/** Test regular operation, including command line parameter parsing. */
@Test(timeout = 60000) // timeout after a minute.
public void testDetachedMode() {
LOG.info("Starting testDetachedMode()");
addTestAppender(FlinkYarnSessionCli.class, Level.INFO);
Runner runner =
startWithArgs(
new String[] {
"-j",
flinkUberjar.getAbsolutePath(),
"-t",
flinkLibFolder.getAbsolutePath(),
"-n",
"1",
"-jm",
"768",
"-tm",
"1024",
"--name",
"MyCustomName", // test setting a custom name
"--detached"
},
"Flink JobManager is now running on",
RunTypes.YARN_SESSION);
checkForLogString("The Flink YARN client has been started in detached mode");
Assert.assertFalse("The runner should detach.", runner.isAlive());
LOG.info("Waiting until two containers are running");
// wait until two containers are running
while (getRunningContainers() < 2) {
sleep(500);
}
LOG.info("Two containers are running. Killing the application");
// kill application "externally".
try {
YarnClient yc = YarnClient.createYarnClient();
yc.init(yarnConfiguration);
yc.start();
List<ApplicationReport> apps = yc.getApplications(EnumSet.of(YarnApplicationState.RUNNING));
Assert.assertEquals(1, apps.size()); // Only one running
ApplicationReport app = apps.get(0);
Assert.assertEquals("MyCustomName", app.getName());
ApplicationId id = app.getApplicationId();
yc.killApplication(id);
while (yc.getApplications(EnumSet.of(YarnApplicationState.KILLED)).size() == 0) {
sleep(500);
}
} catch (Throwable t) {
LOG.warn("Killing failed", t);
Assert.fail();
}
LOG.info("Finished testDetachedMode()");
}
@Test(timeout = 10000)
public void testGetApplications() throws YarnException, IOException {
Configuration conf = new Configuration();
final YarnClient client = new MockYarnClient();
client.init(conf);
client.start();
List<ApplicationReport> expectedReports = ((MockYarnClient) client).getReports();
List<ApplicationReport> reports = client.getApplications();
Assert.assertEquals(reports, expectedReports);
Set<String> appTypes = new HashSet<String>();
appTypes.add("YARN");
appTypes.add("NON-YARN");
reports = client.getApplications(appTypes, null);
Assert.assertEquals(reports.size(), 2);
Assert.assertTrue(
(reports.get(0).getApplicationType().equals("YARN")
&& reports.get(1).getApplicationType().equals("NON-YARN"))
|| (reports.get(1).getApplicationType().equals("YARN")
&& reports.get(0).getApplicationType().equals("NON-YARN")));
for (ApplicationReport report : reports) {
Assert.assertTrue(expectedReports.contains(report));
}
EnumSet<YarnApplicationState> appStates = EnumSet.noneOf(YarnApplicationState.class);
appStates.add(YarnApplicationState.FINISHED);
appStates.add(YarnApplicationState.FAILED);
reports = client.getApplications(null, appStates);
Assert.assertEquals(reports.size(), 2);
Assert.assertTrue(
(reports.get(0).getApplicationType().equals("NON-YARN")
&& reports.get(1).getApplicationType().equals("NON-MAPREDUCE"))
|| (reports.get(1).getApplicationType().equals("NON-YARN")
&& reports.get(0).getApplicationType().equals("NON-MAPREDUCE")));
for (ApplicationReport report : reports) {
Assert.assertTrue(expectedReports.contains(report));
}
reports = client.getApplications(appTypes, appStates);
Assert.assertEquals(reports.size(), 1);
Assert.assertTrue((reports.get(0).getApplicationType().equals("NON-YARN")));
for (ApplicationReport report : reports) {
Assert.assertTrue(expectedReports.contains(report));
}
client.stop();
}
@Before
public void checkClusterEmpty() throws IOException, YarnException {
if (yarnClient == null) {
yarnClient = YarnClient.createYarnClient();
yarnClient.init(yarnConfiguration);
yarnClient.start();
}
List<ApplicationReport> apps = yarnClient.getApplications();
for (ApplicationReport app : apps) {
if (app.getYarnApplicationState() != YarnApplicationState.FINISHED
&& app.getYarnApplicationState() != YarnApplicationState.KILLED
&& app.getYarnApplicationState() != YarnApplicationState.FAILED) {
Assert.fail(
"There is at least one application on the cluster is not finished."
+ "App "
+ app.getApplicationId()
+ " is in state "
+ app.getYarnApplicationState());
}
}
}
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());
}
}
/** Test TaskManager failure */
@Test(timeout = 100000) // timeout after 100 seconds
public void testTaskManagerFailure() {
LOG.info("Starting testTaskManagerFailure()");
Runner runner =
startWithArgs(
new String[] {
"-j",
flinkUberjar.getAbsolutePath(),
"-t",
flinkLibFolder.getAbsolutePath(),
"-n",
"1",
"-jm",
"768",
"-tm",
"1024",
"-nm",
"customName",
"-Dfancy-configuration-value=veryFancy",
"-Dyarn.maximum-failed-containers=3"
},
"Number of connected TaskManagers changed to 1. Slots available: 1",
RunTypes.YARN_SESSION);
Assert.assertEquals(2, getRunningContainers());
// ------------------------ Test if JobManager web interface is accessible -------
try {
YarnClient yc = YarnClient.createYarnClient();
yc.init(yarnConfiguration);
yc.start();
List<ApplicationReport> apps = yc.getApplications(EnumSet.of(YarnApplicationState.RUNNING));
Assert.assertEquals(1, apps.size()); // Only one running
ApplicationReport app = apps.get(0);
Assert.assertEquals("customName", app.getName());
String url = app.getTrackingUrl();
if (!url.endsWith("/")) {
url += "/";
}
if (!url.startsWith("http://")) {
url = "http://" + url;
}
LOG.info("Got application URL from YARN {}", url);
String response = TestBaseUtils.getFromHTTP(url + "taskmanagers/");
JSONObject parsedTMs = new JSONObject(response);
JSONArray taskManagers = parsedTMs.getJSONArray("taskmanagers");
Assert.assertNotNull(taskManagers);
Assert.assertEquals(1, taskManagers.length());
Assert.assertEquals(1, taskManagers.getJSONObject(0).getInt("slotsNumber"));
// get the configuration from webinterface & check if the dynamic properties from YARN show up
// there.
String jsonConfig = TestBaseUtils.getFromHTTP(url + "jobmanager/config");
JSONArray parsed = new JSONArray(jsonConfig);
Map<String, String> parsedConfig = WebMonitorUtils.fromKeyValueJsonArray(parsed);
Assert.assertEquals("veryFancy", parsedConfig.get("fancy-configuration-value"));
Assert.assertEquals("3", parsedConfig.get("yarn.maximum-failed-containers"));
// -------------- FLINK-1902: check if jobmanager hostname/port are shown in web interface
// first, get the hostname/port
String oC = outContent.toString();
Pattern p = Pattern.compile("Flink JobManager is now running on ([a-zA-Z0-9.-]+):([0-9]+)");
Matcher matches = p.matcher(oC);
String hostname = null;
String port = null;
while (matches.find()) {
hostname = matches.group(1).toLowerCase();
port = matches.group(2);
}
LOG.info("Extracted hostname:port: {} {}", hostname, port);
Assert.assertEquals(
"unable to find hostname in " + parsed,
hostname,
parsedConfig.get(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY));
Assert.assertEquals(
"unable to find port in " + parsed,
port,
parsedConfig.get(ConfigConstants.JOB_MANAGER_IPC_PORT_KEY));
// test logfile access
String logs = TestBaseUtils.getFromHTTP(url + "jobmanager/log");
Assert.assertTrue(logs.contains("Starting YARN ApplicationMaster/JobManager (Version"));
} catch (Throwable e) {
LOG.warn("Error while running test", e);
Assert.fail(e.getMessage());
}
// ------------------------ Kill container with TaskManager -------
// find container id of taskManager:
ContainerId taskManagerContainer = null;
NodeManager nodeManager = null;
UserGroupInformation remoteUgi = null;
NMTokenIdentifier nmIdent = null;
try {
remoteUgi = UserGroupInformation.getCurrentUser();
} catch (IOException e) {
LOG.warn("Unable to get curr user", e);
Assert.fail();
}
for (int nmId = 0; nmId < NUM_NODEMANAGERS; nmId++) {
NodeManager nm = yarnCluster.getNodeManager(nmId);
ConcurrentMap<ContainerId, Container> containers = nm.getNMContext().getContainers();
for (Map.Entry<ContainerId, Container> entry : containers.entrySet()) {
String command = Joiner.on(" ").join(entry.getValue().getLaunchContext().getCommands());
if (command.contains(YarnTaskManagerRunner.class.getSimpleName())) {
taskManagerContainer = entry.getKey();
nodeManager = nm;
nmIdent =
new NMTokenIdentifier(taskManagerContainer.getApplicationAttemptId(), null, "", 0);
// allow myself to do stuff with the container
// remoteUgi.addCredentials(entry.getValue().getCredentials());
remoteUgi.addTokenIdentifier(nmIdent);
}
}
sleep(500);
}
Assert.assertNotNull("Unable to find container with TaskManager", taskManagerContainer);
Assert.assertNotNull("Illegal state", nodeManager);
List<ContainerId> toStop = new LinkedList<ContainerId>();
toStop.add(taskManagerContainer);
StopContainersRequest scr = StopContainersRequest.newInstance(toStop);
try {
nodeManager.getNMContext().getContainerManager().stopContainers(scr);
} catch (Throwable e) {
LOG.warn("Error stopping container", e);
Assert.fail("Error stopping container: " + e.getMessage());
}
// stateful termination check:
// wait until we saw a container being killed and AFTERWARDS a new one launched
boolean ok = false;
do {
LOG.debug("Waiting for correct order of events. Output: {}", errContent.toString());
String o = errContent.toString();
int killedOff = o.indexOf("Container killed by the ApplicationMaster");
if (killedOff != -1) {
o = o.substring(killedOff);
ok = o.indexOf("Launching container") > 0;
}
sleep(1000);
} while (!ok);
// send "stop" command to command line interface
runner.sendStop();
// wait for the thread to stop
try {
runner.join(1000);
} catch (InterruptedException e) {
LOG.warn("Interrupted while stopping runner", e);
}
LOG.warn("stopped");
// ----------- Send output to logger
System.setOut(originalStdout);
System.setErr(originalStderr);
String oC = outContent.toString();
String eC = errContent.toString();
LOG.info("Sending stdout content through logger: \n\n{}\n\n", oC);
LOG.info("Sending stderr content through logger: \n\n{}\n\n", eC);
// ------ Check if everything happened correctly
Assert.assertTrue(
"Expect to see failed container", eC.contains("New messages from the YARN cluster"));
Assert.assertTrue(
"Expect to see failed container", eC.contains("Container killed by the ApplicationMaster"));
Assert.assertTrue(
"Expect to see new container started",
eC.contains("Launching container") && eC.contains("on host"));
// cleanup auth for the subsequent tests.
remoteUgi.getTokenIdentifiers().remove(nmIdent);
LOG.info("Finished testTaskManagerFailure()");
}