/**
* Get all the nodes in the cluster, this method generate RPC
*
* @return host names
* @throws YarnException
*/
private List<String> getClusterNodes() throws YarnException {
List<String> result = new ArrayList<String>();
GetClusterNodesRequest clusterNodesReq = Records.newRecord(GetClusterNodesRequest.class);
try {
GetClusterNodesResponse clusterNodesResp =
applicationsManager.getClusterNodes(clusterNodesReq);
List<NodeReport> nodeReports = clusterNodesResp.getNodeReports();
for (NodeReport nodeReport : nodeReports) {
result.add(nodeReport.getNodeId().getHost());
}
} catch (IOException e) {
LOG.error("error getting cluster nodes from AM");
throw new YarnException(e);
}
return result;
}
public static NodeReport newNodeReport(
NodeId nodeId,
NodeState nodeState,
String httpAddress,
String rackName,
Resource used,
Resource capability,
int numContainers,
String healthReport,
long lastHealthReportTime) {
NodeReport nodeReport = recordFactory.newRecordInstance(NodeReport.class);
nodeReport.setNodeId(nodeId);
nodeReport.setNodeState(nodeState);
nodeReport.setHttpAddress(httpAddress);
nodeReport.setRackName(rackName);
nodeReport.setUsed(used);
nodeReport.setCapability(capability);
nodeReport.setNumContainers(numContainers);
nodeReport.setHealthReport(healthReport);
nodeReport.setLastHealthReportTime(lastHealthReportTime);
return nodeReport;
}
@Test
public void testSetupShutdown() throws Exception {
GetClusterNodesRequest request = Records.newRecord(GetClusterNodesRequest.class);
ClientRMService clientRMService = yarnCluster.getResourceManager().getClientRMService();
GetClusterNodesResponse response = clientRMService.getClusterNodes(request);
List<NodeReport> nodeReports = response.getNodeReports();
LOG.info("{}", nodeReports);
for (NodeReport nr : nodeReports) {
LOG.info("Node: {}", nr.getNodeId());
LOG.info("Total memory: {}", nr.getCapability());
LOG.info("Used memory: {}", nr.getUsed());
LOG.info("Number containers: {}", nr.getNumContainers());
}
String appMasterJar = JarFinder.getJar(StreamingAppMaster.class);
LOG.info("appmaster jar: " + appMasterJar);
String testJar = JarFinder.getJar(StramMiniClusterTest.class);
LOG.info("testJar: " + testJar);
// create test application
Properties dagProps = new Properties();
// input module (ensure shutdown works while windows are generated)
dagProps.put(
StreamingApplication.DT_PREFIX + "operator.numGen.classname",
TestGeneratorInputOperator.class.getName());
dagProps.put(StreamingApplication.DT_PREFIX + "operator.numGen.maxTuples", "1");
// fake output adapter - to be ignored when determine shutdown
// props.put(DAGContext.DT_PREFIX + "stream.output.classname",
// HDFSOutputStream.class.getName());
// props.put(DAGContext.DT_PREFIX + "stream.output.inputNode", "module2");
// props.put(DAGContext.DT_PREFIX + "stream.output.filepath",
// "miniclustertest-testSetupShutdown.out");
dagProps.put(
StreamingApplication.DT_PREFIX + "operator.module1.classname",
GenericTestOperator.class.getName());
dagProps.put(
StreamingApplication.DT_PREFIX + "operator.module2.classname",
GenericTestOperator.class.getName());
dagProps.put(StreamingApplication.DT_PREFIX + "stream.fromNumGen.source", "numGen.outport");
dagProps.put(StreamingApplication.DT_PREFIX + "stream.fromNumGen.sinks", "module1.inport1");
dagProps.put(StreamingApplication.DT_PREFIX + "stream.n1n2.source", "module1.outport1");
dagProps.put(StreamingApplication.DT_PREFIX + "stream.n1n2.sinks", "module2.inport1");
dagProps.setProperty(
StreamingApplication.DT_PREFIX + LogicalPlan.MASTER_MEMORY_MB.getName(), "128");
dagProps.setProperty(
StreamingApplication.DT_PREFIX + LogicalPlan.CONTAINER_JVM_OPTIONS.getName(),
"-Dlog4j.properties=custom_log4j.properties");
dagProps.setProperty(
StreamingApplication.DT_PREFIX + "operator.*." + OperatorContext.MEMORY_MB.getName(), "64");
dagProps.setProperty(
StreamingApplication.DT_PREFIX + "operator.*." + OperatorContext.VCORES.getName(), "1");
dagProps.setProperty(
StreamingApplication.DT_PREFIX
+ "operator.*.port.*."
+ Context.PortContext.BUFFER_MEMORY_MB.getName(),
"32");
dagProps.setProperty(StreamingApplication.DT_PREFIX + LogicalPlan.DEBUG.getName(), "true");
// dagProps.setProperty(StreamingApplication.DT_PREFIX +
// LogicalPlan.CONTAINERS_MAX_COUNT.getName(), "2");
LOG.info("dag properties: {}", dagProps);
LOG.info("Initializing Client");
LogicalPlanConfiguration tb = new LogicalPlanConfiguration(conf);
tb.addFromProperties(dagProps, null);
LogicalPlan dag = createDAG(tb);
Configuration yarnConf = new Configuration(yarnCluster.getConfig());
StramClient client = new StramClient(yarnConf, dag);
try {
client.start();
if (StringUtils.isBlank(System.getenv("JAVA_HOME"))) {
client.javaCmd = "java"; // JAVA_HOME not set in the yarn mini cluster
}
LOG.info("Running client");
client.startApplication();
boolean result = client.monitorApplication();
LOG.info("Client run completed. Result=" + result);
Assert.assertTrue(result);
} finally {
client.stop();
}
}
@Test
public void testAMRMUnusableNodes() throws Exception {
MockNM nm1 = rm.registerNode("h1:1234", 5000);
MockNM nm2 = rm.registerNode("h2:1234", 5000);
MockNM nm3 = rm.registerNode("h3:1234", 5000);
MockNM nm4 = rm.registerNode("h4:1234", 5000);
RMApp app1 = rm.submitApp(2000);
// Trigger the scheduling so the AM gets 'launched' on nm1
nm1.nodeHeartbeat(true);
RMAppAttempt attempt1 = app1.getCurrentAppAttempt();
MockAM am1 = rm.sendAMLaunched(attempt1.getAppAttemptId());
// register AM returns no unusable node
am1.registerAppAttempt();
// allocate request returns no updated node
AllocateRequest allocateRequest1 =
BuilderUtils.newAllocateRequest(attempt1.getAppAttemptId(), 0, 0F, null, null);
AMResponse response1 = amService.allocate(allocateRequest1).getAMResponse();
List<NodeReport> updatedNodes = response1.getUpdatedNodes();
Assert.assertEquals(0, updatedNodes.size());
syncNodeHeartbeat(nm4, false);
// allocate request returns updated node
allocateRequest1 =
BuilderUtils.newAllocateRequest(
attempt1.getAppAttemptId(), response1.getResponseId(), 0F, null, null);
response1 = amService.allocate(allocateRequest1).getAMResponse();
updatedNodes = response1.getUpdatedNodes();
Assert.assertEquals(1, updatedNodes.size());
NodeReport nr = updatedNodes.iterator().next();
Assert.assertEquals(nm4.getNodeId(), nr.getNodeId());
Assert.assertEquals(NodeState.UNHEALTHY, nr.getNodeState());
// resending the allocate request returns the same result
response1 = amService.allocate(allocateRequest1).getAMResponse();
updatedNodes = response1.getUpdatedNodes();
Assert.assertEquals(1, updatedNodes.size());
nr = updatedNodes.iterator().next();
Assert.assertEquals(nm4.getNodeId(), nr.getNodeId());
Assert.assertEquals(NodeState.UNHEALTHY, nr.getNodeState());
syncNodeLost(nm3);
// subsequent allocate request returns delta
allocateRequest1 =
BuilderUtils.newAllocateRequest(
attempt1.getAppAttemptId(), response1.getResponseId(), 0F, null, null);
response1 = amService.allocate(allocateRequest1).getAMResponse();
updatedNodes = response1.getUpdatedNodes();
Assert.assertEquals(1, updatedNodes.size());
nr = updatedNodes.iterator().next();
Assert.assertEquals(nm3.getNodeId(), nr.getNodeId());
Assert.assertEquals(NodeState.LOST, nr.getNodeState());
// registering another AM gives it the complete failed list
RMApp app2 = rm.submitApp(2000);
// Trigger nm2 heartbeat so that AM gets launched on it
nm2.nodeHeartbeat(true);
RMAppAttempt attempt2 = app2.getCurrentAppAttempt();
MockAM am2 = rm.sendAMLaunched(attempt2.getAppAttemptId());
// register AM returns all unusable nodes
am2.registerAppAttempt();
// allocate request returns no updated node
AllocateRequest allocateRequest2 =
BuilderUtils.newAllocateRequest(attempt2.getAppAttemptId(), 0, 0F, null, null);
AMResponse response2 = amService.allocate(allocateRequest2).getAMResponse();
updatedNodes = response2.getUpdatedNodes();
Assert.assertEquals(0, updatedNodes.size());
syncNodeHeartbeat(nm4, true);
// both AM's should get delta updated nodes
allocateRequest1 =
BuilderUtils.newAllocateRequest(
attempt1.getAppAttemptId(), response1.getResponseId(), 0F, null, null);
response1 = amService.allocate(allocateRequest1).getAMResponse();
updatedNodes = response1.getUpdatedNodes();
Assert.assertEquals(1, updatedNodes.size());
nr = updatedNodes.iterator().next();
Assert.assertEquals(nm4.getNodeId(), nr.getNodeId());
Assert.assertEquals(NodeState.RUNNING, nr.getNodeState());
allocateRequest2 =
BuilderUtils.newAllocateRequest(
attempt2.getAppAttemptId(), response2.getResponseId(), 0F, null, null);
response2 = amService.allocate(allocateRequest2).getAMResponse();
updatedNodes = response2.getUpdatedNodes();
Assert.assertEquals(1, updatedNodes.size());
nr = updatedNodes.iterator().next();
Assert.assertEquals(nm4.getNodeId(), nr.getNodeId());
Assert.assertEquals(NodeState.RUNNING, nr.getNodeState());
// subsequent allocate calls should return no updated nodes
allocateRequest2 =
BuilderUtils.newAllocateRequest(
attempt2.getAppAttemptId(), response2.getResponseId(), 0F, null, null);
response2 = amService.allocate(allocateRequest2).getAMResponse();
updatedNodes = response2.getUpdatedNodes();
Assert.assertEquals(0, updatedNodes.size());
// how to do the above for LOST node
}