use of org.apache.helix.tools.ClusterStateVerifier.BestPossAndExtViewZkVerifier in project helix by apache.
the class TestDistributedClusterController method testDistributedClusterController.
@Test
public void testDistributedClusterController() throws Exception {
String className = TestHelper.getTestClassName();
String methodName = TestHelper.getTestMethodName();
String clusterNamePrefix = className + "_" + methodName;
final int n = 5;
final int clusterNb = 10;
System.out.println("START " + clusterNamePrefix + " at " + new Date(System.currentTimeMillis()));
// setup 10 clusters
for (int i = 0; i < clusterNb; i++) {
String clusterName = clusterNamePrefix + "0_" + i;
String participantName = "localhost" + i;
String resourceName = "TestDB" + i;
// participant port
TestHelper.setupCluster(// participant port
clusterName, // participant port
ZK_ADDR, // participant port
12918, // participant name prefix
participantName, // resource name prefix
resourceName, // resources
1, // partitions per resource
8, // number of nodes
n, // replicas
3, "MasterSlave", // do rebalance
true);
}
// setup controller cluster
final String controllerClusterName = "CONTROLLER_" + clusterNamePrefix;
// controller
TestHelper.setupCluster(// controller
"CONTROLLER_" + clusterNamePrefix, // controller
ZK_ADDR, // controller
0, // participant name prefix
"controller", // resource name prefix
clusterNamePrefix, // resources
1, // partitions per resource
clusterNb, // number of nodes
n, // replicas
3, "LeaderStandby", // do rebalance
true);
// start distributed cluster controllers
ClusterDistributedController[] controllers = new ClusterDistributedController[n];
for (int i = 0; i < n; i++) {
controllers[i] = new ClusterDistributedController(ZK_ADDR, controllerClusterName, "controller_" + i);
controllers[i].syncStart();
}
boolean result = ClusterStateVerifier.verifyByZkCallback(new ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR, controllerClusterName), 30000);
Assert.assertTrue(result, "Controller cluster NOT in ideal state");
// start first cluster
MockParticipantManager[] participants = new MockParticipantManager[n];
final String firstClusterName = clusterNamePrefix + "0_0";
for (int i = 0; i < n; i++) {
String instanceName = "localhost0_" + (12918 + i);
participants[i] = new MockParticipantManager(ZK_ADDR, firstClusterName, instanceName);
participants[i].syncStart();
}
result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, firstClusterName));
Assert.assertTrue(result, "first cluster NOT in ideal state");
// stop current leader in controller cluster
ZkBaseDataAccessor<ZNRecord> baseAccessor = new ZkBaseDataAccessor<ZNRecord>(_gZkClient);
ZKHelixDataAccessor accessor = new ZKHelixDataAccessor(controllerClusterName, baseAccessor);
Builder keyBuilder = accessor.keyBuilder();
LiveInstance leader = accessor.getProperty(keyBuilder.controllerLeader());
String leaderName = leader.getId();
int j = Integer.parseInt(leaderName.substring(leaderName.lastIndexOf('_') + 1));
controllers[j].syncStop();
// setup the second cluster
MockParticipantManager[] participants2 = new MockParticipantManager[n];
final String secondClusterName = clusterNamePrefix + "0_1";
for (int i = 0; i < n; i++) {
String instanceName = "localhost1_" + (12918 + i);
participants2[i] = new MockParticipantManager(ZK_ADDR, secondClusterName, instanceName);
participants2[i].syncStart();
}
result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, secondClusterName));
Assert.assertTrue(result, "second cluster NOT in ideal state");
// clean up
// wait for all zk callbacks done
System.out.println("Cleaning up...");
for (int i = 0; i < 5; i++) {
result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, controllerClusterName));
controllers[i].syncStop();
}
for (int i = 0; i < 5; i++) {
participants[i].syncStop();
}
System.out.println("END " + clusterNamePrefix + " at " + new Date(System.currentTimeMillis()));
}
use of org.apache.helix.tools.ClusterStateVerifier.BestPossAndExtViewZkVerifier in project helix by apache.
the class TestReelectedPipelineCorrectness method testReelection.
@Test
public void testReelection() throws Exception {
final int NUM_CONTROLLERS = 2;
final int NUM_PARTICIPANTS = 4;
final int NUM_PARTITIONS = 8;
final int NUM_REPLICAS = 2;
String className = TestHelper.getTestClassName();
String methodName = TestHelper.getTestMethodName();
String clusterName = className + "_" + methodName;
System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
ClusterSetup setupTool = new ClusterSetup(ZK_ADDR);
// 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, "MasterSlave", RebalanceMode.FULL_AUTO, // do rebalance
true);
// configure distributed controllers
String controllerCluster = clusterName + "_controllers";
setupTool.addCluster(controllerCluster, true);
for (int i = 0; i < NUM_CONTROLLERS; i++) {
setupTool.addInstanceToCluster(controllerCluster, "controller_" + i);
}
setupTool.activateCluster(clusterName, controllerCluster, true);
// start participants
MockParticipantManager[] participants = new MockParticipantManager[NUM_PARTICIPANTS];
for (int i = 0; i < NUM_PARTICIPANTS; i++) {
final String instanceName = "localhost_" + (12918 + i);
participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
participants[i].syncStart();
}
// start controllers
ClusterDistributedController[] controllers = new ClusterDistributedController[NUM_CONTROLLERS];
for (int i = 0; i < NUM_CONTROLLERS; i++) {
controllers[i] = new ClusterDistributedController(ZK_ADDR, controllerCluster, "controller_" + i);
controllers[i].syncStart();
}
Thread.sleep(1000);
// Ensure a balanced cluster
boolean result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
// Disable the leader, resulting in a leader election
HelixDataAccessor accessor = participants[0].getHelixDataAccessor();
LiveInstance leader = accessor.getProperty(accessor.keyBuilder().controllerLeader());
int totalWait = 0;
while (leader == null && totalWait < CHECK_TIMEOUT) {
Thread.sleep(CHECK_INTERVAL);
totalWait += CHECK_INTERVAL;
leader = accessor.getProperty(accessor.keyBuilder().controllerLeader());
}
if (totalWait >= CHECK_TIMEOUT) {
Assert.fail("No leader was ever elected!");
}
String leaderId = leader.getId();
String standbyId = (leaderId.equals("controller_0")) ? "controller_1" : "controller_0";
HelixAdmin admin = setupTool.getClusterManagementTool();
admin.enableInstance(controllerCluster, leaderId, false);
// Stop a participant to make sure that the leader election worked
Thread.sleep(500);
participants[0].syncStop();
Thread.sleep(500);
result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
// Disable the original standby (leaving 0 active controllers) and kill another participant
admin.enableInstance(controllerCluster, standbyId, false);
Thread.sleep(500);
participants[1].syncStop();
// Also change the ideal state
IdealState idealState = admin.getResourceIdealState(clusterName, "TestDB0");
idealState.setMaxPartitionsPerInstance(1);
admin.setResourceIdealState(clusterName, "TestDB0", idealState);
Thread.sleep(500);
// Also disable an instance in the main cluster
admin.enableInstance(clusterName, "localhost_12920", false);
// Re-enable the original leader
admin.enableInstance(controllerCluster, leaderId, true);
// Now check that both the ideal state and the live instances are adhered to by the rebalance
Thread.sleep(500);
result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
// cleanup
for (int i = 0; i < NUM_CONTROLLERS; i++) {
controllers[i].syncStop();
}
for (int i = 2; i < NUM_PARTICIPANTS; i++) {
participants[i].syncStop();
}
System.out.println("STOP " + clusterName + " at " + new Date(System.currentTimeMillis()));
}
use of org.apache.helix.tools.ClusterStateVerifier.BestPossAndExtViewZkVerifier in project helix by apache.
the class TestSchemataSM method testSchemataSM.
@Test
public void testSchemataSM() throws Exception {
String className = TestHelper.getTestClassName();
String methodName = TestHelper.getTestMethodName();
String clusterName = className + "_" + methodName;
int n = 5;
MockParticipantManager[] participants = new MockParticipantManager[n];
System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
// participant start port
TestHelper.setupCluster(// participant start port
clusterName, // participant start port
ZK_ADDR, // participant start port
12918, // participant name prefix
"localhost", // resource name prefix
"TestSchemata", // resources
1, // partitions per resource
1, // number of nodes
n, // replicas
0, "STORAGE_DEFAULT_SM_SCHEMATA", // don't rebalance
false);
// rebalance ideal-state to use ANY_LIVEINSTANCE for preference list
ZKHelixDataAccessor accessor = new ZKHelixDataAccessor(clusterName, new ZkBaseDataAccessor<ZNRecord>(_gZkClient));
PropertyKey.Builder keyBuilder = accessor.keyBuilder();
PropertyKey key = keyBuilder.idealStates("TestSchemata0");
IdealState idealState = accessor.getProperty(key);
idealState.setReplicas(IdealState.IdealStateConstants.ANY_LIVEINSTANCE.toString());
idealState.getRecord().setListField("TestSchemata0_0", Arrays.asList(IdealState.IdealStateConstants.ANY_LIVEINSTANCE.toString()));
accessor.setProperty(key, idealState);
ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller");
controller.syncStart();
// start n-1 participants
for (int i = 1; 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);
// start the remaining 1 participant
participants[0] = new MockParticipantManager(ZK_ADDR, clusterName, "localhost_12918");
participants[0].syncStart();
// make sure we have all participants in MASTER state
result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
key = keyBuilder.externalView("TestSchemata0");
ExternalView externalView = accessor.getProperty(key);
Map<String, String> stateMap = externalView.getStateMap("TestSchemata0_0");
Assert.assertNotNull(stateMap);
Assert.assertEquals(stateMap.size(), n, "all " + n + " participants should be in Master state");
for (int i = 0; i < n; i++) {
String instanceName = "localhost_" + (12918 + i);
Assert.assertNotNull(stateMap.get(instanceName));
Assert.assertEquals(stateMap.get(instanceName), "MASTER");
}
// clean up
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 TestNullReplica method testNullReplica.
@Test
public void testNullReplica() throws Exception {
// Logger.getRootLogger().setLevel(Level.INFO);
String className = TestHelper.getTestClassName();
String methodName = TestHelper.getTestMethodName();
String clusterName = className + "_" + methodName;
System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
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);
// set replica in ideal state to null
String idealStatePath = PropertyPathBuilder.idealState(clusterName, "TestDB0");
ZNRecord idealState = _gZkClient.readData(idealStatePath);
idealState.getSimpleFields().remove(IdealState.IdealStateProperty.REPLICAS.toString());
_gZkClient.writeData(idealStatePath, idealState);
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 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 " + clusterName + " at " + new Date(System.currentTimeMillis()));
}
use of org.apache.helix.tools.ClusterStateVerifier.BestPossAndExtViewZkVerifier in project helix by apache.
the class TestPauseSignal method testPauseSignal.
@Test()
public void testPauseSignal() throws Exception {
// Logger.getRootLogger().setLevel(Level.INFO);
String className = TestHelper.getTestClassName();
String methodName = TestHelper.getTestMethodName();
final String clusterName = className + "_" + methodName;
System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
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);
// start controller
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 BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
Assert.assertTrue(result);
// pause the cluster and make sure pause is persistent
ZkClient zkClient = new ZkClient(ZK_ADDR);
zkClient.setZkSerializer(new ZNRecordSerializer());
final HelixDataAccessor tmpAccessor = new ZKHelixDataAccessor(clusterName, new ZkBaseDataAccessor<ZNRecord>(zkClient));
String cmd = "-zkSvr " + ZK_ADDR + " -enableCluster " + clusterName + " false";
ClusterSetup.processCommandLineArgs(cmd.split(" "));
tmpAccessor.setProperty(tmpAccessor.keyBuilder().pause(), new PauseSignal("pause"));
zkClient.close();
// wait for controller to be signaled by pause
Thread.sleep(1000);
// add a new resource group
ClusterSetup setupTool = new ClusterSetup(ZK_ADDR);
setupTool.addResourceToCluster(clusterName, "TestDB1", 10, "MasterSlave");
setupTool.rebalanceStorageCluster(clusterName, "TestDB1", 3);
// make sure TestDB1 external view is empty
TestHelper.verifyWithTimeout("verifyEmptyCurStateAndExtView", 1000, clusterName, "TestDB1", TestHelper.<String>setOf("localhost_12918", "localhost_12919", "localhost_12920", "localhost_12921", "localhost_12922"), ZK_ADDR);
// resume controller
final HelixDataAccessor accessor = new ZKHelixDataAccessor(clusterName, new ZkBaseDataAccessor<ZNRecord>(_gZkClient));
cmd = "-zkSvr " + ZK_ADDR + " -enableCluster " + clusterName + " true";
ClusterSetup.processCommandLineArgs(cmd.split(" "));
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 " + clusterName + " at " + new Date(System.currentTimeMillis()));
}
Aggregations