Search in sources :

Example 1 with ContainerInfo

use of org.apache.hadoop.hdds.scm.container.ContainerInfo 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);
}
Also used : ReconStorageContainerManagerFacade(org.apache.hadoop.ozone.recon.scm.ReconStorageContainerManagerFacade) ReconNodeManager(org.apache.hadoop.ozone.recon.scm.ReconNodeManager) NodeManager(org.apache.hadoop.hdds.scm.node.NodeManager) OzoneStorageContainerManager(org.apache.hadoop.hdds.scm.server.OzoneStorageContainerManager) StorageContainerManager(org.apache.hadoop.hdds.scm.server.StorageContainerManager) OzoneStorageContainerManager(org.apache.hadoop.hdds.scm.server.OzoneStorageContainerManager) ContainerManager(org.apache.hadoop.hdds.scm.container.ContainerManager) StorageContainerManager(org.apache.hadoop.hdds.scm.server.StorageContainerManager) PipelineManager(org.apache.hadoop.hdds.scm.pipeline.PipelineManager) XceiverClientGrpc(org.apache.hadoop.hdds.scm.XceiverClientGrpc) ContainerInfo(org.apache.hadoop.hdds.scm.container.ContainerInfo) PipelineNotFoundException(org.apache.hadoop.hdds.scm.pipeline.PipelineNotFoundException) GenericTestUtils(org.apache.ozone.test.GenericTestUtils) Pipeline(org.apache.hadoop.hdds.scm.pipeline.Pipeline) Test(org.junit.Test)

Example 2 with ContainerInfo

use of org.apache.hadoop.hdds.scm.container.ContainerInfo 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))));
}
Also used : OzoneConfiguration(org.apache.hadoop.hdds.conf.OzoneConfiguration) ReconNodeManager(org.apache.hadoop.ozone.recon.scm.ReconNodeManager) NodeManager(org.apache.hadoop.hdds.scm.node.NodeManager) MiniOzoneCluster(org.apache.hadoop.ozone.MiniOzoneCluster) OzoneStorageContainerManager(org.apache.hadoop.hdds.scm.server.OzoneStorageContainerManager) ONE(org.apache.hadoop.hdds.protocol.proto.HddsProtos.ReplicationFactor.ONE) PipelineNotFoundException(org.apache.hadoop.hdds.scm.pipeline.PipelineNotFoundException) After(org.junit.After) Level(org.slf4j.event.Level) Timeout(org.junit.rules.Timeout) ContainerManager(org.apache.hadoop.hdds.scm.container.ContainerManager) StorageContainerManager(org.apache.hadoop.hdds.scm.server.StorageContainerManager) LambdaTestUtils(org.apache.ozone.test.LambdaTestUtils) TestOzoneContainer.runTestOzoneContainerViaDataNode(org.apache.hadoop.ozone.container.ozoneimpl.TestOzoneContainer.runTestOzoneContainerViaDataNode) Before(org.junit.Before) HDDS_CONTAINER_REPORT_INTERVAL(org.apache.hadoop.hdds.HddsConfigKeys.HDDS_CONTAINER_REPORT_INTERVAL) RatisReplicationConfig(org.apache.hadoop.hdds.client.RatisReplicationConfig) Assert.assertNotNull(org.junit.Assert.assertNotNull) Assert.assertTrue(org.junit.Assert.assertTrue) ContainerID(org.apache.hadoop.hdds.scm.container.ContainerID) Pipeline(org.apache.hadoop.hdds.scm.pipeline.Pipeline) Test(org.junit.Test) PipelineManager(org.apache.hadoop.hdds.scm.pipeline.PipelineManager) XceiverClientGrpc(org.apache.hadoop.hdds.scm.XceiverClientGrpc) CLOSE_CONTAINER(org.apache.hadoop.hdds.scm.events.SCMEvents.CLOSE_CONTAINER) Rule(org.junit.Rule) HDDS_PIPELINE_REPORT_INTERVAL(org.apache.hadoop.hdds.HddsConfigKeys.HDDS_PIPELINE_REPORT_INTERVAL) Assert.assertFalse(org.junit.Assert.assertFalse) Optional(java.util.Optional) ReconStorageContainerManagerFacade(org.apache.hadoop.ozone.recon.scm.ReconStorageContainerManagerFacade) Assert(org.junit.Assert) ContainerInfo(org.apache.hadoop.hdds.scm.container.ContainerInfo) GenericTestUtils(org.apache.ozone.test.GenericTestUtils) TemporaryFolder(org.junit.rules.TemporaryFolder) Assert.assertEquals(org.junit.Assert.assertEquals) ReconNodeManager(org.apache.hadoop.ozone.recon.scm.ReconNodeManager) NodeManager(org.apache.hadoop.hdds.scm.node.NodeManager) OzoneStorageContainerManager(org.apache.hadoop.hdds.scm.server.OzoneStorageContainerManager) StorageContainerManager(org.apache.hadoop.hdds.scm.server.StorageContainerManager) OzoneStorageContainerManager(org.apache.hadoop.hdds.scm.server.OzoneStorageContainerManager) ContainerManager(org.apache.hadoop.hdds.scm.container.ContainerManager) StorageContainerManager(org.apache.hadoop.hdds.scm.server.StorageContainerManager) PipelineManager(org.apache.hadoop.hdds.scm.pipeline.PipelineManager) XceiverClientGrpc(org.apache.hadoop.hdds.scm.XceiverClientGrpc) ContainerInfo(org.apache.hadoop.hdds.scm.container.ContainerInfo) OzoneStorageContainerManager(org.apache.hadoop.hdds.scm.server.OzoneStorageContainerManager) Pipeline(org.apache.hadoop.hdds.scm.pipeline.Pipeline) Test(org.junit.Test)

Example 3 with ContainerInfo

use of org.apache.hadoop.hdds.scm.container.ContainerInfo 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);
}
Also used : ReconStorageContainerManagerFacade(org.apache.hadoop.ozone.recon.scm.ReconStorageContainerManagerFacade) ReconNodeManager(org.apache.hadoop.ozone.recon.scm.ReconNodeManager) ContainerManager(org.apache.hadoop.hdds.scm.container.ContainerManager) PipelineManager(org.apache.hadoop.hdds.scm.pipeline.PipelineManager) ContainerInfo(org.apache.hadoop.hdds.scm.container.ContainerInfo) GenericTestUtils(org.apache.ozone.test.GenericTestUtils)

Example 4 with ContainerInfo

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

the class TestDecommissionAndMaintenance method waitForAndReturnContainer.

/**
 * Get any container present in the cluster and wait to ensure 3 replicas
 * have been reported before returning the container.
 * @return A single container present on the cluster
 * @throws Exception
 */
private ContainerInfo waitForAndReturnContainer() throws Exception {
    final ContainerInfo container = cm.getContainers().get(0);
    // Ensure all 3 replicas of the container have been reported via ICR
    waitForContainerReplicas(container, 3);
    return container;
}
Also used : ContainerInfo(org.apache.hadoop.hdds.scm.container.ContainerInfo)

Example 5 with ContainerInfo

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

the class TestDecommissionAndMaintenance method testContainerIsReplicatedWhenAllNodesGotoMaintenance.

@Test
public // return, the excess replicas should be removed.
void testContainerIsReplicatedWhenAllNodesGotoMaintenance() throws Exception {
    // Generate some data on the empty cluster to create some containers
    generateData(20, "key", ReplicationFactor.THREE, ReplicationType.RATIS);
    // Locate any container and find its open pipeline
    final ContainerInfo container = waitForAndReturnContainer();
    Set<ContainerReplica> replicas = getContainerReplicas(container);
    List<DatanodeDetails> forMaintenance = new ArrayList<>();
    replicas.forEach(r -> forMaintenance.add(r.getDatanodeDetails()));
    scmClient.startMaintenanceNodes(forMaintenance.stream().map(d -> getDNHostAndPort(d)).collect(Collectors.toList()), 0);
    // Ensure all 3 DNs go to maintenance
    for (DatanodeDetails dn : forMaintenance) {
        waitForDnToReachPersistedOpState(dn, IN_MAINTENANCE);
    }
    // There should now be 5-6 replicas of the container we are tracking
    Set<ContainerReplica> newReplicas = cm.getContainerReplicas(container.containerID());
    assertTrue(newReplicas.size() >= 5);
    scmClient.recommissionNodes(forMaintenance.stream().map(d -> getDNHostAndPort(d)).collect(Collectors.toList()));
    // Ensure all 3 DNs go to maintenance
    for (DatanodeDetails dn : forMaintenance) {
        waitForDnToReachOpState(dn, IN_SERVICE);
    }
    waitForContainerReplicas(container, 3);
}
Also used : ContainerReplica(org.apache.hadoop.hdds.scm.container.ContainerReplica) DatanodeDetails(org.apache.hadoop.hdds.protocol.DatanodeDetails) ContainerInfo(org.apache.hadoop.hdds.scm.container.ContainerInfo) ArrayList(java.util.ArrayList) Test(org.junit.Test)

Aggregations

ContainerInfo (org.apache.hadoop.hdds.scm.container.ContainerInfo)158 Test (org.junit.Test)100 ContainerReplica (org.apache.hadoop.hdds.scm.container.ContainerReplica)54 Pipeline (org.apache.hadoop.hdds.scm.pipeline.Pipeline)48 DatanodeDetails (org.apache.hadoop.hdds.protocol.DatanodeDetails)42 ContainerReplicaCount (org.apache.hadoop.hdds.scm.container.ContainerReplicaCount)38 ArrayList (java.util.ArrayList)28 ContainerWithPipeline (org.apache.hadoop.hdds.scm.container.common.helpers.ContainerWithPipeline)26 ContainerID (org.apache.hadoop.hdds.scm.container.ContainerID)25 OmKeyArgs (org.apache.hadoop.ozone.om.helpers.OmKeyArgs)24 OmKeyLocationInfo (org.apache.hadoop.ozone.om.helpers.OmKeyLocationInfo)22 OzoneOutputStream (org.apache.hadoop.ozone.client.io.OzoneOutputStream)21 IOException (java.io.IOException)19 OmKeyInfo (org.apache.hadoop.ozone.om.helpers.OmKeyInfo)18 StorageContainerManager (org.apache.hadoop.hdds.scm.server.StorageContainerManager)17 OzoneConfiguration (org.apache.hadoop.hdds.conf.OzoneConfiguration)16 ContainerManager (org.apache.hadoop.hdds.scm.container.ContainerManager)15 HddsProtos (org.apache.hadoop.hdds.protocol.proto.HddsProtos)12 KeyOutputStream (org.apache.hadoop.ozone.client.io.KeyOutputStream)12 ContainerNotFoundException (org.apache.hadoop.hdds.scm.container.ContainerNotFoundException)10