use of io.mantisrx.server.core.Status in project mantis by Netflix.
the class StatusEventBrokerActor method onStatusEvent.
// sends JobStatus messages to active connections by jobId
private void onStatusEvent(final LifecycleEventsProto.StatusEvent se) {
Status status = LifecycleEventsProto.from(se);
String jobId = status.getJobId();
// add Status to job event history
jobIdToStatusEventsBuf.computeIfAbsent(jobId, (j) -> EvictingQueue.create(MAX_STATUS_HISTORY_PER_JOB)).add(status);
cleanupIfTerminalState(se);
Set<ActorRef> jobStatusActiveConnections = jobIdToActorMap.get(jobId);
if (jobStatusActiveConnections != null && !jobStatusActiveConnections.isEmpty()) {
logger.debug("Sending job status {}", se);
jobStatusActiveConnections.forEach(connActor -> connActor.tell(new JobStatus(status), self()));
} else {
logger.debug("Job status dropped, no active subscribers for {}", jobId);
}
if (se instanceof LifecycleEventsProto.WorkerStatusEvent) {
this.agentsErrorMonitorActorRef.tell(se, getSelf());
}
}
use of io.mantisrx.server.core.Status in project mantis by Netflix.
the class JobClusterTest method testZombieWorkerKilledOnMessage.
@Test
public void testZombieWorkerKilledOnMessage() {
String clusterName = "testZombieWorkerKilledOnMessage";
TestKit probe = new TestKit(system);
MantisScheduler schedulerMock = mock(MantisScheduler.class);
MantisJobStore jobStoreMock = mock(MantisJobStore.class);
final JobClusterDefinitionImpl fakeJobCluster = createFakeJobClusterDefn(clusterName);
ActorRef jobClusterActor = system.actorOf(props(clusterName, jobStoreMock, schedulerMock, eventPublisher));
jobClusterActor.tell(new JobClusterProto.InitializeJobClusterRequest(fakeJobCluster, user, probe.getRef()), probe.getRef());
JobClusterProto.InitializeJobClusterResponse createResp = probe.expectMsgClass(JobClusterProto.InitializeJobClusterResponse.class);
assertEquals(SUCCESS, createResp.responseCode);
try {
String jobId = clusterName + "-1";
WorkerId workerId = new WorkerId(clusterName, jobId, 0, 1);
WorkerEvent heartBeat2 = new WorkerHeartbeat(new Status(jobId, 1, workerId.getWorkerIndex(), workerId.getWorkerNum(), TYPE.HEARTBEAT, "", MantisJobState.Started, System.currentTimeMillis()));
jobClusterActor.tell(heartBeat2, probe.getRef());
jobClusterActor.tell(new GetJobClusterRequest(clusterName), probe.getRef());
GetJobClusterResponse resp = probe.expectMsgClass(GetJobClusterResponse.class);
assertEquals(clusterName, resp.getJobCluster().get().getName());
verify(schedulerMock, times(1)).unscheduleAndTerminateWorker(workerId, empty());
} catch (Exception e) {
e.printStackTrace();
fail();
}
}
use of io.mantisrx.server.core.Status in project mantis by Netflix.
the class JobClusterManagerTest method testBootStrapJobClustersAndJobs.
@Test
public void testBootStrapJobClustersAndJobs() {
TestKit probe = new TestKit(system);
JobTestHelper.deleteAllFiles();
MantisJobStore jobStore = new MantisJobStore(new MantisStorageProviderAdapter(new io.mantisrx.server.master.store.SimpleCachedFileStorageProvider(), eventPublisher));
MantisJobStore jobStoreSpied = Mockito.spy(jobStore);
MantisScheduler schedulerMock = mock(MantisScheduler.class);
ActorRef jobClusterManagerActor = system.actorOf(JobClustersManagerActor.props(jobStoreSpied, eventPublisher));
jobClusterManagerActor.tell(new JobClusterManagerProto.JobClustersManagerInitialize(schedulerMock, false), probe.getRef());
JobClustersManagerInitializeResponse iResponse = probe.expectMsgClass(Duration.of(10, ChronoUnit.MINUTES), JobClustersManagerInitializeResponse.class);
List<String> clusterNames = Lists.newArrayList("testBootStrapJobClustersAndJobs1", "testBootStrapJobClustersAndJobs2", "testBootStrapJobClustersAndJobs3");
String clusterWithNoJob = "testBootStrapJobClusterWithNoJob";
createJobClusterAndAssert(jobClusterManagerActor, clusterWithNoJob);
WorkerMigrationConfig migrationConfig = new WorkerMigrationConfig(MigrationStrategyEnum.PERCENTAGE, "{\"percentToMove\":60, \"intervalMs\":30000}");
// Create 3 clusters and submit 1 job each
for (String cluster : clusterNames) {
createJobClusterAndAssert(jobClusterManagerActor, cluster, migrationConfig);
submitJobAndAssert(jobClusterManagerActor, cluster);
if (cluster.equals("testBootStrapJobClustersAndJobs1")) {
// send worker events for job 1 so it goes to started state
String jobId = "testBootStrapJobClustersAndJobs1-1";
WorkerId workerId = new WorkerId(jobId, 0, 1);
WorkerEvent launchedEvent = new WorkerLaunched(workerId, 0, "host1", "vm1", empty(), new WorkerPorts(Lists.newArrayList(8000, 9000, 9010, 9020, 9030)));
jobClusterManagerActor.tell(launchedEvent, probe.getRef());
WorkerEvent startInitEvent = new WorkerStatus(new Status(workerId.getJobId(), 1, workerId.getWorkerIndex(), workerId.getWorkerNum(), TYPE.INFO, "test START_INIT", MantisJobState.StartInitiated));
jobClusterManagerActor.tell(startInitEvent, probe.getRef());
WorkerEvent heartBeat = new WorkerHeartbeat(new Status(jobId, 1, workerId.getWorkerIndex(), workerId.getWorkerNum(), TYPE.HEARTBEAT, "", MantisJobState.Started));
jobClusterManagerActor.tell(heartBeat, probe.getRef());
// get Job status
jobClusterManagerActor.tell(new GetJobDetailsRequest("user", JobId.fromId(jobId).get()), probe.getRef());
GetJobDetailsResponse resp2 = probe.expectMsgClass(GetJobDetailsResponse.class);
// Ensure its launched
assertEquals(SUCCESS, resp2.responseCode);
assertEquals(JobState.Launched, resp2.getJobMetadata().get().getState());
}
}
// kill 1 of the jobs to test archive path
JobClusterManagerProto.KillJobRequest killRequest = new JobClusterManagerProto.KillJobRequest("testBootStrapJobClustersAndJobs2-1", JobCompletedReason.Killed.toString(), "njoshi");
jobClusterManagerActor.tell(killRequest, probe.getRef());
JobClusterManagerProto.KillJobResponse killJobResponse = probe.expectMsgClass(JobClusterManagerProto.KillJobResponse.class);
assertEquals(SUCCESS, killJobResponse.responseCode);
JobTestHelper.sendWorkerTerminatedEvent(probe, jobClusterManagerActor, "testBootStrapJobClustersAndJobs2-1", new WorkerId("testBootStrapJobClustersAndJobs2-1", 0, 1));
try {
Thread.sleep(500);
} catch (InterruptedException e) {
e.printStackTrace();
}
// Stop job cluster Manager Actor
system.stop(jobClusterManagerActor);
// create new instance
jobClusterManagerActor = system.actorOf(JobClustersManagerActor.props(jobStoreSpied, eventPublisher));
// initialize it
jobClusterManagerActor.tell(new JobClusterManagerProto.JobClustersManagerInitialize(schedulerMock, true), probe.getRef());
JobClustersManagerInitializeResponse initializeResponse = probe.expectMsgClass(JobClustersManagerInitializeResponse.class);
// probe.expectMsgClass(Duration.of(10, ChronoUnit.MINUTES),JobClusterManagerProto.JobClustersManagerInitializeResponse.class);
// probe.expectMsgClass(JobClusterManagerProto.JobClustersManagerInitializeResponse.class);
assertEquals(SUCCESS, initializeResponse.responseCode);
// Get Cluster Config
jobClusterManagerActor.tell(new GetJobClusterRequest("testBootStrapJobClustersAndJobs1"), probe.getRef());
GetJobClusterResponse clusterResponse = probe.expectMsgClass(GetJobClusterResponse.class);
assertEquals(SUCCESS, clusterResponse.responseCode);
assertTrue(clusterResponse.getJobCluster().isPresent());
WorkerMigrationConfig mConfig = clusterResponse.getJobCluster().get().getMigrationConfig();
assertEquals(migrationConfig.getStrategy(), mConfig.getStrategy());
assertEquals(migrationConfig.getConfigString(), migrationConfig.getConfigString());
// get Job status
jobClusterManagerActor.tell(new GetJobDetailsRequest("user", JobId.fromId("testBootStrapJobClustersAndJobs1-1").get()), probe.getRef());
GetJobDetailsResponse resp2 = probe.expectMsgClass(GetJobDetailsResponse.class);
// Ensure its launched
System.out.println("Resp2 -> " + resp2.message);
assertEquals(SUCCESS, resp2.responseCode);
assertEquals(JobState.Launched, resp2.getJobMetadata().get().getState());
// 1 jobs should be in completed state
jobClusterManagerActor.tell(new GetJobDetailsRequest("user", JobId.fromId("testBootStrapJobClustersAndJobs2-1").get()), probe.getRef());
resp2 = probe.expectMsgClass(Duration.of(10, ChronoUnit.MINUTES), GetJobDetailsResponse.class);
// Ensure its completed
assertEquals(SUCCESS, resp2.responseCode);
assertEquals(JobState.Completed, resp2.getJobMetadata().get().getState());
jobClusterManagerActor.tell(new GetJobDetailsRequest("user", JobId.fromId("testBootStrapJobClustersAndJobs3-1").get()), probe.getRef());
resp2 = probe.expectMsgClass(Duration.of(10, ChronoUnit.MINUTES), GetJobDetailsResponse.class);
// Ensure its Accepted
assertEquals(SUCCESS, resp2.responseCode);
assertEquals(JobState.Accepted, resp2.getJobMetadata().get().getState());
try {
Optional<JobWorker> workerByIndex = resp2.getJobMetadata().get().getWorkerByIndex(1, 0);
assertTrue(workerByIndex.isPresent());
Optional<IMantisStageMetadata> stageMetadata = resp2.getJobMetadata().get().getStageMetadata(1);
assertTrue(stageMetadata.isPresent());
JobWorker workerByIndex1 = stageMetadata.get().getWorkerByIndex(0);
System.out.println("Got worker by index : " + workerByIndex1);
Optional<JobWorker> worker = resp2.getJobMetadata().get().getWorkerByNumber(1);
assertTrue(worker.isPresent());
} catch (io.mantisrx.server.master.persistence.exceptions.InvalidJobException e) {
e.printStackTrace();
}
jobClusterManagerActor.tell(new GetLastSubmittedJobIdStreamRequest("testBootStrapJobClustersAndJobs1"), probe.getRef());
GetLastSubmittedJobIdStreamResponse lastSubmittedJobIdStreamResponse = probe.expectMsgClass(Duration.of(10, ChronoUnit.MINUTES), GetLastSubmittedJobIdStreamResponse.class);
lastSubmittedJobIdStreamResponse.getjobIdBehaviorSubject().get().take(1).toBlocking().subscribe((jId) -> {
assertEquals(new JobId("testBootStrapJobClustersAndJobs1", 1), jId);
});
jobClusterManagerActor.tell(new GetJobClusterRequest(clusterWithNoJob), probe.getRef());
GetJobClusterResponse jobClusterResponse = probe.expectMsgClass(Duration.of(10, ChronoUnit.MINUTES), GetJobClusterResponse.class);
assertEquals(SUCCESS, jobClusterResponse.responseCode);
assertTrue(jobClusterResponse.getJobCluster().isPresent());
assertEquals(clusterWithNoJob, jobClusterResponse.getJobCluster().get().getName());
// 1 running worker
verify(schedulerMock, timeout(100_1000).times(1)).initializeRunningWorker(any(), any());
// 2 worker schedule requests
verify(schedulerMock, timeout(100_000).times(4)).scheduleWorker(any());
try {
Mockito.verify(jobStoreSpied).loadAllArchivedJobsAsync();
Mockito.verify(jobStoreSpied).loadAllActiveJobs();
Mockito.verify(jobStoreSpied).loadAllCompletedJobs();
Mockito.verify(jobStoreSpied).archiveWorker(any());
Mockito.verify(jobStoreSpied).archiveJob(any());
} catch (IOException e) {
e.printStackTrace();
fail();
}
}
use of io.mantisrx.server.core.Status in project mantis by Netflix.
the class JobTestHelper method sendStartInitiatedEvent.
public static void sendStartInitiatedEvent(final TestKit probe, final ActorRef jobActor, final int stageNum, WorkerId workerId) {
WorkerEvent startInitEvent = new WorkerStatus(new Status(workerId.getJobId(), stageNum, workerId.getWorkerIndex(), workerId.getWorkerNum(), TYPE.INFO, "test START_INITIATED event", MantisJobState.StartInitiated));
jobActor.tell(startInitEvent, probe.getRef());
}
use of io.mantisrx.server.core.Status in project mantis by Netflix.
the class WorkerExecutionOperationsNetworkStage method connectToObservableAtPreviousStages.
@SuppressWarnings({ "rawtypes" })
private WorkerConsumer connectToObservableAtPreviousStages(Observable<JobSchedulingInfo> selfSchedulingInfo, final String jobId, final int previousStageNum, int numInstanceAtPreviousStage, final StageConfig previousStage, final AtomicBoolean acceptSchedulingChanges, final Observer<Status> jobStatus, final int stageNumToExecute, final int workerIndex, final int workerNumber) {
logger.info("Watching for scheduling changes");
// Observable<List<Endpoint>> schedulingUpdates = mantisMasterApi.schedulingChanges(jobId)
Observable<List<Endpoint>> schedulingUpdates = selfSchedulingInfo.flatMap((Func1<JobSchedulingInfo, Observable<WorkerAssignments>>) schedulingChange -> {
Map<Integer, WorkerAssignments> assignments = schedulingChange.getWorkerAssignments();
if (assignments != null && !assignments.isEmpty()) {
return Observable.from(assignments.values());
} else {
return Observable.empty();
}
}).filter(assignments -> (assignments.getStage() == previousStageNum) && acceptSchedulingChanges.get()).map(assignments -> {
List<Endpoint> endpoints = new LinkedList<>();
for (WorkerHost host : assignments.getHosts().values()) {
if (host.getState() == MantisJobState.Started) {
logger.info("Received scheduling update from master, connect request for host: " + host.getHost() + " port: " + host.getPort() + " state: " + host.getState() + " adding: " + connectionsPerEndpoint + " connections to host");
for (int i = 1; i <= connectionsPerEndpoint; i++) {
final String endpointId = "stage_" + stageNumToExecute + "_index_" + Integer.toString(workerIndex) + "_partition_" + i;
logger.info("Adding endpoint to endpoint injector to be considered for add, with id: " + endpointId);
endpoints.add(new Endpoint(host.getHost(), host.getPort().get(0), endpointId));
}
}
}
return endpoints;
}).filter(t1 -> (t1.size() > 0));
String name = jobId + "_" + previousStageNum;
return new WorkerConsumerRemoteObservable(name, new ToDeltaEndpointInjector(schedulingUpdates));
}
Aggregations