public void launchSupervisorOnContainer(Container container) throws IOException {
LOG.info("Connecting to ContainerManager for containerid=" + container.getId());
String cmIpPortStr = container.getNodeId().getHost() + ":" + container.getNodeId().getPort();
InetSocketAddress cmAddress = NetUtils.createSocketAddr(cmIpPortStr);
LOG.info("Connecting to ContainerManager at " + cmIpPortStr);
ContainerManager proxy =
((ContainerManager) rpc.getProxy(ContainerManager.class, cmAddress, hadoopConf));
LOG.info("launchSupervisorOnContainer( id:" + container.getId() + " )");
ContainerLaunchContext launchContext = Records.newRecord(ContainerLaunchContext.class);
launchContext.setContainerId(container.getId());
launchContext.setResource(container.getResource());
try {
launchContext.setUser(UserGroupInformation.getCurrentUser().getShortUserName());
} catch (IOException e) {
LOG.info(
"Getting current user info failed when trying to launch the container" + e.getMessage());
}
Map<String, String> env = new HashMap<String, String>();
launchContext.setEnvironment(env);
Map<String, LocalResource> localResources = new HashMap<String, LocalResource>();
String stormVersion = Util.getStormVersion(this.storm_conf);
Path zip = new Path("/lib/storm/" + stormVersion + "/storm.zip");
FileSystem fs = FileSystem.get(this.hadoopConf);
localResources.put(
"storm",
Util.newYarnAppResource(
fs, zip, LocalResourceType.ARCHIVE, LocalResourceVisibility.PUBLIC));
String appHome = Util.getApplicationHomeForId(this.appAttemptId.toString());
Path dirDst = Util.createConfigurationFileInFs(fs, appHome, this.storm_conf, this.hadoopConf);
localResources.put("conf", Util.newYarnAppResource(fs, dirDst));
launchContext.setLocalResources(localResources);
List<String> supervisorArgs = Util.buildSupervisorCommands(this.storm_conf);
launchContext.setCommands(supervisorArgs);
StartContainerRequest startRequest = Records.newRecord(StartContainerRequest.class);
startRequest.setContainerLaunchContext(launchContext);
LOG.info(
"launchSupervisorOnContainer: startRequest prepared, calling startContainer. "
+ startRequest);
try {
StartContainerResponse response = proxy.startContainer(startRequest);
LOG.info("Got a start container response " + response);
} catch (Exception e) {
LOG.error("Caught an exception while trying to start a container", e);
System.exit(-1);
}
}
@SuppressWarnings("unchecked")
public synchronized void launch(NMCommunicatorLaunchRequestEvent event) {
LOG.info("Launching Container with Id: " + event.getContainerId());
if (this.state == ContainerState.KILLED_BEFORE_LAUNCH) {
state = ContainerState.DONE;
sendContainerLaunchFailedMsg(
event.getContainerId(), "Container was killed before it was launched");
return;
}
ContainerManagementProtocolProxyData proxy = null;
try {
proxy = getCMProxy(containerID, containerMgrAddress, containerToken);
// Construct the actual Container
ContainerLaunchContext containerLaunchContext = event.getContainerLaunchContext();
// Now launch the actual container
StartContainerRequest startRequest = Records.newRecord(StartContainerRequest.class);
startRequest.setContainerToken(event.getContainerToken());
startRequest.setContainerLaunchContext(containerLaunchContext);
StartContainersResponse response =
proxy
.getContainerManagementProtocol()
.startContainers(
StartContainersRequest.newInstance(Collections.singletonList(startRequest)));
if (response.getFailedRequests() != null && !response.getFailedRequests().isEmpty()) {
throw response.getFailedRequests().get(containerID).deSerialize();
}
// after launching, send launched event to task attempt to move
// it from ASSIGNED to RUNNING state
context.getEventHandler().handle(new AMContainerEventLaunched(containerID));
ContainerLaunchedEvent lEvt =
new ContainerLaunchedEvent(
containerID, clock.getTime(), context.getApplicationAttemptId());
context.getHistoryHandler().handle(new DAGHistoryEvent(null, lEvt));
this.state = ContainerState.RUNNING;
} catch (Throwable t) {
String message =
"Container launch failed for " + containerID + " : " + ExceptionUtils.getStackTrace(t);
this.state = ContainerState.FAILED;
sendContainerLaunchFailedMsg(containerID, message);
} finally {
if (proxy != null) {
cmProxy.mayBeCloseProxy(proxy);
}
}
}
@Test
public void testSuccessfulContainerLaunch()
throws InterruptedException, IOException, YarnException {
FileContext localFS = FileContext.getLocalFSFileContext();
localFS.delete(new Path(localDir.getAbsolutePath()), true);
localFS.delete(new Path(localLogDir.getAbsolutePath()), true);
localFS.delete(new Path(remoteLogDir.getAbsolutePath()), true);
localDir.mkdir();
localLogDir.mkdir();
remoteLogDir.mkdir();
YarnConfiguration conf = new YarnConfiguration();
Context context =
new NMContext(
new NMContainerTokenSecretManager(conf),
new NMTokenSecretManagerInNM(),
null,
null,
new NMNullStateStoreService()) {
@Override
public int getHttpPort() {
return 1234;
}
};
conf.set(YarnConfiguration.NM_LOCAL_DIRS, localDir.getAbsolutePath());
conf.set(YarnConfiguration.NM_LOG_DIRS, localLogDir.getAbsolutePath());
conf.set(YarnConfiguration.NM_REMOTE_APP_LOG_DIR, remoteLogDir.getAbsolutePath());
ContainerExecutor exec = new DefaultContainerExecutor();
exec.setConf(conf);
DeletionService del = new DeletionService(exec);
Dispatcher dispatcher = new AsyncDispatcher();
LocalDirsHandlerService dirsHandler = new LocalDirsHandlerService();
NodeHealthCheckerService healthChecker =
new NodeHealthCheckerService(NodeManager.getNodeHealthScriptRunner(conf), dirsHandler);
healthChecker.init(conf);
NodeManagerMetrics metrics = NodeManagerMetrics.create();
NodeStatusUpdater nodeStatusUpdater =
new NodeStatusUpdaterImpl(context, dispatcher, healthChecker, metrics) {
@Override
protected ResourceTracker getRMClient() {
return new LocalRMInterface();
};
@Override
protected void stopRMProxy() {
return;
}
@Override
protected void startStatusUpdater() {
return; // Don't start any updating thread.
}
@Override
public long getRMIdentifier() {
return SIMULATED_RM_IDENTIFIER;
}
};
DummyContainerManager containerManager =
new DummyContainerManager(
context,
exec,
del,
nodeStatusUpdater,
metrics,
new ApplicationACLsManager(conf),
dirsHandler);
nodeStatusUpdater.init(conf);
((NMContext) context).setContainerManager(containerManager);
nodeStatusUpdater.start();
containerManager.init(conf);
containerManager.start();
ContainerLaunchContext launchContext =
recordFactory.newRecordInstance(ContainerLaunchContext.class);
ApplicationId applicationId = ApplicationId.newInstance(0, 0);
ApplicationAttemptId applicationAttemptId = ApplicationAttemptId.newInstance(applicationId, 0);
ContainerId cID = ContainerId.newContainerId(applicationAttemptId, 0);
String user = "******";
StartContainerRequest scRequest =
StartContainerRequest.newInstance(
launchContext,
TestContainerManager.createContainerToken(
cID,
SIMULATED_RM_IDENTIFIER,
context.getNodeId(),
user,
context.getContainerTokenSecretManager()));
List<StartContainerRequest> list = new ArrayList<StartContainerRequest>();
list.add(scRequest);
StartContainersRequest allRequests = StartContainersRequest.newInstance(list);
containerManager.startContainers(allRequests);
BaseContainerManagerTest.waitForContainerState(containerManager, cID, ContainerState.RUNNING);
List<ContainerId> containerIds = new ArrayList<ContainerId>();
containerIds.add(cID);
StopContainersRequest stopRequest = StopContainersRequest.newInstance(containerIds);
containerManager.stopContainers(stopRequest);
BaseContainerManagerTest.waitForContainerState(containerManager, cID, ContainerState.COMPLETE);
containerManager.stop();
}
@Test
public void testSuccessfulContainerLaunch() throws InterruptedException, IOException {
FileContext localFS = FileContext.getLocalFSFileContext();
localFS.delete(new Path(localDir.getAbsolutePath()), true);
localFS.delete(new Path(localLogDir.getAbsolutePath()), true);
localFS.delete(new Path(remoteLogDir.getAbsolutePath()), true);
localDir.mkdir();
localLogDir.mkdir();
remoteLogDir.mkdir();
Context context = new NMContext();
YarnConfiguration conf = new YarnConfiguration();
conf.set(YarnConfiguration.NM_LOCAL_DIRS, localDir.getAbsolutePath());
conf.set(YarnConfiguration.NM_LOG_DIRS, localLogDir.getAbsolutePath());
conf.set(YarnConfiguration.NM_REMOTE_APP_LOG_DIR, remoteLogDir.getAbsolutePath());
ContainerExecutor exec = new DefaultContainerExecutor();
exec.setConf(conf);
DeletionService del = new DeletionService(exec);
Dispatcher dispatcher = new AsyncDispatcher();
NodeHealthCheckerService healthChecker = new NodeHealthCheckerService();
healthChecker.init(conf);
LocalDirsHandlerService dirsHandler = healthChecker.getDiskHandler();
NodeManagerMetrics metrics = NodeManagerMetrics.create();
ContainerTokenSecretManager containerTokenSecretManager = new ContainerTokenSecretManager();
NodeStatusUpdater nodeStatusUpdater =
new NodeStatusUpdaterImpl(
context, dispatcher, healthChecker, metrics, containerTokenSecretManager) {
@Override
protected ResourceTracker getRMClient() {
return new LocalRMInterface();
};
@Override
protected void startStatusUpdater() {
return; // Don't start any updating thread.
}
};
DummyContainerManager containerManager =
new DummyContainerManager(
context,
exec,
del,
nodeStatusUpdater,
metrics,
containerTokenSecretManager,
new ApplicationACLsManager(conf),
dirsHandler);
containerManager.init(conf);
containerManager.start();
ContainerLaunchContext launchContext =
recordFactory.newRecordInstance(ContainerLaunchContext.class);
ContainerId cID = recordFactory.newRecordInstance(ContainerId.class);
ApplicationId applicationId = recordFactory.newRecordInstance(ApplicationId.class);
applicationId.setClusterTimestamp(0);
applicationId.setId(0);
ApplicationAttemptId applicationAttemptId =
recordFactory.newRecordInstance(ApplicationAttemptId.class);
applicationAttemptId.setApplicationId(applicationId);
applicationAttemptId.setAttemptId(0);
cID.setApplicationAttemptId(applicationAttemptId);
launchContext.setContainerId(cID);
launchContext.setUser("testing");
launchContext.setResource(recordFactory.newRecordInstance(Resource.class));
StartContainerRequest request = recordFactory.newRecordInstance(StartContainerRequest.class);
request.setContainerLaunchContext(launchContext);
containerManager.startContainer(request);
BaseContainerManagerTest.waitForContainerState(containerManager, cID, ContainerState.RUNNING);
StopContainerRequest stopRequest = recordFactory.newRecordInstance(StopContainerRequest.class);
stopRequest.setContainerId(cID);
containerManager.stopContainer(stopRequest);
BaseContainerManagerTest.waitForContainerState(containerManager, cID, ContainerState.COMPLETE);
containerManager.stop();
}
