Examples with ClusterManagerConfig org.apache.samza.config.ClusterManagerConfig used on opensource projects

Example 1 with ClusterManagerConfig

use of org.apache.samza.config.ClusterManagerConfig in project samza by apache.

the class TestContainerProcessManager method testOnInit.

@Test
public void testOnInit() throws Exception {
    Config conf = getConfig();
    SamzaApplicationState state = new SamzaApplicationState(getJobModelManager(1));
    MockClusterResourceManagerCallback callback = new MockClusterResourceManagerCallback();
    ClusterResourceManager clusterResourceManager = new MockClusterResourceManager(callback, state);
    ClusterManagerConfig clusterManagerConfig = spy(new ClusterManagerConfig(conf));
    ContainerManager containerManager = buildContainerManager(containerPlacementMetadataStore, state, clusterResourceManager, clusterManagerConfig.getHostAffinityEnabled(), false);
    ContainerProcessManager cpm = buildContainerProcessManager(clusterManagerConfig, state, clusterResourceManager, Optional.empty());
    MockContainerAllocatorWithoutHostAffinity allocator = new MockContainerAllocatorWithoutHostAffinity(clusterResourceManager, conf, state, containerManager);
    getPrivateFieldFromCpm("containerAllocator", cpm).set(cpm, allocator);
    CountDownLatch latch = new CountDownLatch(1);
    getPrivateFieldFromCpm("allocatorThread", cpm).set(cpm, new Thread() {

        public void run() {
            isRunning = true;
            latch.countDown();
        }
    });
    cpm.start();
    if (!latch.await(2, TimeUnit.SECONDS)) {
        Assert.fail("timed out waiting for the latch to expire");
    }
    // Verify Allocator thread has started running
    assertTrue(isRunning);
    // Verify the remaining state
    assertEquals(1, state.neededProcessors.get());
    assertEquals(1, allocator.requestedContainers);
    cpm.stop();
}

Example 2 with ClusterManagerConfig

use of org.apache.samza.config.ClusterManagerConfig in project samza by apache.

the class TestContainerProcessManager method testContainerProcessManager.

@Test
public void testContainerProcessManager() throws Exception {
    Map<String, String> conf = new HashMap<>();
    conf.putAll(getConfig());
    conf.put("cluster-manager.container.memory.mb", "500");
    conf.put("cluster-manager.container.cpu.cores", "5");
    SamzaApplicationState state = new SamzaApplicationState(getJobModelManager(1));
    MockClusterResourceManagerCallback callback = new MockClusterResourceManagerCallback();
    MockClusterResourceManager clusterResourceManager = new MockClusterResourceManager(callback, state);
    FaultDomainManager faultDomainManager = mock(FaultDomainManager.class);
    LocalityManager mockLocalityManager = mock(LocalityManager.class);
    when(mockLocalityManager.readLocality()).thenReturn(new LocalityModel(ImmutableMap.of("0", new ProcessorLocality("0", "host1"))));
    ContainerManager containerManager = buildContainerManager(containerPlacementMetadataStore, state, clusterResourceManager, true, false, mockLocalityManager, faultDomainManager);
    ContainerProcessManager cpm = buildContainerProcessManager(new ClusterManagerConfig(new MapConfig(conf)), state, clusterResourceManager, Optional.empty());
    ContainerAllocator allocator = (ContainerAllocator) getPrivateFieldFromCpm("containerAllocator", cpm).get(cpm);
    assertEquals(ContainerAllocator.class, allocator.getClass());
    // Asserts that samza exposed container configs is honored by allocator thread
    assertEquals(500, allocator.containerMemoryMb);
    assertEquals(5, allocator.containerNumCpuCores);
    conf.clear();
    conf.putAll(getConfigWithHostAffinity());
    conf.put("cluster-manager.container.memory.mb", "500");
    conf.put("cluster-manager.container.cpu.cores", "5");
    state = new SamzaApplicationState(getJobModelManager(1));
    callback = new MockClusterResourceManagerCallback();
    clusterResourceManager = new MockClusterResourceManager(callback, state);
    cpm = new ContainerProcessManager(new ClusterManagerConfig(new MapConfig(conf)), state, new MetricsRegistryMap(), clusterResourceManager, Optional.empty(), containerManager, mockLocalityManager, false);
    allocator = (ContainerAllocator) getPrivateFieldFromCpm("containerAllocator", cpm).get(cpm);
    assertEquals(ContainerAllocator.class, allocator.getClass());
    // Asserts that samza exposed container configs is honored by allocator thread
    assertEquals(500, allocator.containerMemoryMb);
    assertEquals(5, allocator.containerNumCpuCores);
}

Example 3 with ClusterManagerConfig

use of org.apache.samza.config.ClusterManagerConfig in project samza by apache.

the class TestContainerProcessManager method testContainerRequestedRetriesExceedingWindowOnFailureWithUnknownCode.

private void testContainerRequestedRetriesExceedingWindowOnFailureWithUnknownCode(boolean withHostAffinity, boolean failAfterRetries) throws Exception {
    int maxRetries = 3;
    String processorId = "0";
    ClusterManagerConfig clusterManagerConfig = new ClusterManagerConfig(getConfigWithHostAffinityAndRetries(withHostAffinity, maxRetries, failAfterRetries));
    SamzaApplicationState state = new SamzaApplicationState(getJobModelManager(1));
    MockClusterResourceManagerCallback callback = new MockClusterResourceManagerCallback();
    MockClusterResourceManager clusterResourceManager = new MockClusterResourceManager(callback, state);
    ContainerManager containerManager = buildContainerManager(containerPlacementMetadataStore, state, clusterResourceManager, clusterManagerConfig.getHostAffinityEnabled(), false);
    MockContainerAllocatorWithoutHostAffinity allocator = new MockContainerAllocatorWithoutHostAffinity(clusterResourceManager, clusterManagerConfig, state, containerManager);
    ContainerProcessManager cpm = buildContainerProcessManager(clusterManagerConfig, state, clusterResourceManager, Optional.of(allocator));
    // start triggers a request
    cpm.start();
    assertFalse(cpm.shouldShutdown());
    assertEquals(1, allocator.getContainerRequestState().numPendingRequests());
    SamzaResource container = new SamzaResource(1, 1024, "host1", "id0");
    cpm.onResourceAllocated(container);
    // Allow container to run and update state
    if (!allocator.awaitContainersStart(1, 2, TimeUnit.SECONDS)) {
        fail("timed out waiting for the containers to start");
    }
    cpm.onStreamProcessorLaunchSuccess(container);
    // Mock 2nd failure exceeding retry window.
    int longWindow = clusterManagerConfig.getContainerRetryWindowMs() + 10;
    cpm.getProcessorFailures().put(processorId, new ProcessorFailure(1, Instant.now().minusMillis(longWindow), Duration.ZERO));
    assertEquals(1, cpm.getProcessorFailures().get(processorId).getCount());
    cpm.onResourceCompleted(new SamzaResourceStatus(container.getContainerId(), "diagnostics", 1));
    assertEquals(false, cpm.getJobFailureCriteriaMet());
    assertEquals(1, cpm.getProcessorFailures().get(processorId).getCount());
    cpm.onResourceAllocated(container);
    // Allow container to run and update state
    if (!allocator.awaitContainersStart(1, 2, TimeUnit.SECONDS)) {
        fail("timed out waiting for the containers to start");
    }
    cpm.onStreamProcessorLaunchSuccess(container);
    // Mock 3rd failure exceeding retry window.
    cpm.getProcessorFailures().put(processorId, new ProcessorFailure(2, Instant.now().minusMillis(longWindow), Duration.ZERO));
    cpm.onResourceCompleted(new SamzaResourceStatus(container.getContainerId(), "diagnostics", 1));
    assertEquals(false, cpm.getJobFailureCriteriaMet());
    assertEquals(1, cpm.getProcessorFailures().get(processorId).getCount());
    cpm.onResourceAllocated(container);
    // Allow container to run and update state
    if (!allocator.awaitContainersStart(1, 2, TimeUnit.SECONDS)) {
        fail("timed out waiting for the containers to start");
    }
    cpm.onStreamProcessorLaunchSuccess(container);
    // Mock 4th failure exceeding retry window.
    cpm.getProcessorFailures().put(processorId, new ProcessorFailure(3, Instant.now().minusMillis(longWindow), Duration.ZERO));
    cpm.onResourceCompleted(new SamzaResourceStatus(container.getContainerId(), "diagnostics", 1));
    assertEquals(false, cpm.getJobFailureCriteriaMet());
    assertEquals(1, cpm.getProcessorFailures().get(processorId).getCount());
    cpm.stop();
}

Example 4 with ClusterManagerConfig

use of org.apache.samza.config.ClusterManagerConfig in project samza by apache.

the class TestContainerProcessManager method testRerequestOnAnyHostIfContainerStartFails.

@Test
public void testRerequestOnAnyHostIfContainerStartFails() throws Exception {
    SamzaApplicationState state = new SamzaApplicationState(getJobModelManager(1));
    Map<String, String> configMap = new HashMap<>();
    configMap.putAll(getConfig());
    MockClusterResourceManagerCallback callback = new MockClusterResourceManagerCallback();
    MockClusterResourceManager clusterResourceManager = new MockClusterResourceManager(callback, state);
    FaultDomainManager faultDomainManager = mock(FaultDomainManager.class);
    LocalityManager mockLocalityManager = mock(LocalityManager.class);
    when(mockLocalityManager.readLocality()).thenReturn(new LocalityModel(ImmutableMap.of("0", new ProcessorLocality("1", "host1"))));
    ContainerManager containerManager = buildContainerManager(containerPlacementMetadataStore, state, clusterResourceManager, Boolean.valueOf(config.get(ClusterManagerConfig.HOST_AFFINITY_ENABLED)), false, mockLocalityManager, faultDomainManager);
    MockContainerAllocatorWithoutHostAffinity allocator = new MockContainerAllocatorWithoutHostAffinity(clusterResourceManager, new MapConfig(config), state, containerManager);
    ContainerProcessManager manager = new ContainerProcessManager(new ClusterManagerConfig(config), state, new MetricsRegistryMap(), clusterResourceManager, Optional.of(allocator), containerManager, mockLocalityManager, false);
    manager.start();
    SamzaResource resource = new SamzaResource(1, 1024, "host1", "resource-1");
    state.pendingProcessors.put("1", resource);
    Assert.assertEquals(clusterResourceManager.resourceRequests.size(), 1);
    manager.onStreamProcessorLaunchFailure(resource, new Exception("cannot launch container!"));
    Assert.assertEquals(clusterResourceManager.resourceRequests.size(), 2);
    Assert.assertEquals(clusterResourceManager.resourceRequests.get(1).getHost(), ResourceRequestState.ANY_HOST);
    manager.stop();
}

Example 5 with ClusterManagerConfig

use of org.apache.samza.config.ClusterManagerConfig in project samza by apache.

the class TestContainerProcessManager method testNewContainerRequestedOnFailureWithUnknownCode.

/**
 * Test Container Process Manager should request a new container when a task fails with unknown exit code
 * When host-affinity is not enabled, it will always request for ANY_HOST
 */
@Test
public void testNewContainerRequestedOnFailureWithUnknownCode() throws Exception {
    Config conf = getConfig();
    SamzaApplicationState state = new SamzaApplicationState(getJobModelManager(1));
    MockClusterResourceManagerCallback callback = new MockClusterResourceManagerCallback();
    MockClusterResourceManager clusterResourceManager = new MockClusterResourceManager(callback, state);
    ClusterManagerConfig clusterManagerConfig = spy(new ClusterManagerConfig(conf));
    ContainerManager containerManager = buildContainerManager(containerPlacementMetadataStore, state, clusterResourceManager, clusterManagerConfig.getHostAffinityEnabled(), false);
    MockContainerAllocatorWithoutHostAffinity allocator = new MockContainerAllocatorWithoutHostAffinity(clusterResourceManager, conf, state, containerManager);
    ContainerProcessManager cpm = spy(buildContainerProcessManager(clusterManagerConfig, state, clusterResourceManager, Optional.of(allocator)));
    // start triggers a request
    cpm.start();
    verify(clusterManagerConfig, never()).getContainerPreferredHostLastRetryDelayMs();
    verify(cpm, never()).onResourceCompletedWithUnknownStatus(any(), anyString(), anyString(), anyInt());
    assertFalse(cpm.shouldShutdown());
    assertEquals(1, allocator.getContainerRequestState().numPendingRequests());
    SamzaResource container = new SamzaResource(1, 1024, "host1", "id0");
    cpm.onResourceAllocated(container);
    // Allow container to run and update state
    if (!allocator.awaitContainersStart(1, 2, TimeUnit.SECONDS)) {
        fail("timed out waiting for the containers to start");
    }
    cpm.onStreamProcessorLaunchSuccess(container);
    // Create first container failure
    SamzaResourceStatus samzaResourceStatus = new SamzaResourceStatus(container.getContainerId(), "diagnostics", 1);
    cpm.onResourceCompleted(samzaResourceStatus);
    // The above failure should trigger a container request
    verify(cpm).onResourceCompletedWithUnknownStatus(eq(samzaResourceStatus), eq(container.getContainerId()), eq("0"), eq(1));
    verify(clusterManagerConfig, never()).getContainerPreferredHostLastRetryDelayMs();
    assertEquals(1, allocator.getContainerRequestState().numPendingRequests());
    assertEquals(ResourceRequestState.ANY_HOST, allocator.getContainerRequestState().peekPendingRequest().getPreferredHost());
    assertFalse(cpm.shouldShutdown());
    assertFalse(state.jobHealthy.get());
    assertEquals(2, clusterResourceManager.resourceRequests.size());
    assertEquals(0, clusterResourceManager.releasedResources.size());
    cpm.onResourceAllocated(container);
    // Allow container to run and update state
    if (!allocator.awaitContainersStart(1, 2, TimeUnit.SECONDS)) {
        fail("timed out waiting for the containers to start");
    }
    cpm.onStreamProcessorLaunchSuccess(container);
    assertTrue(state.jobHealthy.get());
    // Create a second failure
    cpm.onResourceCompleted(samzaResourceStatus);
    // The above failure should trigger a job shutdown because our retry count is set to 1
    verify(cpm, times(2)).onResourceCompletedWithUnknownStatus(eq(samzaResourceStatus), eq(container.getContainerId()), eq("0"), eq(1));
    verify(clusterManagerConfig, never()).getContainerPreferredHostLastRetryDelayMs();
    assertEquals(0, allocator.getContainerRequestState().numPendingRequests());
    assertEquals(2, clusterResourceManager.resourceRequests.size());
    assertEquals(0, clusterResourceManager.releasedResources.size());
    assertFalse(state.jobHealthy.get());
    assertTrue(cpm.shouldShutdown());
    assertEquals(SamzaApplicationState.SamzaAppStatus.FAILED, state.status);
    cpm.stop();
}