private void waitTillAccepted(YarnClient rmClient, ApplicationId appId) throws Exception {
try {
long start = System.currentTimeMillis();
ApplicationReport report = rmClient.getApplicationReport(appId);
while (YarnApplicationState.ACCEPTED != report.getYarnApplicationState()) {
if (System.currentTimeMillis() - start > 20 * 1000) {
throw new Exception("App '" + appId + "' time out, failed to reach ACCEPTED state");
}
Thread.sleep(200);
report = rmClient.getApplicationReport(appId);
}
} catch (Exception ex) {
throw new Exception(ex);
}
}
public TezSessionStatus getSessionStatus() throws TezException, IOException {
try {
ApplicationReport appReport = yarnClient.getApplicationReport(applicationId);
switch (appReport.getYarnApplicationState()) {
case NEW:
case NEW_SAVING:
case ACCEPTED:
case SUBMITTED:
return TezSessionStatus.INITIALIZING;
case FINISHED:
case FAILED:
case KILLED:
return TezSessionStatus.SHUTDOWN;
case RUNNING:
try {
DAGClientAMProtocolBlockingPB proxy =
TezClientUtils.getSessionAMProxy(
yarnClient, sessionConfig.getYarnConfiguration(), applicationId);
if (proxy == null) {
return TezSessionStatus.INITIALIZING;
}
GetAMStatusResponseProto response =
proxy.getAMStatus(null, GetAMStatusRequestProto.newBuilder().build());
return DagTypeConverters.convertTezSessionStatusFromProto(response.getStatus());
} catch (TezException e) {
LOG.info("Failed to retrieve AM Status via proxy", e);
} catch (ServiceException e) {
LOG.info("Failed to retrieve AM Status via proxy", e);
}
}
} catch (YarnException e) {
throw new TezException(e);
}
return TezSessionStatus.INITIALIZING;
}
@Override
public void run() {
while (running.get() && yarnClient.isInState(Service.STATE.STARTED)) {
try {
ApplicationReport report = yarnClient.getApplicationReport(appId);
synchronized (lock) {
lastReport = report;
}
} catch (Exception e) {
LOG.warn("Error while getting application report", e);
}
try {
Thread.sleep(FlinkYarnCluster.POLLING_THREAD_INTERVAL_MS);
} catch (InterruptedException e) {
LOG.error("Polling thread got interrupted", e);
Thread.currentThread().interrupt(); // pass interrupt.
}
}
if (running.get() && !yarnClient.isInState(Service.STATE.STARTED)) {
// == if the polling thread is still running but the yarn client is stopped.
LOG.warn("YARN client is unexpected in state " + yarnClient.getServiceState());
}
}
/**
* Create a new Flink on YARN cluster.
*
* @param yarnClient
* @param appId the YARN application ID
* @param hadoopConfig
* @param flinkConfig
* @param sessionFilesDir
* @param detached Set to true if no actor system or RPC communication with the cluster should be
* established
* @throws IOException
* @throws YarnException
*/
public FlinkYarnCluster(
final YarnClient yarnClient,
final ApplicationId appId,
Configuration hadoopConfig,
org.apache.flink.configuration.Configuration flinkConfig,
Path sessionFilesDir,
boolean detached)
throws IOException, YarnException {
this.akkaDuration = AkkaUtils.getTimeout(flinkConfig);
this.akkaTimeout = Timeout.durationToTimeout(akkaDuration);
this.yarnClient = yarnClient;
this.hadoopConfig = hadoopConfig;
this.sessionFilesDir = sessionFilesDir;
this.applicationId = appId;
this.detached = detached;
this.flinkConfig = flinkConfig;
this.appId = appId;
// get one application report manually
intialAppReport = yarnClient.getApplicationReport(appId);
String jobManagerHost = intialAppReport.getHost();
int jobManagerPort = intialAppReport.getRpcPort();
this.jobManagerAddress = new InetSocketAddress(jobManagerHost, jobManagerPort);
}
public ApplicationReport getApplicationReport() throws YarnException, IOException {
return yarnClient.getApplicationReport(this.appId);
}
private void testDetachedPerJobYarnClusterInternal(String job) {
YarnClient yc = YarnClient.createYarnClient();
yc.init(yarnConfiguration);
yc.start();
// get temporary folder for writing output of wordcount example
File tmpOutFolder = null;
try {
tmpOutFolder = tmp.newFolder();
} catch (IOException e) {
throw new RuntimeException(e);
}
// get temporary file for reading input data for wordcount example
File tmpInFile;
try {
tmpInFile = tmp.newFile();
FileUtils.writeStringToFile(tmpInFile, WordCountData.TEXT);
} catch (IOException e) {
throw new RuntimeException(e);
}
Runner runner =
startWithArgs(
new String[] {
"run",
"-m",
"yarn-cluster",
"-yj",
flinkUberjar.getAbsolutePath(),
"-yt",
flinkLibFolder.getAbsolutePath(),
"-yn",
"1",
"-yjm",
"768",
"-yD",
"yarn.heap-cutoff-ratio=0.5", // test if the cutoff is passed correctly
"-ytm",
"1024",
"-ys",
"2", // test requesting slots from YARN.
"--yarndetached",
job,
tmpInFile.getAbsoluteFile().toString(),
tmpOutFolder.getAbsoluteFile().toString()
},
"Job has been submitted with JobID",
RunTypes.CLI_FRONTEND);
// it should usually be 2, but on slow machines, the number varies
Assert.assertTrue("There should be at most 2 containers running", getRunningContainers() <= 2);
// give the runner some time to detach
for (int attempt = 0; runner.isAlive() && attempt < 5; attempt++) {
try {
Thread.sleep(500);
} catch (InterruptedException e) {
}
}
Assert.assertFalse("The runner should detach.", runner.isAlive());
LOG.info("CLI Frontend has returned, so the job is running");
// find out the application id and wait until it has finished.
try {
List<ApplicationReport> apps = yc.getApplications(EnumSet.of(YarnApplicationState.RUNNING));
ApplicationId tmpAppId;
if (apps.size() == 1) {
// Better method to find the right appId. But sometimes the app is shutting down very fast
// Only one running
tmpAppId = apps.get(0).getApplicationId();
LOG.info("waiting for the job with appId {} to finish", tmpAppId);
// wait until the app has finished
while (yc.getApplications(EnumSet.of(YarnApplicationState.RUNNING)).size() > 0) {
sleep(500);
}
} else {
// get appId by finding the latest finished appid
apps = yc.getApplications();
Collections.sort(
apps,
new Comparator<ApplicationReport>() {
@Override
public int compare(ApplicationReport o1, ApplicationReport o2) {
return o1.getApplicationId().compareTo(o2.getApplicationId()) * -1;
}
});
tmpAppId = apps.get(0).getApplicationId();
LOG.info(
"Selected {} as the last appId from {}", tmpAppId, Arrays.toString(apps.toArray()));
}
final ApplicationId id = tmpAppId;
// now it has finished.
// check the output files.
File[] listOfOutputFiles = tmpOutFolder.listFiles();
Assert.assertNotNull("Taskmanager output not found", listOfOutputFiles);
LOG.info("The job has finished. TaskManager output files found in {}", tmpOutFolder);
// read all output files in output folder to one output string
String content = "";
for (File f : listOfOutputFiles) {
if (f.isFile()) {
content += FileUtils.readFileToString(f) + "\n";
}
}
// String content = FileUtils.readFileToString(taskmanagerOut);
// check for some of the wordcount outputs.
Assert.assertTrue(
"Expected string 'da 5' or '(all,2)' not found in string '" + content + "'",
content.contains("da 5") || content.contains("(da,5)") || content.contains("(all,2)"));
Assert.assertTrue(
"Expected string 'der 29' or '(mind,1)' not found in string'" + content + "'",
content.contains("der 29")
|| content.contains("(der,29)")
|| content.contains("(mind,1)"));
// check if the heap size for the TaskManager was set correctly
File jobmanagerLog =
YarnTestBase.findFile(
"..",
new FilenameFilter() {
@Override
public boolean accept(File dir, String name) {
return name.contains("jobmanager.log")
&& dir.getAbsolutePath().contains(id.toString());
}
});
Assert.assertNotNull("Unable to locate JobManager log", jobmanagerLog);
content = FileUtils.readFileToString(jobmanagerLog);
// TM was started with 1024 but we cut off 50% (NOT THE DEFAULT VALUE)
String expected = "Starting TM with command=$JAVA_HOME/bin/java -Xms424m -Xmx424m";
Assert.assertTrue(
"Expected string '" + expected + "' not found in JobManager log: '" + jobmanagerLog + "'",
content.contains(expected));
expected = " (2/2) (attempt #0) to ";
Assert.assertTrue(
"Expected string '"
+ expected
+ "' not found in JobManager log."
+ "This string checks that the job has been started with a parallelism of 2. Log contents: '"
+ jobmanagerLog
+ "'",
content.contains(expected));
// make sure the detached app is really finished.
LOG.info("Checking again that app has finished");
ApplicationReport rep;
do {
sleep(500);
rep = yc.getApplicationReport(id);
LOG.info("Got report {}", rep);
} while (rep.getYarnApplicationState() == YarnApplicationState.RUNNING);
} catch (Throwable t) {
LOG.warn("Error while detached yarn session was running", t);
Assert.fail(t.getMessage());
}
}
public void run(String[] args) throws Exception {
final String command = args[0];
final int n = Integer.valueOf(args[1]);
final Path jarPath = new Path(args[2]);
// Create yarnClient
YarnConfiguration conf = new YarnConfiguration();
YarnClient yarnClient = YarnClient.createYarnClient();
yarnClient.init(conf);
yarnClient.start();
// Create application via yarnClient
YarnClientApplication app = yarnClient.createApplication();
// Set up the container launch context for the application master
ContainerLaunchContext amContainer = Records.newRecord(ContainerLaunchContext.class);
amContainer.setCommands(
Collections.singletonList(
"$JAVA_HOME/bin/java"
+ " -Xmx256M"
+ " com.hortonworks.simpleyarnapp.ApplicationMaster"
+ " "
+ command
+ " "
+ String.valueOf(n)
+ " 1>"
+ ApplicationConstants.LOG_DIR_EXPANSION_VAR
+ "/stdout"
+ " 2>"
+ ApplicationConstants.LOG_DIR_EXPANSION_VAR
+ "/stderr"));
// Setup jar for ApplicationMaster
LocalResource appMasterJar = Records.newRecord(LocalResource.class);
setupAppMasterJar(jarPath, appMasterJar);
System.out.println("Jar name is " + jarPath.getName());
amContainer.setLocalResources(Collections.singletonMap(jarPath.getName(), appMasterJar));
// Setup CLASSPATH for ApplicationMaster
Map<String, String> appMasterEnv = new HashMap<String, String>();
setupAppMasterEnv(appMasterEnv);
amContainer.setEnvironment(appMasterEnv);
// Set up resource type requirements for ApplicationMaster
Resource capability = Records.newRecord(Resource.class);
capability.setMemory(256);
capability.setVirtualCores(1);
// Finally, set-up ApplicationSubmissionContext for the application
ApplicationSubmissionContext appContext = app.getApplicationSubmissionContext();
appContext.setApplicationName("apache-yarn-example"); // application name
appContext.setAMContainerSpec(amContainer);
appContext.setResource(capability);
appContext.setQueue("default"); // queue
// Submit application
ApplicationId appId = appContext.getApplicationId();
System.out.println("Submitting application " + appId);
yarnClient.submitApplication(appContext);
ApplicationReport appReport = yarnClient.getApplicationReport(appId);
YarnApplicationState appState = appReport.getYarnApplicationState();
while (appState != YarnApplicationState.FINISHED
&& appState != YarnApplicationState.KILLED
&& appState != YarnApplicationState.FAILED) {
System.out.println("App Status = " + appState);
Thread.sleep(100);
appReport = yarnClient.getApplicationReport(appId);
appState = appReport.getYarnApplicationState();
}
System.out.println(
"Application "
+ appId
+ " finished with"
+ " state "
+ appState
+ " at "
+ appReport.getFinishTime());
}