Examples with ContainerManager org.apache.hadoop.hdds.scm.container.ContainerManager used on opensource projects

Example 1 with ContainerManager

use of org.apache.hadoop.hdds.scm.container.ContainerManager in project ozone by apache.

the class TestReconAsPassiveScm method testDatanodeRegistrationAndReports.

@Test
public void testDatanodeRegistrationAndReports() throws Exception {
    ReconStorageContainerManagerFacade reconScm = (ReconStorageContainerManagerFacade) cluster.getReconServer().getReconStorageContainerManager();
    StorageContainerManager scm = cluster.getStorageContainerManager();
    PipelineManager reconPipelineManager = reconScm.getPipelineManager();
    PipelineManager scmPipelineManager = scm.getPipelineManager();
    LambdaTestUtils.await(60000, 5000, () -> (reconPipelineManager.getPipelines().size() >= 4));
    // Verify if Recon has all the pipelines from SCM.
    scmPipelineManager.getPipelines().forEach(p -> {
        try {
            assertNotNull(reconPipelineManager.getPipeline(p.getId()));
        } catch (PipelineNotFoundException e) {
            Assert.fail();
        }
    });
    // Verify we can never create a pipeline in Recon.
    LambdaTestUtils.intercept(UnsupportedOperationException.class, "Trying to create pipeline in Recon, which is prohibited!", () -> reconPipelineManager.createPipeline(RatisReplicationConfig.getInstance(ONE)));
    ContainerManager scmContainerManager = scm.getContainerManager();
    assertTrue(scmContainerManager.getContainers().isEmpty());
    // Verify if all the 3 nodes are registered with Recon.
    NodeManager reconNodeManager = reconScm.getScmNodeManager();
    NodeManager scmNodeManager = scm.getScmNodeManager();
    assertEquals(scmNodeManager.getAllNodes().size(), reconNodeManager.getAllNodes().size());
    // Create container
    ContainerManager reconContainerManager = reconScm.getContainerManager();
    ContainerInfo containerInfo = scmContainerManager.allocateContainer(RatisReplicationConfig.getInstance(ONE), "test");
    long containerID = containerInfo.getContainerID();
    Pipeline pipeline = scmPipelineManager.getPipeline(containerInfo.getPipelineID());
    XceiverClientGrpc client = new XceiverClientGrpc(pipeline, conf);
    runTestOzoneContainerViaDataNode(containerID, client);
    // Verify Recon picked up the new container that was created.
    assertEquals(scmContainerManager.getContainerIDs(), reconContainerManager.getContainerIDs());
    GenericTestUtils.LogCapturer logCapturer = GenericTestUtils.LogCapturer.captureLogs(ReconNodeManager.LOG);
    GenericTestUtils.setLogLevel(ReconNodeManager.LOG, Level.DEBUG);
    reconScm.getEventQueue().fireEvent(CLOSE_CONTAINER, containerInfo.containerID());
    GenericTestUtils.waitFor(() -> logCapturer.getOutput().contains("Ignoring unsupported command closeContainerCommand"), 1000, 20000);
}

Example 2 with ContainerManager

use of org.apache.hadoop.hdds.scm.container.ContainerManager in project ozone by apache.

the class TestReconAsPassiveScm method testReconRestart.

@Test
public void testReconRestart() throws Exception {
    final OzoneStorageContainerManager reconScm = cluster.getReconServer().getReconStorageContainerManager();
    StorageContainerManager scm = cluster.getStorageContainerManager();
    // Stop Recon
    ContainerManager scmContainerManager = scm.getContainerManager();
    assertTrue(scmContainerManager.getContainers().isEmpty());
    ContainerManager reconContainerManager = reconScm.getContainerManager();
    assertTrue(reconContainerManager.getContainers().isEmpty());
    LambdaTestUtils.await(60000, 5000, () -> (reconScm.getScmNodeManager().getAllNodes().size() == 3));
    cluster.stopRecon();
    // Create container in SCM.
    ContainerInfo containerInfo = scmContainerManager.allocateContainer(RatisReplicationConfig.getInstance(ONE), "test");
    long containerID = containerInfo.getContainerID();
    PipelineManager scmPipelineManager = scm.getPipelineManager();
    Pipeline pipeline = scmPipelineManager.getPipeline(containerInfo.getPipelineID());
    XceiverClientGrpc client = new XceiverClientGrpc(pipeline, conf);
    runTestOzoneContainerViaDataNode(containerID, client);
    assertFalse(scmContainerManager.getContainers().isEmpty());
    // Close a pipeline
    Optional<Pipeline> pipelineToClose = scmPipelineManager.getPipelines(RatisReplicationConfig.getInstance(ONE)).stream().filter(p -> !p.getId().equals(containerInfo.getPipelineID())).findFirst();
    assertTrue(pipelineToClose.isPresent());
    scmPipelineManager.closePipeline(pipelineToClose.get(), false);
    // Start Recon
    cluster.startRecon();
    // Verify if Recon has all the nodes on restart (even if heartbeats are
    // not yet received).
    NodeManager reconNodeManager = reconScm.getScmNodeManager();
    NodeManager scmNodeManager = scm.getScmNodeManager();
    assertEquals(scmNodeManager.getAllNodes().size(), reconNodeManager.getAllNodes().size());
    // Verify Recon picks up new container, close pipeline SCM actions.
    OzoneStorageContainerManager newReconScm = cluster.getReconServer().getReconStorageContainerManager();
    PipelineManager reconPipelineManager = newReconScm.getPipelineManager();
    assertFalse(reconPipelineManager.containsPipeline(pipelineToClose.get().getId()));
    LambdaTestUtils.await(90000, 5000, () -> (newReconScm.getContainerManager().containerExist(ContainerID.valueOf(containerID))));
}

Example 3 with ContainerManager

use of org.apache.hadoop.hdds.scm.container.ContainerManager in project ozone by apache.

the class TestReconScmSnapshot method testSnapshot.

public static void testSnapshot(MiniOzoneCluster cluster) throws Exception {
    GenericTestUtils.LogCapturer logCapturer = GenericTestUtils.LogCapturer.captureLogs(LoggerFactory.getLogger(ReconStorageContainerManagerFacade.class));
    List<ContainerInfo> reconContainers = cluster.getReconServer().getReconStorageContainerManager().getContainerManager().getContainers();
    assertEquals(0, reconContainers.size());
    ReconNodeManager nodeManager;
    nodeManager = (ReconNodeManager) cluster.getReconServer().getReconStorageContainerManager().getScmNodeManager();
    long keyCountBefore = nodeManager.getNodeDBKeyCount();
    // Stopping Recon to add Containers in SCM
    cluster.stopRecon();
    ContainerManager containerManager;
    containerManager = cluster.getStorageContainerManager().getContainerManager();
    for (int i = 0; i < 10; i++) {
        containerManager.allocateContainer(RatisReplicationConfig.getInstance(HddsProtos.ReplicationFactor.ONE), "testOwner");
    }
    cluster.startRecon();
    // ContainerCount after Recon DB is updated with SCM DB
    containerManager = cluster.getStorageContainerManager().getContainerManager();
    ContainerManager reconContainerManager = cluster.getReconServer().getReconStorageContainerManager().getContainerManager();
    assertTrue(logCapturer.getOutput().contains("Recon Container Count: " + reconContainers.size() + ", SCM Container Count: " + containerManager.getContainers().size()));
    assertEquals(containerManager.getContainers().size(), reconContainerManager.getContainers().size());
    // PipelineCount after Recon DB is updated with SCM DB
    PipelineManager scmPipelineManager = cluster.getStorageContainerManager().getPipelineManager();
    PipelineManager reconPipelineManager = cluster.getReconServer().getReconStorageContainerManager().getPipelineManager();
    assertEquals(scmPipelineManager.getPipelines().size(), reconPipelineManager.getPipelines().size());
    // NodeCount after Recon DB updated with SCM DB
    nodeManager = (ReconNodeManager) cluster.getReconServer().getReconStorageContainerManager().getScmNodeManager();
    long keyCountAfter = nodeManager.getNodeDBKeyCount();
    assertEquals(keyCountAfter, keyCountBefore);
}

Example 4 with ContainerManager

use of org.apache.hadoop.hdds.scm.container.ContainerManager in project ozone by apache.

the class TestScmSafeMode method testSCMSafeMode.

@Test(timeout = 300_000)
public void testSCMSafeMode() throws Exception {
    // Test1: Test safe mode  when there are no containers in system.
    cluster.stop();
    try {
        cluster = builder.build();
    } catch (IOException e) {
        Assert.fail("Cluster startup failed.");
    }
    assertTrue(cluster.getStorageContainerManager().isInSafeMode());
    cluster.startHddsDatanodes();
    cluster.waitForClusterToBeReady();
    cluster.waitTobeOutOfSafeMode();
    assertFalse(cluster.getStorageContainerManager().isInSafeMode());
    // Test2: Test safe mode  when containers are there in system.
    // Create {numKeys} random names keys.
    TestStorageContainerManagerHelper helper = new TestStorageContainerManagerHelper(cluster, conf);
    Map<String, OmKeyInfo> keyLocations = helper.createKeys(100 * 2, 4096);
    final List<ContainerInfo> containers = cluster.getStorageContainerManager().getContainerManager().getContainers();
    GenericTestUtils.waitFor(() -> containers.size() >= 3, 100, 1000 * 30);
    // Removing some container to keep them open.
    containers.remove(0);
    containers.remove(0);
    // Close remaining containers
    ContainerManager mapping = cluster.getStorageContainerManager().getContainerManager();
    containers.forEach(c -> {
        try {
            mapping.updateContainerState(c.containerID(), HddsProtos.LifeCycleEvent.FINALIZE);
            mapping.updateContainerState(c.containerID(), LifeCycleEvent.CLOSE);
        } catch (IOException | InvalidStateTransitionException e) {
            LOG.info("Failed to change state of open containers.", e);
        }
    });
    cluster.stop();
    GenericTestUtils.LogCapturer logCapturer = GenericTestUtils.LogCapturer.captureLogs(SCMSafeModeManager.getLogger());
    logCapturer.clearOutput();
    try {
        cluster = builder.build();
    } catch (IOException ex) {
        fail("failed");
    }
    StorageContainerManager scm;
    scm = cluster.getStorageContainerManager();
    assertTrue(scm.isInSafeMode());
    assertFalse(logCapturer.getOutput().contains("SCM exiting safe mode."));
    assertTrue(scm.getCurrentContainerThreshold() == 0);
    for (HddsDatanodeService dn : cluster.getHddsDatanodes()) {
        dn.start();
    }
    GenericTestUtils.waitFor(() -> scm.getCurrentContainerThreshold() == 1.0, 100, 20000);
    EventQueue eventQueue = (EventQueue) cluster.getStorageContainerManager().getEventQueue();
    eventQueue.processAll(5000L);
    double safeModeCutoff = conf.getDouble(HddsConfigKeys.HDDS_SCM_SAFEMODE_THRESHOLD_PCT, HddsConfigKeys.HDDS_SCM_SAFEMODE_THRESHOLD_PCT_DEFAULT);
    assertTrue(scm.getCurrentContainerThreshold() >= safeModeCutoff);
    assertTrue(logCapturer.getOutput().contains("SCM exiting safe mode."));
    assertFalse(scm.isInSafeMode());
}

Example 5 with ContainerManager

use of org.apache.hadoop.hdds.scm.container.ContainerManager in project ozone by apache.

the class PipelineManagerImpl method closeContainersForPipeline.

/**
 * Fire events to close all containers related to the input pipeline.
 * @param pipelineId - ID of the pipeline.
 * @throws IOException
 */
protected void closeContainersForPipeline(final PipelineID pipelineId) throws IOException {
    Set<ContainerID> containerIDs = stateManager.getContainers(pipelineId);
    ContainerManager containerManager = scmContext.getScm().getContainerManager();
    for (ContainerID containerID : containerIDs) {
        if (containerManager.getContainer(containerID).getState() == HddsProtos.LifeCycleState.OPEN) {
            try {
                containerManager.updateContainerState(containerID, HddsProtos.LifeCycleEvent.FINALIZE);
            } catch (InvalidStateTransitionException ex) {
                throw new IOException(ex);
            }
        }
        eventPublisher.fireEvent(SCMEvents.CLOSE_CONTAINER, containerID);
        LOG.info("Container {} closed for pipeline={}", containerID, pipelineId);
    }
}