLER, _zkaddr);
controller.connect();
boolean result =
ClusterStateVerifier.verifyByZkCallback(
new BestPossAndExtViewZkVerifier(_zkaddr, clusterName));
Assert.assertTrue(result);
// cleanup
controller.disconnect();
participant.disconnect();
// verify all live-instances and leader nodes are gone
ZKHelixDataAccessor accessor = new ZKHelixDataAccessor(clusterName, _baseAccessor);
PropertyKey.Builder keyBuilder = accessor.keyBuilder();
Assert.assertNull(accessor.getProperty(keyBuilder.liveInstance("localhost_12918")));
Assert.assertNull(accessor.getProperty(keyBuilder.controllerLeader()));
System.out.println("END " + clusterName + " at " + new Date(System.currentTimeMillis()));
}
public static void addFakeBrokerInstancesToAutoJoinHelixCluster(
String helixClusterName, String zkServer, int numInstances, boolean isSingleTenant)
throws Exception {
for (int i = 0; i < numInstances; ++i) {
final String brokerId = "Broker_localhost_" + i;
final HelixManager helixZkManager =
HelixManagerFactory.getZKHelixManager(
helixClusterName, brokerId, InstanceType.PARTICIPANT, zkServer);
final StateMachineEngine stateMachineEngine = helixZkManager.getStateMachineEngine();
final StateModelFactory<?> stateModelFactory =
new EmptyBrokerOnlineOfflineStateModelFactory();
stateMachineEngine.registerStateModelFactory(
EmptyBrokerOnlineOfflineStateModelFactory.getStateModelDef(), stateModelFactory);
helixZkManager.connect();
if (isSingleTenant) {
helixZkManager
.getClusterManagmentTool()
.addInstanceTag(
helixClusterName,
brokerId,
ControllerTenantNameBuilder.getBrokerTenantNameForTenant(
ControllerTenantNameBuilder.DEFAULT_TENANT_NAME));
} else {
helixZkManager
.getClusterManagmentTool()
.addInstanceTag(helixClusterName, brokerId, UNTAGGED_BROKER_INSTANCE);
}
Thread.sleep(1000);
}
}
public static void addFakeDataInstancesToAutoJoinHelixCluster(
String helixClusterName,
String zkServer,
int numInstances,
boolean isSingleTenant,
int adminPort)
throws Exception {
for (int i = 0; i < numInstances; ++i) {
final String instanceId = "Server_localhost_" + i;
final HelixManager helixZkManager =
HelixManagerFactory.getZKHelixManager(
helixClusterName, instanceId, InstanceType.PARTICIPANT, zkServer);
final StateMachineEngine stateMachineEngine = helixZkManager.getStateMachineEngine();
final StateModelFactory<?> stateModelFactory =
new EmptySegmentOnlineOfflineStateModelFactory();
stateMachineEngine.registerStateModelFactory(
EmptySegmentOnlineOfflineStateModelFactory.getStateModelDef(), stateModelFactory);
helixZkManager.connect();
if (isSingleTenant) {
helixZkManager
.getClusterManagmentTool()
.addInstanceTag(
helixClusterName,
instanceId,
TableNameBuilder.OFFLINE_TABLE_NAME_BUILDER.forTable(
ControllerTenantNameBuilder.DEFAULT_TENANT_NAME));
helixZkManager
.getClusterManagmentTool()
.addInstanceTag(
helixClusterName,
instanceId,
TableNameBuilder.REALTIME_TABLE_NAME_BUILDER.forTable(
ControllerTenantNameBuilder.DEFAULT_TENANT_NAME));
} else {
helixZkManager
.getClusterManagmentTool()
.addInstanceTag(helixClusterName, instanceId, UNTAGGED_SERVER_INSTANCE);
}
HelixConfigScope scope =
new HelixConfigScopeBuilder(
HelixConfigScope.ConfigScopeProperty.PARTICIPANT, helixClusterName)
.forParticipant(instanceId)
.build();
Map<String, String> props = new HashMap<>();
props.put(CommonConstants.Helix.Instance.ADMIN_PORT_KEY, String.valueOf(adminPort + i));
helixZkManager.getClusterManagmentTool().setConfig(scope, props);
}
}
void start() {
try {
participantManager
.getMessagingService()
.registerMessageHandlerFactory(
Message.MessageType.USER_DEFINE_MSG.toString(),
new MessageHandlerResolverWrapper().convert());
participantManager
.getStateMachineEngine()
.registerStateModelFactory("LeaderStandby", new LockTransitionalFactory());
participantManager.connect();
offlinePartition();
} catch (final Exception ex) {
throw new RuntimeException(ex);
}
}
@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 testZKReconnect() throws Exception {
final AtomicReference<ZkServer> zkServerRef = new AtomicReference<ZkServer>();
final int zkPort = TestHelper.getRandomPort();
final String zkAddr = String.format("localhost:%d", zkPort);
ZkServer zkServer = TestHelper.startZkServer(zkAddr);
zkServerRef.set(zkServer);
String className = TestHelper.getTestClassName();
String methodName = TestHelper.getTestMethodName();
String clusterName = className + "_" + methodName;
// Setup cluster
LOG.info("Setup clusters");
ClusterSetup clusterSetup = new ClusterSetup(zkAddr);
clusterSetup.addCluster(clusterName, true);
// Registers and starts controller
LOG.info("Starts controller");
HelixManager controller =
HelixManagerFactory.getZKHelixManager(clusterName, null, InstanceType.CONTROLLER, zkAddr);
controller.connect();
// Registers and starts participant
LOG.info("Starts participant");
String hostname = "localhost";
String instanceId = String.format("%s_%d", hostname, 1);
clusterSetup.addInstanceToCluster(clusterName, instanceId);
HelixManager participant =
HelixManagerFactory.getZKHelixManager(
clusterName, instanceId, InstanceType.PARTICIPANT, zkAddr);
participant.connect();
LOG.info("Register state machine");
final CountDownLatch latch = new CountDownLatch(1);
participant
.getStateMachineEngine()
.registerStateModelFactory(
"OnlineOffline",
new StateModelFactory<StateModel>() {
@Override
public StateModel createNewStateModel(String stateUnitKey) {
return new SimpleStateModel(latch);
}
},
"test");
String resourceName = "test-resource";
LOG.info("Ideal state assignment");
HelixAdmin helixAdmin = participant.getClusterManagmentTool();
helixAdmin.addResource(
clusterName,
resourceName,
1,
"OnlineOffline",
IdealState.RebalanceMode.CUSTOMIZED.toString());
IdealState idealState = helixAdmin.getResourceIdealState(clusterName, resourceName);
idealState.setReplicas("1");
idealState.setStateModelFactoryName("test");
idealState.setPartitionState(resourceName + "_0", instanceId, "ONLINE");
LOG.info("Shutdown ZK server");
TestHelper.stopZkServer(zkServerRef.get());
Executors.newSingleThreadScheduledExecutor()
.schedule(
new Runnable() {
@Override
public void run() {
try {
LOG.info("Restart ZK server");
// zkServer.set(TestUtils.startZookeeper(zkDir, zkPort));
zkServerRef.set(TestHelper.startZkServer(zkAddr, null, false));
} catch (Exception e) {
LOG.error(e.getMessage(), e);
}
}
},
2L,
TimeUnit.SECONDS);
// future.get();
LOG.info("Before update ideal state");
helixAdmin.setResourceIdealState(clusterName, resourceName, idealState);
LOG.info("After update ideal state");
LOG.info("Wait for OFFLINE->ONLINE state transition");
try {
Assert.assertTrue(latch.await(10, TimeUnit.SECONDS));
// wait until stable state
boolean result =
ClusterStateVerifier.verifyByZkCallback(
new BestPossAndExtViewZkVerifier(zkAddr, clusterName));
Assert.assertTrue(result);
} finally {
participant.disconnect();
zkServerRef.get().shutdown();
}
}
public HelixBrokerStarter(
String helixClusterName, String zkServer, Configuration pinotHelixProperties)
throws Exception {
_liveInstancesListener = new LiveInstancesChangeListenerImpl(helixClusterName);
_pinotHelixProperties = DefaultHelixBrokerConfig.getDefaultBrokerConf(pinotHelixProperties);
final String brokerId =
_pinotHelixProperties.getString(
"instanceId",
CommonConstants.Helix.PREFIX_OF_BROKER_INSTANCE
+ NetUtil.getHostAddress()
+ "_"
+ _pinotHelixProperties.getInt(
CommonConstants.Helix.KEY_OF_BROKER_QUERY_PORT,
CommonConstants.Helix.DEFAULT_BROKER_QUERY_PORT));
_pinotHelixProperties.addProperty("pinot.broker.id", brokerId);
RoutingTableBuilder defaultOfflineRoutingTableBuilder =
getRoutingTableBuilder(
_pinotHelixProperties.subset(DEFAULT_OFFLINE_ROUTING_TABLE_BUILDER_KEY));
RoutingTableBuilder defaultRealtimeRoutingTableBuilder =
getRoutingTableBuilder(
_pinotHelixProperties.subset(DEFAULT_REALTIME_ROUTING_TABLE_BUILDER_KEY));
Map<String, RoutingTableBuilder> tableToRoutingTableBuilderMap =
getTableToRoutingTableBuilderMap(_pinotHelixProperties.subset(ROUTING_TABLE_BUILDER_KEY));
ZkClient zkClient =
new ZkClient(
StringUtil.join(
"/", StringUtils.chomp(zkServer, "/"), helixClusterName, "PROPERTYSTORE"),
ZkClient.DEFAULT_SESSION_TIMEOUT,
ZkClient.DEFAULT_CONNECTION_TIMEOUT,
new ZNRecordSerializer());
_propertyStore =
new ZkHelixPropertyStore<ZNRecord>(new ZkBaseDataAccessor<ZNRecord>(zkClient), "/", null);
_helixExternalViewBasedRouting =
new HelixExternalViewBasedRouting(
defaultOfflineRoutingTableBuilder,
defaultRealtimeRoutingTableBuilder,
tableToRoutingTableBuilderMap,
_propertyStore);
// _brokerServerBuilder = startBroker();
_brokerServerBuilder = startBroker(_pinotHelixProperties);
_helixManager =
HelixManagerFactory.getZKHelixManager(
helixClusterName, brokerId, InstanceType.PARTICIPANT, zkServer);
final StateMachineEngine stateMachineEngine = _helixManager.getStateMachineEngine();
final StateModelFactory<?> stateModelFactory =
new BrokerResourceOnlineOfflineStateModelFactory(
_helixManager, _helixExternalViewBasedRouting);
stateMachineEngine.registerStateModelFactory(
BrokerResourceOnlineOfflineStateModelFactory.getStateModelDef(), stateModelFactory);
_helixManager.connect();
_helixAdmin = _helixManager.getClusterManagmentTool();
_helixBrokerRoutingTable =
new HelixBrokerRoutingTable(_helixExternalViewBasedRouting, brokerId, _helixManager);
addInstanceTagIfNeeded(helixClusterName, brokerId);
_helixManager.addExternalViewChangeListener(_helixBrokerRoutingTable);
_helixManager.addInstanceConfigChangeListener(_helixBrokerRoutingTable);
_helixManager.addLiveInstanceChangeListener(_liveInstancesListener);
}