/**
* Test serialization of the container manager with mock execution layer.
*
* @throws Exception
*/
private void testContainerManager(StorageAgent agent) throws Exception {
FileUtils.deleteDirectory(new File(testMeta.dir)); // clean any state from previous run
LogicalPlan dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_PATH, testMeta.dir);
dag.setAttribute(OperatorContext.STORAGE_AGENT, agent);
StatsListeningOperator o1 = dag.addOperator("o1", StatsListeningOperator.class);
FSRecoveryHandler recoveryHandler =
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false));
StreamingContainerManager scm =
StreamingContainerManager.getInstance(recoveryHandler, dag, false);
File expFile = new File(recoveryHandler.getDir(), FSRecoveryHandler.FILE_SNAPSHOT);
Assert.assertTrue("snapshot file " + expFile, expFile.exists());
PhysicalPlan plan = scm.getPhysicalPlan();
assertEquals("number required containers", 1, plan.getContainers().size());
PTOperator o1p1 = plan.getOperators(dag.getMeta(o1)).get(0);
@SuppressWarnings(
"UnusedAssignment") /* sneaky: the constructor does some changes to the container */
MockContainer mc = new MockContainer(scm, o1p1.getContainer());
PTContainer originalContainer = o1p1.getContainer();
Assert.assertNotNull(o1p1.getContainer().bufferServerAddress);
assertEquals(PTContainer.State.ACTIVE, o1p1.getContainer().getState());
assertEquals("state " + o1p1, PTOperator.State.PENDING_DEPLOY, o1p1.getState());
// test restore initial snapshot + log
dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_PATH, testMeta.dir);
scm =
StreamingContainerManager.getInstance(
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false)), dag, false);
dag = scm.getLogicalPlan();
plan = scm.getPhysicalPlan();
o1p1 = plan.getOperators(dag.getOperatorMeta("o1")).get(0);
assertEquals("post restore state " + o1p1, PTOperator.State.PENDING_DEPLOY, o1p1.getState());
o1 = (StatsListeningOperator) o1p1.getOperatorMeta().getOperator();
assertEquals(
"containerId", originalContainer.getExternalId(), o1p1.getContainer().getExternalId());
assertEquals("stats listener", 1, o1p1.statsListeners.size());
assertEquals("number stats calls", 0, o1.processStatsCnt); // stats are not logged
assertEquals("post restore 1", PTContainer.State.ALLOCATED, o1p1.getContainer().getState());
assertEquals(
"post restore 1",
originalContainer.bufferServerAddress,
o1p1.getContainer().bufferServerAddress);
StreamingContainerAgent sca = scm.getContainerAgent(originalContainer.getExternalId());
Assert.assertNotNull("allocated container restored " + originalContainer, sca);
assertEquals(
"memory usage allocated container",
(int) OperatorContext.MEMORY_MB.defaultValue,
sca.container.getAllocatedMemoryMB());
// YARN-1490 - simulate container terminated on AM recovery
scm.scheduleContainerRestart(originalContainer.getExternalId());
assertEquals("memory usage of failed container", 0, sca.container.getAllocatedMemoryMB());
Checkpoint firstCheckpoint = new Checkpoint(3, 0, 0);
mc = new MockContainer(scm, o1p1.getContainer());
checkpoint(scm, o1p1, firstCheckpoint);
mc.stats(o1p1.getId())
.deployState(OperatorHeartbeat.DeployState.ACTIVE)
.currentWindowId(3)
.checkpointWindowId(3);
mc.sendHeartbeat();
assertEquals("state " + o1p1, PTOperator.State.ACTIVE, o1p1.getState());
// logical plan modification triggers snapshot
CreateOperatorRequest cor = new CreateOperatorRequest();
cor.setOperatorFQCN(GenericTestOperator.class.getName());
cor.setOperatorName("o2");
CreateStreamRequest csr = new CreateStreamRequest();
csr.setSourceOperatorName("o1");
csr.setSourceOperatorPortName("outport");
csr.setSinkOperatorName("o2");
csr.setSinkOperatorPortName("inport1");
FutureTask<?> lpmf = scm.logicalPlanModification(Lists.newArrayList(cor, csr));
while (!lpmf.isDone()) {
scm.monitorHeartbeat();
}
Assert.assertNull(lpmf.get()); // unmask exception, if any
Assert.assertSame("dag references", dag, scm.getLogicalPlan());
assertEquals("number operators after plan modification", 2, dag.getAllOperators().size());
// set operator state triggers journal write
o1p1.setState(PTOperator.State.INACTIVE);
Checkpoint offlineCheckpoint = new Checkpoint(10, 0, 0);
// write checkpoint while AM is out,
// it needs to be picked up as part of restore
checkpoint(scm, o1p1, offlineCheckpoint);
// test restore
dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_PATH, testMeta.dir);
scm =
StreamingContainerManager.getInstance(
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false)), dag, false);
Assert.assertNotSame("dag references", dag, scm.getLogicalPlan());
assertEquals(
"number operators after restore", 2, scm.getLogicalPlan().getAllOperators().size());
dag = scm.getLogicalPlan();
plan = scm.getPhysicalPlan();
o1p1 = plan.getOperators(dag.getOperatorMeta("o1")).get(0);
assertEquals("post restore state " + o1p1, PTOperator.State.INACTIVE, o1p1.getState());
o1 = (StatsListeningOperator) o1p1.getOperatorMeta().getOperator();
assertEquals("stats listener", 1, o1p1.statsListeners.size());
assertEquals("number stats calls post restore", 1, o1.processStatsCnt);
assertEquals("post restore 1", PTContainer.State.ACTIVE, o1p1.getContainer().getState());
assertEquals(
"post restore 1",
originalContainer.bufferServerAddress,
o1p1.getContainer().bufferServerAddress);
// offline checkpoint detection
assertEquals(
"checkpoints after recovery",
Lists.newArrayList(firstCheckpoint, offlineCheckpoint),
o1p1.checkpoints);
}
private void testRestartApp(StorageAgent agent, String appPath1) throws Exception {
FileUtils.deleteDirectory(new File(testMeta.dir)); // clean any state from previous run
String appId1 = "app1";
String appId2 = "app2";
String appPath2 = testMeta.dir + "/" + appId2;
LogicalPlan dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_ID, appId1);
dag.setAttribute(LogicalPlan.APPLICATION_PATH, appPath1);
dag.setAttribute(LogicalPlan.APPLICATION_ATTEMPT_ID, 1);
dag.setAttribute(OperatorContext.STORAGE_AGENT, agent);
dag.addOperator("o1", StatsListeningOperator.class);
FSRecoveryHandler recoveryHandler =
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false));
StreamingContainerManager.getInstance(recoveryHandler, dag, false);
// test restore initial snapshot + log
dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_PATH, appPath1);
StreamingContainerManager scm =
StreamingContainerManager.getInstance(
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false)), dag, false);
PhysicalPlan plan = scm.getPhysicalPlan();
dag = plan.getLogicalPlan(); // original plan
Assert.assertNotNull("operator", dag.getOperatorMeta("o1"));
PTOperator o1p1 = plan.getOperators(dag.getOperatorMeta("o1")).get(0);
long[] ids =
new FSStorageAgent(appPath1 + "/" + LogicalPlan.SUBDIR_CHECKPOINTS, new Configuration())
.getWindowIds(o1p1.getId());
Assert.assertArrayEquals(new long[] {o1p1.getRecoveryCheckpoint().getWindowId()}, ids);
Assert.assertNull(o1p1.getContainer().getExternalId());
// trigger journal write
o1p1.getContainer().setExternalId("cid1");
scm.writeJournal(o1p1.getContainer().getSetContainerState());
dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_PATH, appPath2);
dag.setAttribute(LogicalPlan.APPLICATION_ID, appId2);
StramClient sc = new StramClient(new Configuration(), dag);
try {
sc.start();
sc.copyInitialState(new Path(appPath1));
} finally {
sc.stop();
}
scm =
StreamingContainerManager.getInstance(
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false)), dag, false);
plan = scm.getPhysicalPlan();
dag = plan.getLogicalPlan();
assertEquals("modified appId", appId2, dag.getValue(LogicalPlan.APPLICATION_ID));
assertEquals("modified appPath", appPath2, dag.getValue(LogicalPlan.APPLICATION_PATH));
Assert.assertNotNull("operator", dag.getOperatorMeta("o1"));
o1p1 = plan.getOperators(dag.getOperatorMeta("o1")).get(0);
assertEquals("journal copied", "cid1", o1p1.getContainer().getExternalId());
ids =
new FSStorageAgent(appPath2 + "/" + LogicalPlan.SUBDIR_CHECKPOINTS, new Configuration())
.getWindowIds(o1p1.getId());
Assert.assertArrayEquals(
"checkpoints copied", new long[] {o1p1.getRecoveryCheckpoint().getWindowId()}, ids);
}
/**
* Launch application for the dag represented by this client.
*
* @throws YarnException
* @throws IOException
*/
public void startApplication() throws YarnException, IOException {
Class<?>[] defaultClasses;
if (applicationType.equals(YARN_APPLICATION_TYPE)) {
// TODO restrict the security check to only check if security is enabled for webservices.
if (UserGroupInformation.isSecurityEnabled()) {
defaultClasses = DATATORRENT_SECURITY_CLASSES;
} else {
defaultClasses = DATATORRENT_CLASSES;
}
} else {
throw new IllegalStateException(applicationType + " is not a valid application type.");
}
LinkedHashSet<String> localJarFiles = findJars(dag, defaultClasses);
if (resources != null) {
localJarFiles.addAll(resources);
}
YarnClusterMetrics clusterMetrics = yarnClient.getYarnClusterMetrics();
LOG.info(
"Got Cluster metric info from ASM"
+ ", numNodeManagers="
+ clusterMetrics.getNumNodeManagers());
// GetClusterNodesRequest clusterNodesReq = Records.newRecord(GetClusterNodesRequest.class);
// GetClusterNodesResponse clusterNodesResp =
// rmClient.clientRM.getClusterNodes(clusterNodesReq);
// LOG.info("Got Cluster node info from ASM");
// for (NodeReport node : clusterNodesResp.getNodeReports()) {
// LOG.info("Got node report from ASM for"
// + ", nodeId=" + node.getNodeId()
// + ", nodeAddress" + node.getHttpAddress()
// + ", nodeRackName" + node.getRackName()
// + ", nodeNumContainers" + node.getNumContainers()
// + ", nodeHealthStatus" + node.getHealthReport());
// }
List<QueueUserACLInfo> listAclInfo = yarnClient.getQueueAclsInfo();
for (QueueUserACLInfo aclInfo : listAclInfo) {
for (QueueACL userAcl : aclInfo.getUserAcls()) {
LOG.info(
"User ACL Info for Queue"
+ ", queueName="
+ aclInfo.getQueueName()
+ ", userAcl="
+ userAcl.name());
}
}
// Get a new application id
YarnClientApplication newApp = yarnClient.createApplication();
appId = newApp.getNewApplicationResponse().getApplicationId();
// Dump out information about cluster capability as seen by the resource manager
int maxMem = newApp.getNewApplicationResponse().getMaximumResourceCapability().getMemory();
LOG.info("Max mem capabililty of resources in this cluster " + maxMem);
int amMemory = dag.getMasterMemoryMB();
if (amMemory > maxMem) {
LOG.info(
"AM memory specified above max threshold of cluster. Using max value."
+ ", specified="
+ amMemory
+ ", max="
+ maxMem);
amMemory = maxMem;
}
if (dag.getAttributes().get(LogicalPlan.APPLICATION_ID) == null) {
dag.setAttribute(LogicalPlan.APPLICATION_ID, appId.toString());
}
// Create launch context for app master
LOG.info("Setting up application submission context for ASM");
ApplicationSubmissionContext appContext = Records.newRecord(ApplicationSubmissionContext.class);
// set the application id
appContext.setApplicationId(appId);
// set the application name
appContext.setApplicationName(dag.getValue(LogicalPlan.APPLICATION_NAME));
appContext.setApplicationType(this.applicationType);
if (YARN_APPLICATION_TYPE.equals(this.applicationType)) {
// appContext.setMaxAppAttempts(1); // no retries until Stram is HA
}
// Set up the container launch context for the application master
ContainerLaunchContext amContainer = Records.newRecord(ContainerLaunchContext.class);
// Setup security tokens
// If security is enabled get ResourceManager and NameNode delegation tokens.
// Set these tokens on the container so that they are sent as part of application submission.
// This also sets them up for renewal by ResourceManager. The NameNode delegation rmToken
// is also used by ResourceManager to fetch the jars from HDFS and set them up for the
// application master launch.
if (UserGroupInformation.isSecurityEnabled()) {
Credentials credentials = new Credentials();
String tokenRenewer = conf.get(YarnConfiguration.RM_PRINCIPAL);
if (tokenRenewer == null || tokenRenewer.length() == 0) {
throw new IOException("Can't get Master Kerberos principal for the RM to use as renewer");
}
// For now, only getting tokens for the default file-system.
FileSystem fs = StramClientUtils.newFileSystemInstance(conf);
try {
final Token<?> tokens[] = fs.addDelegationTokens(tokenRenewer, credentials);
if (tokens != null) {
for (Token<?> token : tokens) {
LOG.info("Got dt for " + fs.getUri() + "; " + token);
}
}
} finally {
fs.close();
}
addRMDelegationToken(tokenRenewer, credentials);
DataOutputBuffer dob = new DataOutputBuffer();
credentials.writeTokenStorageToStream(dob);
ByteBuffer fsTokens = ByteBuffer.wrap(dob.getData(), 0, dob.getLength());
amContainer.setTokens(fsTokens);
}
// set local resources for the application master
// local files or archives as needed
// In this scenario, the jar file for the application master is part of the local resources
Map<String, LocalResource> localResources = new HashMap<String, LocalResource>();
// copy required jar files to dfs, to be localized for containers
FileSystem fs = StramClientUtils.newFileSystemInstance(conf);
try {
Path appsBasePath =
new Path(StramClientUtils.getDTDFSRootDir(fs, conf), StramClientUtils.SUBDIR_APPS);
Path appPath = new Path(appsBasePath, appId.toString());
String libJarsCsv = copyFromLocal(fs, appPath, localJarFiles.toArray(new String[] {}));
LOG.info("libjars: {}", libJarsCsv);
dag.getAttributes().put(LogicalPlan.LIBRARY_JARS, libJarsCsv);
LaunchContainerRunnable.addFilesToLocalResources(
LocalResourceType.FILE, libJarsCsv, localResources, fs);
if (archives != null) {
String[] localFiles = archives.split(",");
String archivesCsv = copyFromLocal(fs, appPath, localFiles);
LOG.info("archives: {}", archivesCsv);
dag.getAttributes().put(LogicalPlan.ARCHIVES, archivesCsv);
LaunchContainerRunnable.addFilesToLocalResources(
LocalResourceType.ARCHIVE, archivesCsv, localResources, fs);
}
if (files != null) {
String[] localFiles = files.split(",");
String filesCsv = copyFromLocal(fs, appPath, localFiles);
LOG.info("files: {}", filesCsv);
dag.getAttributes().put(LogicalPlan.FILES, filesCsv);
LaunchContainerRunnable.addFilesToLocalResources(
LocalResourceType.FILE, filesCsv, localResources, fs);
}
dag.getAttributes().put(LogicalPlan.APPLICATION_PATH, appPath.toString());
if (dag.getAttributes().get(OperatorContext.STORAGE_AGENT) == null) {
/* which would be the most likely case */
Path checkpointPath = new Path(appPath, LogicalPlan.SUBDIR_CHECKPOINTS);
// use conf client side to pickup any proxy settings from dt-site.xml
dag.setAttribute(
OperatorContext.STORAGE_AGENT, new FSStorageAgent(checkpointPath.toString(), conf));
}
if (dag.getAttributes().get(LogicalPlan.CONTAINER_OPTS_CONFIGURATOR) == null) {
dag.setAttribute(
LogicalPlan.CONTAINER_OPTS_CONFIGURATOR, new BasicContainerOptConfigurator());
}
// Set the log4j properties if needed
if (!log4jPropFile.isEmpty()) {
Path log4jSrc = new Path(log4jPropFile);
Path log4jDst = new Path(appPath, "log4j.props");
fs.copyFromLocalFile(false, true, log4jSrc, log4jDst);
FileStatus log4jFileStatus = fs.getFileStatus(log4jDst);
LocalResource log4jRsrc = Records.newRecord(LocalResource.class);
log4jRsrc.setType(LocalResourceType.FILE);
log4jRsrc.setVisibility(LocalResourceVisibility.APPLICATION);
log4jRsrc.setResource(ConverterUtils.getYarnUrlFromURI(log4jDst.toUri()));
log4jRsrc.setTimestamp(log4jFileStatus.getModificationTime());
log4jRsrc.setSize(log4jFileStatus.getLen());
localResources.put("log4j.properties", log4jRsrc);
}
if (originalAppId != null) {
Path origAppPath = new Path(appsBasePath, this.originalAppId);
LOG.info("Restart from {}", origAppPath);
copyInitialState(origAppPath);
}
// push logical plan to DFS location
Path cfgDst = new Path(appPath, LogicalPlan.SER_FILE_NAME);
FSDataOutputStream outStream = fs.create(cfgDst, true);
LogicalPlan.write(this.dag, outStream);
outStream.close();
Path launchConfigDst = new Path(appPath, LogicalPlan.LAUNCH_CONFIG_FILE_NAME);
outStream = fs.create(launchConfigDst, true);
conf.writeXml(outStream);
outStream.close();
FileStatus topologyFileStatus = fs.getFileStatus(cfgDst);
LocalResource topologyRsrc = Records.newRecord(LocalResource.class);
topologyRsrc.setType(LocalResourceType.FILE);
topologyRsrc.setVisibility(LocalResourceVisibility.APPLICATION);
topologyRsrc.setResource(ConverterUtils.getYarnUrlFromURI(cfgDst.toUri()));
topologyRsrc.setTimestamp(topologyFileStatus.getModificationTime());
topologyRsrc.setSize(topologyFileStatus.getLen());
localResources.put(LogicalPlan.SER_FILE_NAME, topologyRsrc);
// Set local resource info into app master container launch context
amContainer.setLocalResources(localResources);
// Set the necessary security tokens as needed
// amContainer.setContainerTokens(containerToken);
// Set the env variables to be setup in the env where the application master will be run
LOG.info("Set the environment for the application master");
Map<String, String> env = new HashMap<String, String>();
// Add application jar(s) location to classpath
// At some point we should not be required to add
// the hadoop specific classpaths to the env.
// It should be provided out of the box.
// For now setting all required classpaths including
// the classpath to "." for the application jar(s)
// including ${CLASSPATH} will duplicate the class path in app master, removing it for now
// StringBuilder classPathEnv = new StringBuilder("${CLASSPATH}:./*");
StringBuilder classPathEnv = new StringBuilder("./*");
String classpath = conf.get(YarnConfiguration.YARN_APPLICATION_CLASSPATH);
for (String c :
StringUtils.isBlank(classpath)
? YarnConfiguration.DEFAULT_YARN_APPLICATION_CLASSPATH
: classpath.split(",")) {
if (c.equals("$HADOOP_CLIENT_CONF_DIR")) {
// SPOI-2501
continue;
}
classPathEnv.append(':');
classPathEnv.append(c.trim());
}
env.put("CLASSPATH", classPathEnv.toString());
// propagate to replace node managers user name (effective in non-secure mode)
env.put("HADOOP_USER_NAME", UserGroupInformation.getLoginUser().getUserName());
amContainer.setEnvironment(env);
// Set the necessary command to execute the application master
ArrayList<CharSequence> vargs = new ArrayList<CharSequence>(30);
// Set java executable command
LOG.info("Setting up app master command");
vargs.add(javaCmd);
if (dag.isDebug()) {
vargs.add("-agentlib:jdwp=transport=dt_socket,server=y,suspend=n");
}
// Set Xmx based on am memory size
// default heap size 75% of total memory
vargs.add("-Xmx" + (amMemory * 3 / 4) + "m");
vargs.add("-XX:+HeapDumpOnOutOfMemoryError");
vargs.add("-XX:HeapDumpPath=/tmp/dt-heap-" + appId.getId() + ".bin");
vargs.add("-Dhadoop.root.logger=" + (dag.isDebug() ? "DEBUG" : "INFO") + ",RFA");
vargs.add("-Dhadoop.log.dir=" + ApplicationConstants.LOG_DIR_EXPANSION_VAR);
vargs.add(String.format("-D%s=%s", StreamingContainer.PROP_APP_PATH, dag.assertAppPath()));
if (dag.isDebug()) {
vargs.add("-Dlog4j.debug=true");
}
String loggersLevel = conf.get(DTLoggerFactory.DT_LOGGERS_LEVEL);
if (loggersLevel != null) {
vargs.add(String.format("-D%s=%s", DTLoggerFactory.DT_LOGGERS_LEVEL, loggersLevel));
}
vargs.add(StreamingAppMaster.class.getName());
vargs.add("1>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/AppMaster.stdout");
vargs.add("2>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/AppMaster.stderr");
// Get final command
StringBuilder command = new StringBuilder(9 * vargs.size());
for (CharSequence str : vargs) {
command.append(str).append(" ");
}
LOG.info("Completed setting up app master command " + command.toString());
List<String> commands = new ArrayList<String>();
commands.add(command.toString());
amContainer.setCommands(commands);
// Set up resource type requirements
// For now, only memory is supported so we set memory requirements
Resource capability = Records.newRecord(Resource.class);
capability.setMemory(amMemory);
appContext.setResource(capability);
// Service data is a binary blob that can be passed to the application
// Not needed in this scenario
// amContainer.setServiceData(serviceData);
appContext.setAMContainerSpec(amContainer);
// Set the priority for the application master
Priority pri = Records.newRecord(Priority.class);
pri.setPriority(amPriority);
appContext.setPriority(pri);
// Set the queue to which this application is to be submitted in the RM
appContext.setQueue(queueName);
// Submit the application to the applications manager
// SubmitApplicationResponse submitResp = rmClient.submitApplication(appRequest);
// Ignore the response as either a valid response object is returned on success
// or an exception thrown to denote some form of a failure
String specStr =
Objects.toStringHelper("Submitting application: ")
.add("name", appContext.getApplicationName())
.add("queue", appContext.getQueue())
.add("user", UserGroupInformation.getLoginUser())
.add("resource", appContext.getResource())
.toString();
LOG.info(specStr);
if (dag.isDebug()) {
// LOG.info("Full submission context: " + appContext);
}
yarnClient.submitApplication(appContext);
} finally {
fs.close();
}
}