/** Decommissioning using a post-configured exclude hosts file */
@Test
public void testAddNewExcludePathToConfiguration() throws Exception {
Configuration conf = new Configuration();
rm = new MockRM(conf);
rm.start();
MockNM nm1 = rm.registerNode("host1:1234", 5120);
MockNM nm2 = rm.registerNode("host2:5678", 10240);
ClusterMetrics metrics = ClusterMetrics.getMetrics();
assert (metrics != null);
int initialMetricCount = metrics.getNumDecommisionedNMs();
NodeHeartbeatResponse nodeHeartbeat = nm1.nodeHeartbeat(true);
Assert.assertEquals(NodeAction.NORMAL, nodeHeartbeat.getNodeAction());
nodeHeartbeat = nm2.nodeHeartbeat(true);
Assert.assertEquals(NodeAction.NORMAL, nodeHeartbeat.getNodeAction());
writeToHostsFile("host2");
conf.set(YarnConfiguration.RM_NODES_EXCLUDE_FILE_PATH, hostFile.getAbsolutePath());
rm.getNodesListManager().refreshNodes(conf);
checkDecommissionedNMCount(rm, ++initialMetricCount);
nodeHeartbeat = nm1.nodeHeartbeat(true);
Assert.assertEquals(
"Node should not have been decomissioned.",
NodeAction.NORMAL,
nodeHeartbeat.getNodeAction());
nodeHeartbeat = nm2.nodeHeartbeat(true);
Assert.assertEquals(
"Node should have been decomissioned but is in state" + nodeHeartbeat.getNodeAction(),
NodeAction.SHUTDOWN,
nodeHeartbeat.getNodeAction());
}
@Test
public void testNodeHeartBeatWithInvalidLabels() throws Exception {
writeToHostsFile("host2");
Configuration conf = new Configuration();
conf.set(YarnConfiguration.RM_NODES_INCLUDE_FILE_PATH, hostFile.getAbsolutePath());
conf.set(
YarnConfiguration.NODELABEL_CONFIGURATION_TYPE,
YarnConfiguration.DISTRIBUTED_NODELABEL_CONFIGURATION_TYPE);
final RMNodeLabelsManager nodeLabelsMgr = new NullRMNodeLabelsManager();
rm =
new MockRM(conf) {
@Override
protected RMNodeLabelsManager createNodeLabelManager() {
return nodeLabelsMgr;
}
};
rm.start();
try {
nodeLabelsMgr.addToCluserNodeLabelsWithDefaultExclusivity(toSet("A", "B", "C"));
} catch (IOException e) {
Assert.fail("Caught Exception while intializing");
e.printStackTrace();
}
ResourceTrackerService resourceTrackerService = rm.getResourceTrackerService();
RegisterNodeManagerRequest registerReq = Records.newRecord(RegisterNodeManagerRequest.class);
NodeId nodeId = NodeId.newInstance("host2", 1234);
Resource capability = BuilderUtils.newResource(1024, 1);
registerReq.setResource(capability);
registerReq.setNodeId(nodeId);
registerReq.setHttpPort(1234);
registerReq.setNMVersion(YarnVersionInfo.getVersion());
registerReq.setNodeLabels(toNodeLabelSet("A"));
RegisterNodeManagerResponse registerResponse =
resourceTrackerService.registerNodeManager(registerReq);
NodeHeartbeatRequest heartbeatReq = Records.newRecord(NodeHeartbeatRequest.class);
heartbeatReq.setNodeLabels(toNodeLabelSet("B", "#C")); // Invalid heart beat labels
heartbeatReq.setNodeStatus(getNodeStatusObject(nodeId));
heartbeatReq.setLastKnownNMTokenMasterKey(registerResponse.getNMTokenMasterKey());
heartbeatReq.setLastKnownContainerTokenMasterKey(registerResponse.getContainerTokenMasterKey());
NodeHeartbeatResponse nodeHeartbeatResponse =
resourceTrackerService.nodeHeartbeat(heartbeatReq);
// response should be NORMAL when RM heartbeat labels are rejected
Assert.assertEquals(
"Response should be NORMAL when RM heartbeat labels" + " are rejected",
NodeAction.NORMAL,
nodeHeartbeatResponse.getNodeAction());
Assert.assertFalse(nodeHeartbeatResponse.getAreNodeLabelsAcceptedByRM());
Assert.assertNotNull(nodeHeartbeatResponse.getDiagnosticsMessage());
rm.stop();
}
@Test
public void testReboot() throws Exception {
Configuration conf = new Configuration();
rm = new MockRM(conf);
rm.start();
MockNM nm1 = rm.registerNode("host1:1234", 5120);
MockNM nm2 = rm.registerNode("host2:1234", 2048);
int initialMetricCount = ClusterMetrics.getMetrics().getNumRebootedNMs();
NodeHeartbeatResponse nodeHeartbeat = nm1.nodeHeartbeat(true);
Assert.assertTrue(NodeAction.NORMAL.equals(nodeHeartbeat.getNodeAction()));
nodeHeartbeat =
nm2.nodeHeartbeat(new HashMap<ApplicationId, List<ContainerStatus>>(), true, -100);
Assert.assertTrue(NodeAction.RESYNC.equals(nodeHeartbeat.getNodeAction()));
Assert.assertEquals(
"Too far behind rm response id:0 nm response id:-100",
nodeHeartbeat.getDiagnosticsMessage());
checkRebootedNMCount(rm, ++initialMetricCount);
}
@Test
public void testNodeHeartbeatWithCentralLabelConfig() throws Exception {
writeToHostsFile("host2");
Configuration conf = new Configuration();
conf.set(YarnConfiguration.RM_NODES_INCLUDE_FILE_PATH, hostFile.getAbsolutePath());
conf.set(
YarnConfiguration.NODELABEL_CONFIGURATION_TYPE,
YarnConfiguration.DEFAULT_NODELABEL_CONFIGURATION_TYPE);
final RMNodeLabelsManager nodeLabelsMgr = new NullRMNodeLabelsManager();
rm =
new MockRM(conf) {
@Override
protected RMNodeLabelsManager createNodeLabelManager() {
return nodeLabelsMgr;
}
};
rm.start();
ResourceTrackerService resourceTrackerService = rm.getResourceTrackerService();
RegisterNodeManagerRequest req = Records.newRecord(RegisterNodeManagerRequest.class);
NodeId nodeId = NodeId.newInstance("host2", 1234);
Resource capability = BuilderUtils.newResource(1024, 1);
req.setResource(capability);
req.setNodeId(nodeId);
req.setHttpPort(1234);
req.setNMVersion(YarnVersionInfo.getVersion());
req.setNodeLabels(toNodeLabelSet("A", "B", "C"));
RegisterNodeManagerResponse registerResponse = resourceTrackerService.registerNodeManager(req);
NodeHeartbeatRequest heartbeatReq = Records.newRecord(NodeHeartbeatRequest.class);
heartbeatReq.setNodeLabels(toNodeLabelSet("B")); // Valid heart beat labels
heartbeatReq.setNodeStatus(getNodeStatusObject(nodeId));
heartbeatReq.setLastKnownNMTokenMasterKey(registerResponse.getNMTokenMasterKey());
heartbeatReq.setLastKnownContainerTokenMasterKey(registerResponse.getContainerTokenMasterKey());
NodeHeartbeatResponse nodeHeartbeatResponse =
resourceTrackerService.nodeHeartbeat(heartbeatReq);
// response should be ok but the RMacceptNodeLabelsUpdate should be false
Assert.assertEquals(NodeAction.NORMAL, nodeHeartbeatResponse.getNodeAction());
// no change in the labels,
Assert.assertNull(nodeLabelsMgr.getNodeLabels().get(nodeId));
// heartbeat labels rejected
Assert.assertFalse(
"Invalid Node Labels should not accepted by RM",
nodeHeartbeatResponse.getAreNodeLabelsAcceptedByRM());
if (rm != null) {
rm.stop();
}
}
@Test
public void testNMUnregistration() throws Exception {
Configuration conf = new Configuration();
rm = new MockRM(conf);
rm.start();
ResourceTrackerService resourceTrackerService = rm.getResourceTrackerService();
MockNM nm1 = rm.registerNode("host1:1234", 5120);
int shutdownNMsCount = ClusterMetrics.getMetrics().getNumShutdownNMs();
NodeHeartbeatResponse nodeHeartbeat = nm1.nodeHeartbeat(true);
Assert.assertTrue(NodeAction.NORMAL.equals(nodeHeartbeat.getNodeAction()));
UnRegisterNodeManagerRequest request = Records.newRecord(UnRegisterNodeManagerRequest.class);
request.setNodeId(nm1.getNodeId());
resourceTrackerService.unRegisterNodeManager(request);
checkShutdownNMCount(rm, ++shutdownNMsCount);
// The RM should remove the node after unregistration, hence send a reboot
// command.
nodeHeartbeat = nm1.nodeHeartbeat(true);
Assert.assertTrue(NodeAction.RESYNC.equals(nodeHeartbeat.getNodeAction()));
}
@Test
public void testInvalidNMUnregistration() throws Exception {
Configuration conf = new Configuration();
rm = new MockRM(conf);
rm.start();
ResourceTrackerService resourceTrackerService = rm.getResourceTrackerService();
int shutdownNMsCount = ClusterMetrics.getMetrics().getNumShutdownNMs();
int decommisionedNMsCount = ClusterMetrics.getMetrics().getNumDecommisionedNMs();
// Node not found for unregister
UnRegisterNodeManagerRequest request = Records.newRecord(UnRegisterNodeManagerRequest.class);
request.setNodeId(BuilderUtils.newNodeId("host", 1234));
resourceTrackerService.unRegisterNodeManager(request);
checkShutdownNMCount(rm, 0);
checkDecommissionedNMCount(rm, 0);
// 1. Register the Node Manager
// 2. Exclude the same Node Manager host
// 3. Give NM heartbeat to RM
// 4. Unregister the Node Manager
MockNM nm1 = new MockNM("host1:1234", 5120, resourceTrackerService);
RegisterNodeManagerResponse response = nm1.registerNode();
Assert.assertEquals(NodeAction.NORMAL, response.getNodeAction());
writeToHostsFile("host2");
conf.set(YarnConfiguration.RM_NODES_INCLUDE_FILE_PATH, hostFile.getAbsolutePath());
rm.getNodesListManager().refreshNodes(conf);
NodeHeartbeatResponse heartbeatResponse = nm1.nodeHeartbeat(true);
Assert.assertEquals(NodeAction.SHUTDOWN, heartbeatResponse.getNodeAction());
checkShutdownNMCount(rm, shutdownNMsCount);
checkDecommissionedNMCount(rm, ++decommisionedNMsCount);
request.setNodeId(nm1.getNodeId());
resourceTrackerService.unRegisterNodeManager(request);
checkShutdownNMCount(rm, shutdownNMsCount);
checkDecommissionedNMCount(rm, decommisionedNMsCount);
// 1. Register the Node Manager
// 2. Exclude the same Node Manager host
// 3. Unregister the Node Manager
MockNM nm2 = new MockNM("host2:1234", 5120, resourceTrackerService);
RegisterNodeManagerResponse response2 = nm2.registerNode();
Assert.assertEquals(NodeAction.NORMAL, response2.getNodeAction());
writeToHostsFile("host1");
conf.set(YarnConfiguration.RM_NODES_INCLUDE_FILE_PATH, hostFile.getAbsolutePath());
rm.getNodesListManager().refreshNodes(conf);
request.setNodeId(nm2.getNodeId());
resourceTrackerService.unRegisterNodeManager(request);
checkShutdownNMCount(rm, shutdownNMsCount);
checkDecommissionedNMCount(rm, ++decommisionedNMsCount);
}
/** Decommissioning using a pre-configured include hosts file */
@Test
public void testDecommissionWithIncludeHosts() throws Exception {
writeToHostsFile("localhost", "host1", "host2");
Configuration conf = new Configuration();
conf.set(YarnConfiguration.RM_NODES_INCLUDE_FILE_PATH, hostFile.getAbsolutePath());
rm = new MockRM(conf);
rm.start();
MockNM nm1 = rm.registerNode("host1:1234", 5120);
MockNM nm2 = rm.registerNode("host2:5678", 10240);
MockNM nm3 = rm.registerNode("localhost:4433", 1024);
ClusterMetrics metrics = ClusterMetrics.getMetrics();
assert (metrics != null);
int metricCount = metrics.getNumDecommisionedNMs();
NodeHeartbeatResponse nodeHeartbeat = nm1.nodeHeartbeat(true);
Assert.assertTrue(NodeAction.NORMAL.equals(nodeHeartbeat.getNodeAction()));
nodeHeartbeat = nm2.nodeHeartbeat(true);
Assert.assertTrue(NodeAction.NORMAL.equals(nodeHeartbeat.getNodeAction()));
nodeHeartbeat = nm3.nodeHeartbeat(true);
Assert.assertTrue(NodeAction.NORMAL.equals(nodeHeartbeat.getNodeAction()));
// To test that IPs also work
String ip = NetUtils.normalizeHostName("localhost");
writeToHostsFile("host1", ip);
rm.getNodesListManager().refreshNodes(conf);
checkDecommissionedNMCount(rm, ++metricCount);
nodeHeartbeat = nm1.nodeHeartbeat(true);
Assert.assertTrue(NodeAction.NORMAL.equals(nodeHeartbeat.getNodeAction()));
Assert.assertEquals(1, ClusterMetrics.getMetrics().getNumDecommisionedNMs());
nodeHeartbeat = nm2.nodeHeartbeat(true);
Assert.assertTrue(
"Node is not decommisioned.", NodeAction.SHUTDOWN.equals(nodeHeartbeat.getNodeAction()));
nodeHeartbeat = nm3.nodeHeartbeat(true);
Assert.assertTrue(NodeAction.NORMAL.equals(nodeHeartbeat.getNodeAction()));
Assert.assertEquals(metricCount, ClusterMetrics.getMetrics().getNumDecommisionedNMs());
}
@Test
public void testReconnectNode() throws Exception {
final DrainDispatcher dispatcher = new DrainDispatcher();
rm =
new MockRM() {
@Override
protected EventHandler<SchedulerEvent> createSchedulerEventDispatcher() {
return new SchedulerEventDispatcher(this.scheduler) {
@Override
public void handle(SchedulerEvent event) {
scheduler.handle(event);
}
};
}
@Override
protected Dispatcher createDispatcher() {
return dispatcher;
}
};
rm.start();
MockNM nm1 = rm.registerNode("host1:1234", 5120);
MockNM nm2 = rm.registerNode("host2:5678", 5120);
nm1.nodeHeartbeat(true);
nm2.nodeHeartbeat(false);
dispatcher.await();
checkUnealthyNMCount(rm, nm2, true, 1);
final int expectedNMs = ClusterMetrics.getMetrics().getNumActiveNMs();
QueueMetrics metrics = rm.getResourceScheduler().getRootQueueMetrics();
// TODO Metrics incorrect in case of the FifoScheduler
Assert.assertEquals(5120, metrics.getAvailableMB());
// reconnect of healthy node
nm1 = rm.registerNode("host1:1234", 5120);
NodeHeartbeatResponse response = nm1.nodeHeartbeat(true);
Assert.assertTrue(NodeAction.NORMAL.equals(response.getNodeAction()));
dispatcher.await();
Assert.assertEquals(expectedNMs, ClusterMetrics.getMetrics().getNumActiveNMs());
checkUnealthyNMCount(rm, nm2, true, 1);
// reconnect of unhealthy node
nm2 = rm.registerNode("host2:5678", 5120);
response = nm2.nodeHeartbeat(false);
Assert.assertTrue(NodeAction.NORMAL.equals(response.getNodeAction()));
dispatcher.await();
Assert.assertEquals(expectedNMs, ClusterMetrics.getMetrics().getNumActiveNMs());
checkUnealthyNMCount(rm, nm2, true, 1);
// unhealthy node changed back to healthy
nm2 = rm.registerNode("host2:5678", 5120);
dispatcher.await();
response = nm2.nodeHeartbeat(true);
response = nm2.nodeHeartbeat(true);
dispatcher.await();
Assert.assertEquals(5120 + 5120, metrics.getAvailableMB());
// reconnect of node with changed capability
nm1 = rm.registerNode("host2:5678", 10240);
dispatcher.await();
response = nm1.nodeHeartbeat(true);
dispatcher.await();
Assert.assertTrue(NodeAction.NORMAL.equals(response.getNodeAction()));
Assert.assertEquals(5120 + 10240, metrics.getAvailableMB());
// reconnect of node with changed capability and running applications
List<ApplicationId> runningApps = new ArrayList<ApplicationId>();
runningApps.add(ApplicationId.newInstance(1, 0));
nm1 = rm.registerNode("host2:5678", 15360, 2, runningApps);
dispatcher.await();
response = nm1.nodeHeartbeat(true);
dispatcher.await();
Assert.assertTrue(NodeAction.NORMAL.equals(response.getNodeAction()));
Assert.assertEquals(5120 + 15360, metrics.getAvailableMB());
// reconnect healthy node changing http port
nm1 = new MockNM("host1:1234", 5120, rm.getResourceTrackerService());
nm1.setHttpPort(3);
nm1.registerNode();
dispatcher.await();
response = nm1.nodeHeartbeat(true);
response = nm1.nodeHeartbeat(true);
dispatcher.await();
RMNode rmNode = rm.getRMContext().getRMNodes().get(nm1.getNodeId());
Assert.assertEquals(3, rmNode.getHttpPort());
Assert.assertEquals(5120, rmNode.getTotalCapability().getMemory());
Assert.assertEquals(5120 + 15360, metrics.getAvailableMB());
}
@Test
public void testNodeHeartBeatWithLabels() throws Exception {
writeToHostsFile("host2");
Configuration conf = new Configuration();
conf.set(YarnConfiguration.RM_NODES_INCLUDE_FILE_PATH, hostFile.getAbsolutePath());
conf.set(
YarnConfiguration.NODELABEL_CONFIGURATION_TYPE,
YarnConfiguration.DISTRIBUTED_NODELABEL_CONFIGURATION_TYPE);
final RMNodeLabelsManager nodeLabelsMgr = new NullRMNodeLabelsManager();
rm =
new MockRM(conf) {
@Override
protected RMNodeLabelsManager createNodeLabelManager() {
return nodeLabelsMgr;
}
};
rm.start();
// adding valid labels
try {
nodeLabelsMgr.addToCluserNodeLabelsWithDefaultExclusivity(toSet("A", "B", "C"));
} catch (IOException e) {
Assert.fail("Caught Exception while intializing");
e.printStackTrace();
}
// Registering of labels and other required info to RM
ResourceTrackerService resourceTrackerService = rm.getResourceTrackerService();
RegisterNodeManagerRequest registerReq = Records.newRecord(RegisterNodeManagerRequest.class);
NodeId nodeId = NodeId.newInstance("host2", 1234);
Resource capability = BuilderUtils.newResource(1024, 1);
registerReq.setResource(capability);
registerReq.setNodeId(nodeId);
registerReq.setHttpPort(1234);
registerReq.setNMVersion(YarnVersionInfo.getVersion());
registerReq.setNodeLabels(toNodeLabelSet("A")); // Node register label
RegisterNodeManagerResponse registerResponse =
resourceTrackerService.registerNodeManager(registerReq);
// modification of labels during heartbeat
NodeHeartbeatRequest heartbeatReq = Records.newRecord(NodeHeartbeatRequest.class);
heartbeatReq.setNodeLabels(toNodeLabelSet("B")); // Node heartbeat label update
NodeStatus nodeStatusObject = getNodeStatusObject(nodeId);
heartbeatReq.setNodeStatus(nodeStatusObject);
heartbeatReq.setLastKnownNMTokenMasterKey(registerResponse.getNMTokenMasterKey());
heartbeatReq.setLastKnownContainerTokenMasterKey(registerResponse.getContainerTokenMasterKey());
NodeHeartbeatResponse nodeHeartbeatResponse =
resourceTrackerService.nodeHeartbeat(heartbeatReq);
Assert.assertEquals(
"InValid Node Labels were not accepted by RM",
NodeAction.NORMAL,
nodeHeartbeatResponse.getNodeAction());
assertCollectionEquals(
nodeLabelsMgr.getNodeLabels().get(nodeId),
ResourceTrackerService.convertToStringSet(heartbeatReq.getNodeLabels()));
Assert.assertTrue(
"Valid Node Labels were not accepted by RM",
nodeHeartbeatResponse.getAreNodeLabelsAcceptedByRM());
// After modification of labels next heartbeat sends null informing no update
Set<String> oldLabels = nodeLabelsMgr.getNodeLabels().get(nodeId);
int responseId = nodeStatusObject.getResponseId();
heartbeatReq = Records.newRecord(NodeHeartbeatRequest.class);
heartbeatReq.setNodeLabels(null); // Node heartbeat label update
nodeStatusObject = getNodeStatusObject(nodeId);
nodeStatusObject.setResponseId(responseId + 2);
heartbeatReq.setNodeStatus(nodeStatusObject);
heartbeatReq.setLastKnownNMTokenMasterKey(registerResponse.getNMTokenMasterKey());
heartbeatReq.setLastKnownContainerTokenMasterKey(registerResponse.getContainerTokenMasterKey());
nodeHeartbeatResponse = resourceTrackerService.nodeHeartbeat(heartbeatReq);
Assert.assertEquals(
"InValid Node Labels were not accepted by RM",
NodeAction.NORMAL,
nodeHeartbeatResponse.getNodeAction());
assertCollectionEquals(nodeLabelsMgr.getNodeLabels().get(nodeId), oldLabels);
Assert.assertFalse(
"Node Labels should not accepted by RM",
nodeHeartbeatResponse.getAreNodeLabelsAcceptedByRM());
rm.stop();
}
/** Decommissioning using a pre-configured exclude hosts file */
@Test
public void testDecommissionWithExcludeHosts() throws Exception {
Configuration conf = new Configuration();
conf.set(YarnConfiguration.RM_NODES_EXCLUDE_FILE_PATH, hostFile.getAbsolutePath());
writeToHostsFile("");
final DrainDispatcher dispatcher = new DrainDispatcher();
rm =
new MockRM(conf) {
@Override
protected Dispatcher createDispatcher() {
return dispatcher;
}
};
rm.start();
MockNM nm1 = rm.registerNode("host1:1234", 5120);
MockNM nm2 = rm.registerNode("host2:5678", 10240);
MockNM nm3 = rm.registerNode("localhost:4433", 1024);
dispatcher.await();
int metricCount = ClusterMetrics.getMetrics().getNumDecommisionedNMs();
NodeHeartbeatResponse nodeHeartbeat = nm1.nodeHeartbeat(true);
Assert.assertTrue(NodeAction.NORMAL.equals(nodeHeartbeat.getNodeAction()));
nodeHeartbeat = nm2.nodeHeartbeat(true);
Assert.assertTrue(NodeAction.NORMAL.equals(nodeHeartbeat.getNodeAction()));
dispatcher.await();
// To test that IPs also work
String ip = NetUtils.normalizeHostName("localhost");
writeToHostsFile("host2", ip);
rm.getNodesListManager().refreshNodes(conf);
checkDecommissionedNMCount(rm, metricCount + 2);
nodeHeartbeat = nm1.nodeHeartbeat(true);
Assert.assertTrue(NodeAction.NORMAL.equals(nodeHeartbeat.getNodeAction()));
nodeHeartbeat = nm2.nodeHeartbeat(true);
Assert.assertTrue(
"The decommisioned metrics are not updated",
NodeAction.SHUTDOWN.equals(nodeHeartbeat.getNodeAction()));
nodeHeartbeat = nm3.nodeHeartbeat(true);
Assert.assertTrue(
"The decommisioned metrics are not updated",
NodeAction.SHUTDOWN.equals(nodeHeartbeat.getNodeAction()));
dispatcher.await();
writeToHostsFile("");
rm.getNodesListManager().refreshNodes(conf);
nm3 = rm.registerNode("localhost:4433", 1024);
dispatcher.await();
nodeHeartbeat = nm3.nodeHeartbeat(true);
dispatcher.await();
Assert.assertTrue(NodeAction.NORMAL.equals(nodeHeartbeat.getNodeAction()));
// decommissined node is 1 since 1 node is rejoined after updating exclude
// file
checkDecommissionedNMCount(rm, metricCount + 1);
}