/** Test a fire-and-forget job submission to a YARN cluster. */
@Test(timeout = 60000)
public void testDetachedPerJobYarnClusterWithStreamingJob() {
LOG.info("Starting testDetachedPerJobYarnClusterWithStreamingJob()");
File exampleJarLocation =
YarnTestBase.findFile(
".." + File.separator + "flink-examples" + File.separator + "flink-examples-streaming",
new ContainsName(new String[] {"-WordCount.jar"}));
Assert.assertNotNull("Could not find streaming wordcount jar", exampleJarLocation);
testDetachedPerJobYarnClusterInternal(exampleJarLocation.getAbsolutePath());
LOG.info("Finished testDetachedPerJobYarnClusterWithStreamingJob()");
}
/**
* Test per-job yarn cluster with the parallelism set at the CliFrontend instead of the YARN
* client.
*/
@Test
public void perJobYarnClusterWithParallelism() {
LOG.info("Starting perJobYarnClusterWithParallelism()");
// write log messages to stdout as well, so that the runWithArgs() method
// is catching the log output
addTestAppender(JobClient.class, Level.INFO);
File exampleJarLocation =
YarnTestBase.findFile(
"..",
new ContainsName(
new String[] {"-WordCount.jar"}, "streaming")); // exclude streaming wordcount here.
Assert.assertNotNull("Could not find wordcount jar", exampleJarLocation);
runWithArgs(
new String[] {
"run",
"-p",
"2", // test that the job is executed with a DOP of 2
"-m",
"yarn-cluster",
"-yj",
flinkUberjar.getAbsolutePath(),
"-yt",
flinkLibFolder.getAbsolutePath(),
"-yn",
"1",
"-yjm",
"768",
"-ytm",
"1024",
exampleJarLocation.getAbsolutePath()
},
/* test succeeded after this string */
"Job execution complete",
/* prohibited strings: (we want to see (2/2)) */
new String[] {"System.out)(1/1) switched to FINISHED "},
RunTypes.CLI_FRONTEND,
0,
true);
LOG.info("Finished perJobYarnClusterWithParallelism()");
}
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());
}
}