private static String[] runTests(
int numHosts, int slotsPerHost, int parallelism, TestLocatableInputSplit[] splits)
throws Exception {
AbstractJobVertex vertex = new AbstractJobVertex("test vertex");
vertex.setParallelism(parallelism);
vertex.setInvokableClass(DummyInvokable.class);
vertex.setInputSplitSource(new TestInputSplitSource(splits));
JobGraph jobGraph = new JobGraph("test job", vertex);
ExecutionGraph eg =
new ExecutionGraph(
jobGraph.getJobID(), jobGraph.getName(), jobGraph.getJobConfiguration(), TIMEOUT);
eg.setQueuedSchedulingAllowed(false);
eg.attachJobGraph(jobGraph.getVerticesSortedTopologicallyFromSources());
Scheduler scheduler = getScheduler(numHosts, slotsPerHost);
eg.scheduleForExecution(scheduler);
ExecutionVertex[] tasks = eg.getVerticesTopologically().iterator().next().getTaskVertices();
assertEquals(parallelism, tasks.length);
String[] hostsForTasks = new String[parallelism];
for (int i = 0; i < parallelism; i++) {
hostsForTasks[i] = tasks[i].getCurrentAssignedResourceLocation().getHostname();
}
return hostsForTasks;
}
/**
* Parameter {@code uploadedJarLocation} is actually used to point to the local jar, because Flink
* does not support uploading a jar file before hand. Jar files are always uploaded directly when
* a program is submitted.
*/
public void submitTopologyWithOpts(
final String name, final String uploadedJarLocation, final FlinkTopology topology)
throws AlreadyAliveException, InvalidTopologyException {
if (this.getTopologyJobId(name) != null) {
throw new AlreadyAliveException();
}
final URI uploadedJarUri;
final URL uploadedJarUrl;
try {
uploadedJarUri = new File(uploadedJarLocation).getAbsoluteFile().toURI();
uploadedJarUrl = uploadedJarUri.toURL();
JobWithJars.checkJarFile(uploadedJarUrl);
} catch (final IOException e) {
throw new RuntimeException("Problem with jar file " + uploadedJarLocation, e);
}
try {
FlinkClient.addStormConfigToTopology(topology, conf);
} catch (ClassNotFoundException e) {
LOG.error("Could not register class for Kryo serialization.", e);
throw new InvalidTopologyException("Could not register class for Kryo serialization.");
}
final StreamGraph streamGraph = topology.getExecutionEnvironment().getStreamGraph();
streamGraph.setJobName(name);
final JobGraph jobGraph = streamGraph.getJobGraph();
jobGraph.addJar(new Path(uploadedJarUri));
final Configuration configuration = jobGraph.getJobConfiguration();
configuration.setString(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY, jobManagerHost);
configuration.setInteger(ConfigConstants.JOB_MANAGER_IPC_PORT_KEY, jobManagerPort);
final Client client;
try {
client = new Client(configuration);
} catch (IOException e) {
throw new RuntimeException("Could not establish a connection to the job manager", e);
}
try {
ClassLoader classLoader =
JobWithJars.buildUserCodeClassLoader(
Lists.newArrayList(uploadedJarUrl),
Collections.<URL>emptyList(),
this.getClass().getClassLoader());
client.runDetached(jobGraph, classLoader);
} catch (final ProgramInvocationException e) {
throw new RuntimeException("Cannot execute job due to ProgramInvocationException", e);
}
}
@Test
public void testMultipleInstancesPerHost() {
TestLocatableInputSplit[] splits =
new TestLocatableInputSplit[] {
new TestLocatableInputSplit(1, "host1"),
new TestLocatableInputSplit(2, "host1"),
new TestLocatableInputSplit(3, "host2"),
new TestLocatableInputSplit(4, "host2"),
new TestLocatableInputSplit(5, "host3"),
new TestLocatableInputSplit(6, "host3")
};
try {
AbstractJobVertex vertex = new AbstractJobVertex("test vertex");
vertex.setParallelism(6);
vertex.setInvokableClass(DummyInvokable.class);
vertex.setInputSplitSource(new TestInputSplitSource(splits));
JobGraph jobGraph = new JobGraph("test job", vertex);
ExecutionGraph eg =
new ExecutionGraph(
jobGraph.getJobID(), jobGraph.getName(), jobGraph.getJobConfiguration(), TIMEOUT);
eg.attachJobGraph(jobGraph.getVerticesSortedTopologicallyFromSources());
eg.setQueuedSchedulingAllowed(false);
// create a scheduler with 6 instances where always two are on the same host
Scheduler scheduler = new Scheduler();
Instance i1 = getInstance(new byte[] {10, 0, 1, 1}, 12345, "host1", 1);
Instance i2 = getInstance(new byte[] {10, 0, 1, 1}, 12346, "host1", 1);
Instance i3 = getInstance(new byte[] {10, 0, 1, 2}, 12345, "host2", 1);
Instance i4 = getInstance(new byte[] {10, 0, 1, 2}, 12346, "host2", 1);
Instance i5 = getInstance(new byte[] {10, 0, 1, 3}, 12345, "host3", 1);
Instance i6 = getInstance(new byte[] {10, 0, 1, 3}, 12346, "host4", 1);
scheduler.newInstanceAvailable(i1);
scheduler.newInstanceAvailable(i2);
scheduler.newInstanceAvailable(i3);
scheduler.newInstanceAvailable(i4);
scheduler.newInstanceAvailable(i5);
scheduler.newInstanceAvailable(i6);
eg.scheduleForExecution(scheduler);
ExecutionVertex[] tasks = eg.getVerticesTopologically().iterator().next().getTaskVertices();
assertEquals(6, tasks.length);
Instance taskInstance1 = tasks[0].getCurrentAssignedResource().getInstance();
Instance taskInstance2 = tasks[1].getCurrentAssignedResource().getInstance();
Instance taskInstance3 = tasks[2].getCurrentAssignedResource().getInstance();
Instance taskInstance4 = tasks[3].getCurrentAssignedResource().getInstance();
Instance taskInstance5 = tasks[4].getCurrentAssignedResource().getInstance();
Instance taskInstance6 = tasks[5].getCurrentAssignedResource().getInstance();
assertTrue(taskInstance1 == i1 || taskInstance1 == i2);
assertTrue(taskInstance2 == i1 || taskInstance2 == i2);
assertTrue(taskInstance3 == i3 || taskInstance3 == i4);
assertTrue(taskInstance4 == i3 || taskInstance4 == i4);
assertTrue(taskInstance5 == i5 || taskInstance5 == i6);
assertTrue(taskInstance6 == i5 || taskInstance6 == i6);
} catch (Exception e) {
e.printStackTrace();
fail(e.getMessage());
}
}
