/**
* create a summary of the application's runtime.
*
* @param app {@link RMApp} whose summary is to be created, cannot be <code>null</code>.
*/
public static SummaryBuilder createAppSummary(RMApp app) {
String trackingUrl = "N/A";
String host = "N/A";
RMAppAttempt attempt = app.getCurrentAppAttempt();
if (attempt != null) {
trackingUrl = attempt.getTrackingUrl();
host = attempt.getHost();
}
RMAppMetrics metrics = app.getRMAppMetrics();
SummaryBuilder summary =
new SummaryBuilder()
.add("appId", app.getApplicationId())
.add("name", app.getName())
.add("user", app.getUser())
.add("queue", app.getQueue())
.add("state", app.getState())
.add("trackingUrl", trackingUrl)
.add("appMasterHost", host)
.add("startTime", app.getStartTime())
.add("finishTime", app.getFinishTime())
.add("finalStatus", app.getFinalApplicationStatus())
.add("memorySeconds", metrics.getMemorySeconds())
.add("vcoreSeconds", metrics.getVcoreSeconds())
.add("preemptedAMContainers", metrics.getNumAMContainersPreempted())
.add("preemptedNonAMContainers", metrics.getNumNonAMContainersPreempted())
.add("preemptedResources", metrics.getResourcePreempted())
.add("applicationType", app.getApplicationType());
return summary;
}
public AppAttemptInfo(RMAppAttempt attempt) {
this.startTime = 0;
this.containerId = "";
this.nodeHttpAddress = "";
this.nodeId = "";
this.logsLink = "";
if (attempt != null) {
this.id = attempt.getAppAttemptId().getAttemptId();
this.startTime = attempt.getStartTime();
Container masterContainer = attempt.getMasterContainer();
if (masterContainer != null) {
this.containerId = masterContainer.getId().toString();
this.nodeHttpAddress = masterContainer.getNodeHttpAddress();
this.nodeId = masterContainer.getNodeId().toString();
this.logsLink =
join(
HttpConfig.getSchemePrefix(),
masterContainer.getNodeHttpAddress(),
"/node",
"/containerlogs/",
ConverterUtils.toString(masterContainer.getId()),
"/",
attempt.getSubmissionContext().getUser());
}
}
}
// This is to test fetching AM container will be retried, if AM container is
// not fetchable since DNS is unavailable causing container token/NMtoken
// creation failure.
@Test(timeout = 20000)
public void testAMContainerAllocationWhenDNSUnavailable() throws Exception {
final YarnConfiguration conf = new YarnConfiguration();
MockRM rm1 =
new MockRM(conf) {
@Override
protected RMSecretManagerService createRMSecretManagerService() {
return new TestRMSecretManagerService(conf, rmContext);
}
};
rm1.start();
MockNM nm1 = rm1.registerNode("unknownhost:1234", 8000);
SecurityUtilTestHelper.setTokenServiceUseIp(true);
RMApp app1 = rm1.submitApp(200);
RMAppAttempt attempt = app1.getCurrentAppAttempt();
nm1.nodeHeartbeat(true);
// fetching am container will fail, keep retrying 5 times.
while (numRetries <= 5) {
nm1.nodeHeartbeat(true);
Thread.sleep(1000);
Assert.assertEquals(RMAppAttemptState.SCHEDULED, attempt.getAppAttemptState());
System.out.println("Waiting for am container to be allocated.");
}
SecurityUtilTestHelper.setTokenServiceUseIp(false);
rm1.waitForState(attempt.getAppAttemptId(), RMAppAttemptState.ALLOCATED);
MockRM.launchAndRegisterAM(app1, rm1, nm1);
}
/**
* create a summary of the application's runtime.
*
* @param app {@link RMApp} whose summary is to be created, cannot be <code>null</code>.
*/
public static SummaryBuilder createAppSummary(RMApp app) {
String trackingUrl = "N/A";
String host = "N/A";
RMAppAttempt attempt = app.getCurrentAppAttempt();
if (attempt != null) {
trackingUrl = attempt.getTrackingUrl();
host = attempt.getHost();
}
SummaryBuilder summary =
new SummaryBuilder()
.add("appId", app.getApplicationId())
.add("name", app.getName())
.add("user", app.getUser())
.add("queue", app.getQueue())
.add("state", app.getState())
.add("trackingUrl", trackingUrl)
.add("appMasterHost", host)
.add("startTime", app.getStartTime())
.add("finishTime", app.getFinishTime())
.add("finalStatus", app.getFinalApplicationStatus());
return summary;
}
@Test
public void testARRMResponseId() throws Exception {
MockNM nm1 = rm.registerNode("h1:1234", 5000);
RMApp app = rm.submitApp(2000);
// Trigger the scheduling so the AM gets 'launched'
nm1.nodeHeartbeat(true);
RMAppAttempt attempt = app.getCurrentAppAttempt();
MockAM am = rm.sendAMLaunched(attempt.getAppAttemptId());
am.registerAppAttempt();
AllocateRequest allocateRequest =
BuilderUtils.newAllocateRequest(attempt.getAppAttemptId(), 0, 0F, null, null);
AllocateResponse response = amService.allocate(allocateRequest);
Assert.assertEquals(1, response.getResponseId());
Assert.assertFalse(response.getReboot());
allocateRequest =
BuilderUtils.newAllocateRequest(
attempt.getAppAttemptId(), response.getResponseId(), 0F, null, null);
response = amService.allocate(allocateRequest);
Assert.assertEquals(2, response.getResponseId());
/* try resending */
response = amService.allocate(allocateRequest);
Assert.assertEquals(2, response.getResponseId());
/** try sending old request again * */
allocateRequest = BuilderUtils.newAllocateRequest(attempt.getAppAttemptId(), 0, 0F, null, null);
response = amService.allocate(allocateRequest);
Assert.assertTrue(response.getReboot());
}
@Test(timeout = 30000)
public void testExcessReservationThanNodeManagerCapacity() throws Exception {
YarnConfiguration conf = new YarnConfiguration();
YarnAPIStorageFactory.setConfiguration(conf);
RMStorageFactory.setConfiguration(conf);
MockRM rm = new MockRM(conf);
try {
rm.start();
// Register node1
MockNM nm1 = rm.registerNode("127.0.0.1:1234", 2 * GB, 4);
MockNM nm2 = rm.registerNode("127.0.0.1:2234", 3 * GB, 4);
nm1.nodeHeartbeat(true);
nm2.nodeHeartbeat(true);
// HOP :: Sleep to allow previous events to be processed
Thread.sleep(
conf.getInt(
YarnConfiguration.HOPS_PENDING_EVENTS_RETRIEVAL_PERIOD,
YarnConfiguration.DEFAULT_HOPS_PENDING_EVENTS_RETRIEVAL_PERIOD)
* 2);
// wait..
int waitCount = 20;
int size = rm.getRMContext().getActiveRMNodes().size();
while ((size = rm.getRMContext().getActiveRMNodes().size()) != 2 && waitCount-- > 0) {
LOG.info("Waiting for node managers to register : " + size);
Thread.sleep(100);
}
Assert.assertEquals(2, rm.getRMContext().getActiveRMNodes().size());
// Submit an application
RMApp app1 = rm.submitApp(128);
// kick the scheduling
nm1.nodeHeartbeat(true);
RMAppAttempt attempt1 = app1.getCurrentAppAttempt();
MockAM am1 = rm.sendAMLaunched(attempt1.getAppAttemptId(), nm1);
am1.registerAppAttempt();
LOG.info("sending container requests ");
am1.addRequests(new String[] {"*"}, 3 * GB, 1, 1);
AllocateResponse alloc1Response = am1.schedule(); // send the request
// kick the scheduler
nm1.nodeHeartbeat(true);
int waitCounter = 20;
LOG.info("heartbeating nm1");
while (alloc1Response.getAllocatedContainers().size() < 1 && waitCounter-- > 0) {
LOG.info("Waiting for containers to be created for app 1...");
Thread.sleep(500);
alloc1Response = am1.schedule();
}
LOG.info("received container : " + alloc1Response.getAllocatedContainers().size());
// No container should be allocated.
// Internally it should not been reserved.
Assert.assertTrue(alloc1Response.getAllocatedContainers().size() == 0);
LOG.info("heartbeating nm2");
waitCounter = 20;
nm2.nodeHeartbeat(true);
while (alloc1Response.getAllocatedContainers().size() < 1 && waitCounter-- > 0) {
LOG.info("Waiting for containers to be created for app 1...");
Thread.sleep(500);
alloc1Response = am1.schedule();
}
LOG.info("received container : " + alloc1Response.getAllocatedContainers().size());
Assert.assertTrue(alloc1Response.getAllocatedContainers().size() == 1);
} finally {
rm.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
}