Examples with SCMContext org.apache.hadoop.hdds.scm.ha.SCMContext used on opensource projects

Example 1 with SCMContext

use of org.apache.hadoop.hdds.scm.ha.SCMContext in project ozone by apache.

the class TestHealthyPipelineSafeModeRule method testHealthyPipelineSafeModeRuleWithMixedPipelines.

@Test
public void testHealthyPipelineSafeModeRuleWithMixedPipelines() throws Exception {
    String storageDir = GenericTestUtils.getTempPath(TestHealthyPipelineSafeModeRule.class.getName() + UUID.randomUUID());
    EventQueue eventQueue = new EventQueue();
    SCMServiceManager serviceManager = new SCMServiceManager();
    SCMContext scmContext = SCMContext.emptyContext();
    List<ContainerInfo> containers = new ArrayList<>(HddsTestUtils.getContainerInfo(1));
    OzoneConfiguration config = new OzoneConfiguration();
    // In Mock Node Manager, first 8 nodes are healthy, next 2 nodes are
    // stale and last one is dead, and this repeats. So for a 12 node, 9
    // healthy, 2 stale and one dead.
    MockNodeManager nodeManager = new MockNodeManager(true, 12);
    config.set(HddsConfigKeys.OZONE_METADATA_DIRS, storageDir);
    // enable pipeline check
    config.setBoolean(HddsConfigKeys.HDDS_SCM_SAFEMODE_PIPELINE_AVAILABILITY_CHECK, true);
    config.setBoolean(HddsConfigKeys.HDDS_SCM_SAFEMODE_PIPELINE_CREATION, false);
    SCMMetadataStore scmMetadataStore = new SCMMetadataStoreImpl(config);
    try {
        PipelineManagerImpl pipelineManager = PipelineManagerImpl.newPipelineManager(config, MockSCMHAManager.getInstance(true), nodeManager, scmMetadataStore.getPipelineTable(), eventQueue, scmContext, serviceManager);
        PipelineProvider mockRatisProvider = new MockRatisPipelineProvider(nodeManager, pipelineManager.getStateManager(), config);
        pipelineManager.setPipelineProvider(HddsProtos.ReplicationType.RATIS, mockRatisProvider);
        // Create 3 pipelines
        Pipeline pipeline1 = pipelineManager.createPipeline(RatisReplicationConfig.getInstance(ReplicationFactor.ONE));
        pipelineManager.openPipeline(pipeline1.getId());
        Pipeline pipeline2 = pipelineManager.createPipeline(RatisReplicationConfig.getInstance(ReplicationFactor.THREE));
        pipelineManager.openPipeline(pipeline2.getId());
        Pipeline pipeline3 = pipelineManager.createPipeline(RatisReplicationConfig.getInstance(ReplicationFactor.THREE));
        pipelineManager.openPipeline(pipeline3.getId());
        // Mark pipeline healthy
        pipeline1 = pipelineManager.getPipeline(pipeline1.getId());
        MockRatisPipelineProvider.markPipelineHealthy(pipeline1);
        pipeline2 = pipelineManager.getPipeline(pipeline2.getId());
        MockRatisPipelineProvider.markPipelineHealthy(pipeline2);
        pipeline3 = pipelineManager.getPipeline(pipeline3.getId());
        MockRatisPipelineProvider.markPipelineHealthy(pipeline3);
        SCMSafeModeManager scmSafeModeManager = new SCMSafeModeManager(config, containers, null, pipelineManager, eventQueue, serviceManager, scmContext);
        HealthyPipelineSafeModeRule healthyPipelineSafeModeRule = scmSafeModeManager.getHealthyPipelineSafeModeRule();
        // No pipeline event have sent to SCMSafemodeManager
        Assert.assertFalse(healthyPipelineSafeModeRule.validate());
        GenericTestUtils.LogCapturer logCapturer = GenericTestUtils.LogCapturer.captureLogs(LoggerFactory.getLogger(SCMSafeModeManager.class));
        // fire event with pipeline create status with ratis type and factor 1
        // pipeline, validate() should return false
        firePipelineEvent(pipeline1, eventQueue);
        GenericTestUtils.waitFor(() -> logCapturer.getOutput().contains("reported count is 1"), 1000, 5000);
        Assert.assertFalse(healthyPipelineSafeModeRule.validate());
        firePipelineEvent(pipeline2, eventQueue);
        firePipelineEvent(pipeline3, eventQueue);
        GenericTestUtils.waitFor(() -> healthyPipelineSafeModeRule.validate(), 1000, 5000);
    } finally {
        scmMetadataStore.getStore().close();
        FileUtil.fullyDelete(new File(storageDir));
    }
}

Example 2 with SCMContext

use of org.apache.hadoop.hdds.scm.ha.SCMContext in project ozone by apache.

the class StorageContainerManager method initializeSystemManagers.

/**
 * This function initializes the following managers. If the configurator
 * specifies a value, we will use it, else we will use the default value.
 *
 *  Node Manager
 *  Pipeline Manager
 *  Container Manager
 *  Block Manager
 *  Replication Manager
 *  Safe Mode Manager
 *
 * @param conf - Ozone Configuration.
 * @param configurator - A customizer which allows different managers to be
 *                    used if needed.
 * @throws IOException - on Failure.
 */
private void initializeSystemManagers(OzoneConfiguration conf, SCMConfigurator configurator) throws IOException {
    if (configurator.getNetworkTopology() != null) {
        clusterMap = configurator.getNetworkTopology();
    } else {
        clusterMap = new NetworkTopologyImpl(conf);
    }
    // This needs to be done before initializing Ratis.
    RatisDropwizardExports.registerRatisMetricReporters(ratisMetricsMap);
    if (configurator.getSCMHAManager() != null) {
        scmHAManager = configurator.getSCMHAManager();
    } else {
        scmHAManager = new SCMHAManagerImpl(conf, this);
    }
    // inline upgrade for SequenceIdGenerator
    SequenceIdGenerator.upgradeToSequenceId(scmMetadataStore);
    // Distributed sequence id generator
    sequenceIdGen = new SequenceIdGenerator(conf, scmHAManager, scmMetadataStore.getSequenceIdTable());
    if (configurator.getScmContext() != null) {
        scmContext = configurator.getScmContext();
    } else {
        // When term equals SCMContext.INVALID_TERM, the isLeader() check
        // and getTermOfLeader() will always pass.
        long term = SCMHAUtils.isSCMHAEnabled(conf) ? 0 : SCMContext.INVALID_TERM;
        // non-leader of term 0, in safe mode, preCheck not completed.
        scmContext = new SCMContext.Builder().setLeader(false).setTerm(term).setIsInSafeMode(true).setIsPreCheckComplete(false).setSCM(this).build();
    }
    if (configurator.getScmNodeManager() != null) {
        scmNodeManager = configurator.getScmNodeManager();
    } else {
        scmNodeManager = new SCMNodeManager(conf, scmStorageConfig, eventQueue, clusterMap, scmContext, scmLayoutVersionManager);
    }
    placementMetrics = SCMContainerPlacementMetrics.create();
    containerPlacementPolicy = ContainerPlacementPolicyFactory.getPolicy(conf, scmNodeManager, clusterMap, true, placementMetrics);
    if (configurator.getPipelineManager() != null) {
        pipelineManager = configurator.getPipelineManager();
    } else {
        pipelineManager = PipelineManagerImpl.newPipelineManager(conf, scmHAManager, scmNodeManager, scmMetadataStore.getPipelineTable(), eventQueue, scmContext, serviceManager);
    }
    if (configurator.getContainerManager() != null) {
        containerManager = configurator.getContainerManager();
    } else {
        containerManager = new ContainerManagerImpl(conf, scmHAManager, sequenceIdGen, pipelineManager, scmMetadataStore.getContainerTable());
    }
    pipelineChoosePolicy = PipelineChoosePolicyFactory.getPolicy(conf);
    if (configurator.getWritableContainerFactory() != null) {
        writableContainerFactory = configurator.getWritableContainerFactory();
    } else {
        writableContainerFactory = new WritableContainerFactory(this);
    }
    if (configurator.getScmBlockManager() != null) {
        scmBlockManager = configurator.getScmBlockManager();
    } else {
        scmBlockManager = new BlockManagerImpl(conf, this);
    }
    if (configurator.getReplicationManager() != null) {
        replicationManager = configurator.getReplicationManager();
    } else {
        replicationManager = new ReplicationManager(conf, containerManager, containerPlacementPolicy, eventQueue, scmContext, serviceManager, scmNodeManager, new MonotonicClock(ZoneOffset.UTC), scmHAManager, getScmMetadataStore().getMoveTable());
    }
    if (configurator.getScmSafeModeManager() != null) {
        scmSafeModeManager = configurator.getScmSafeModeManager();
    } else {
        scmSafeModeManager = new SCMSafeModeManager(conf, containerManager.getContainers(), containerManager, pipelineManager, eventQueue, serviceManager, scmContext);
    }
    scmDecommissionManager = new NodeDecommissionManager(conf, scmNodeManager, containerManager, scmContext, eventQueue, replicationManager);
}

Example 3 with SCMContext

use of org.apache.hadoop.hdds.scm.ha.SCMContext in project ozone by apache.

the class TestHealthyPipelineSafeModeRule method testHealthyPipelineSafeModeRuleWithNoPipelines.

@Test
public void testHealthyPipelineSafeModeRuleWithNoPipelines() throws Exception {
    EventQueue eventQueue = new EventQueue();
    SCMServiceManager serviceManager = new SCMServiceManager();
    SCMContext scmContext = SCMContext.emptyContext();
    List<ContainerInfo> containers = new ArrayList<>(HddsTestUtils.getContainerInfo(1));
    String storageDir = GenericTestUtils.getTempPath(TestHealthyPipelineSafeModeRule.class.getName() + UUID.randomUUID());
    OzoneConfiguration config = new OzoneConfiguration();
    MockNodeManager nodeManager = new MockNodeManager(true, 0);
    config.set(HddsConfigKeys.OZONE_METADATA_DIRS, storageDir);
    // enable pipeline check
    config.setBoolean(HddsConfigKeys.HDDS_SCM_SAFEMODE_PIPELINE_AVAILABILITY_CHECK, true);
    config.setBoolean(HddsConfigKeys.HDDS_SCM_SAFEMODE_PIPELINE_CREATION, false);
    SCMMetadataStore scmMetadataStore = new SCMMetadataStoreImpl(config);
    try {
        PipelineManagerImpl pipelineManager = PipelineManagerImpl.newPipelineManager(config, MockSCMHAManager.getInstance(true), nodeManager, scmMetadataStore.getPipelineTable(), eventQueue, scmContext, serviceManager);
        PipelineProvider mockRatisProvider = new MockRatisPipelineProvider(nodeManager, pipelineManager.getStateManager(), config);
        pipelineManager.setPipelineProvider(HddsProtos.ReplicationType.RATIS, mockRatisProvider);
        SCMSafeModeManager scmSafeModeManager = new SCMSafeModeManager(config, containers, null, pipelineManager, eventQueue, serviceManager, scmContext);
        HealthyPipelineSafeModeRule healthyPipelineSafeModeRule = scmSafeModeManager.getHealthyPipelineSafeModeRule();
        // This should be immediately satisfied, as no pipelines are there yet.
        Assert.assertTrue(healthyPipelineSafeModeRule.validate());
    } finally {
        scmMetadataStore.getStore().close();
        FileUtil.fullyDelete(new File(storageDir));
    }
}

Example 4 with SCMContext

use of org.apache.hadoop.hdds.scm.ha.SCMContext in project ozone by apache.

the class TestHealthyPipelineSafeModeRule method testHealthyPipelineSafeModeRuleWithPipelines.

@Test
public void testHealthyPipelineSafeModeRuleWithPipelines() throws Exception {
    String storageDir = GenericTestUtils.getTempPath(TestHealthyPipelineSafeModeRule.class.getName() + UUID.randomUUID());
    EventQueue eventQueue = new EventQueue();
    SCMServiceManager serviceManager = new SCMServiceManager();
    SCMContext scmContext = SCMContext.emptyContext();
    List<ContainerInfo> containers = new ArrayList<>(HddsTestUtils.getContainerInfo(1));
    OzoneConfiguration config = new OzoneConfiguration();
    // In Mock Node Manager, first 8 nodes are healthy, next 2 nodes are
    // stale and last one is dead, and this repeats. So for a 12 node, 9
    // healthy, 2 stale and one dead.
    MockNodeManager nodeManager = new MockNodeManager(true, 12);
    config.set(HddsConfigKeys.OZONE_METADATA_DIRS, storageDir);
    // enable pipeline check
    config.setBoolean(HddsConfigKeys.HDDS_SCM_SAFEMODE_PIPELINE_AVAILABILITY_CHECK, true);
    config.setBoolean(HddsConfigKeys.HDDS_SCM_SAFEMODE_PIPELINE_CREATION, false);
    SCMMetadataStore scmMetadataStore = new SCMMetadataStoreImpl(config);
    try {
        PipelineManagerImpl pipelineManager = PipelineManagerImpl.newPipelineManager(config, MockSCMHAManager.getInstance(true), nodeManager, scmMetadataStore.getPipelineTable(), eventQueue, scmContext, serviceManager);
        PipelineProvider mockRatisProvider = new MockRatisPipelineProvider(nodeManager, pipelineManager.getStateManager(), config);
        pipelineManager.setPipelineProvider(HddsProtos.ReplicationType.RATIS, mockRatisProvider);
        // Create 3 pipelines
        Pipeline pipeline1 = pipelineManager.createPipeline(RatisReplicationConfig.getInstance(ReplicationFactor.THREE));
        pipelineManager.openPipeline(pipeline1.getId());
        Pipeline pipeline2 = pipelineManager.createPipeline(RatisReplicationConfig.getInstance(ReplicationFactor.THREE));
        pipelineManager.openPipeline(pipeline2.getId());
        Pipeline pipeline3 = pipelineManager.createPipeline(RatisReplicationConfig.getInstance(ReplicationFactor.THREE));
        pipelineManager.openPipeline(pipeline3.getId());
        // Mark pipeline healthy
        pipeline1 = pipelineManager.getPipeline(pipeline1.getId());
        MockRatisPipelineProvider.markPipelineHealthy(pipeline1);
        pipeline2 = pipelineManager.getPipeline(pipeline2.getId());
        MockRatisPipelineProvider.markPipelineHealthy(pipeline2);
        pipeline3 = pipelineManager.getPipeline(pipeline3.getId());
        MockRatisPipelineProvider.markPipelineHealthy(pipeline3);
        SCMSafeModeManager scmSafeModeManager = new SCMSafeModeManager(config, containers, null, pipelineManager, eventQueue, serviceManager, scmContext);
        HealthyPipelineSafeModeRule healthyPipelineSafeModeRule = scmSafeModeManager.getHealthyPipelineSafeModeRule();
        // No datanodes have sent pipelinereport from datanode
        Assert.assertFalse(healthyPipelineSafeModeRule.validate());
        // Fire pipeline report from all datanodes in first pipeline, as here we
        // have 3 pipelines, 10% is 0.3, when doing ceil it is 1. So, we should
        // validate should return true after fire pipeline event
        // Here testing with out pipelinereport handler, so not moving created
        // pipelines to allocated state, as pipelines changing to healthy is
        // handled by pipeline report handler. So, leaving pipeline's in pipeline
        // manager in open state for test case simplicity.
        firePipelineEvent(pipeline1, eventQueue);
        GenericTestUtils.waitFor(() -> healthyPipelineSafeModeRule.validate(), 1000, 5000);
    } finally {
        scmMetadataStore.getStore().close();
        FileUtil.fullyDelete(new File(storageDir));
    }
}

Example 5 with SCMContext

use of org.apache.hadoop.hdds.scm.ha.SCMContext in project ozone by apache.

the class TestStorageContainerManager method testCloseContainerCommandOnRestart.

@Test
@SuppressWarnings("unchecked")
public void testCloseContainerCommandOnRestart() throws Exception {
    int numKeys = 15;
    OzoneConfiguration conf = new OzoneConfiguration();
    conf.setTimeDuration(HDDS_CONTAINER_REPORT_INTERVAL, 1, TimeUnit.SECONDS);
    conf.setInt(ScmConfigKeys.OZONE_SCM_BLOCK_DELETION_MAX_RETRY, 5);
    conf.setTimeDuration(OzoneConfigKeys.OZONE_BLOCK_DELETING_SERVICE_INTERVAL, 100, TimeUnit.MILLISECONDS);
    conf.setInt(ScmConfigKeys.OZONE_SCM_PIPELINE_OWNER_CONTAINER_COUNT, numKeys);
    conf.setBoolean(HDDS_SCM_SAFEMODE_PIPELINE_CREATION, false);
    MiniOzoneCluster cluster = MiniOzoneCluster.newBuilder(conf).setHbInterval(1000).setHbProcessorInterval(3000).setTrace(false).setNumDatanodes(1).build();
    cluster.waitForClusterToBeReady();
    cluster.waitForPipelineTobeReady(HddsProtos.ReplicationFactor.ONE, 30000);
    try {
        TestStorageContainerManagerHelper helper = new TestStorageContainerManagerHelper(cluster, conf);
        helper.createKeys(10, 4096);
        GenericTestUtils.waitFor(() -> cluster.getStorageContainerManager().getContainerManager().getContainers() != null, 1000, 10000);
        StorageContainerManager scm = cluster.getStorageContainerManager();
        List<ContainerInfo> containers = cluster.getStorageContainerManager().getContainerManager().getContainers();
        Assert.assertNotNull(containers);
        ContainerInfo selectedContainer = containers.iterator().next();
        // Stop processing HB
        scm.getDatanodeProtocolServer().stop();
        LOG.info("Current Container State is {}", selectedContainer.getState());
        try {
            scm.getContainerManager().updateContainerState(selectedContainer.containerID(), HddsProtos.LifeCycleEvent.FINALIZE);
        } catch (SCMException ex) {
            if (selectedContainer.getState() != HddsProtos.LifeCycleState.CLOSING) {
                ex.printStackTrace();
                throw (ex);
            }
        }
        cluster.restartStorageContainerManager(false);
        scm = cluster.getStorageContainerManager();
        EventPublisher publisher = mock(EventPublisher.class);
        ReplicationManager replicationManager = scm.getReplicationManager();
        Field f = ReplicationManager.class.getDeclaredField("eventPublisher");
        f.setAccessible(true);
        Field modifiersField = Field.class.getDeclaredField("modifiers");
        modifiersField.setAccessible(true);
        modifiersField.setInt(f, f.getModifiers() & ~Modifier.FINAL);
        f.set(replicationManager, publisher);
        UUID dnUuid = cluster.getHddsDatanodes().iterator().next().getDatanodeDetails().getUuid();
        CloseContainerCommand closeContainerCommand = new CloseContainerCommand(selectedContainer.getContainerID(), selectedContainer.getPipelineID(), false);
        CommandForDatanode commandForDatanode = new CommandForDatanode(dnUuid, closeContainerCommand);
        GenericTestUtils.waitFor(() -> {
            SCMContext scmContext = cluster.getStorageContainerManager().getScmContext();
            return !scmContext.isInSafeMode() && scmContext.isLeader();
        }, 1000, 25000);
        // After safe mode is off, ReplicationManager starts to run with a delay.
        Thread.sleep(5000);
        // Give ReplicationManager some time to process the containers.
        cluster.getStorageContainerManager().getReplicationManager().processAll();
        Thread.sleep(5000);
        verify(publisher).fireEvent(eq(SCMEvents.DATANODE_COMMAND), argThat(new CloseContainerCommandMatcher(dnUuid, commandForDatanode)));
    } finally {
        cluster.shutdown();
    }
}