Examples with FieldLineageWriter io.cdap.cdap.data2.metadata.writer.FieldLineageWriter used on opensource projects

Example 1 with FieldLineageWriter

use of io.cdap.cdap.data2.metadata.writer.FieldLineageWriter in project cdap by caskdata.

the class MetadataSubscriberServiceTest method testSubscriber.

@Test
public void testSubscriber() throws InterruptedException, ExecutionException, TimeoutException {
    LineageStoreReader lineageReader = getInjector().getInstance(LineageStoreReader.class);
    ProgramRunId run1 = service1.run(RunIds.generate());
    // Try to read lineage, which should be empty since we haven't start the MetadataSubscriberService yet.
    Set<NamespacedEntityId> entities = lineageReader.getEntitiesForRun(run1);
    Assert.assertTrue(entities.isEmpty());
    // Write out some lineage information
    LineageWriter lineageWriter = getInjector().getInstance(MessagingLineageWriter.class);
    lineageWriter.addAccess(run1, dataset1, AccessType.READ);
    lineageWriter.addAccess(run1, dataset2, AccessType.WRITE);
    // Write the field level lineage
    FieldLineageWriter fieldLineageWriter = getInjector().getInstance(MessagingLineageWriter.class);
    ProgramRunId spark1Run1 = spark1.run(RunIds.generate(100));
    ReadOperation read = new ReadOperation("read", "some read", EndPoint.of("ns", "endpoint1"), "offset", "body");
    TransformOperation parse = new TransformOperation("parse", "parse body", Collections.singletonList(InputField.of("read", "body")), "name", "address");
    WriteOperation write = new WriteOperation("write", "write data", EndPoint.of("ns", "endpoint2"), Arrays.asList(InputField.of("read", "offset"), InputField.of("parse", "name"), InputField.of("parse", "address")));
    List<Operation> operations = new ArrayList<>();
    operations.add(read);
    operations.add(write);
    operations.add(parse);
    FieldLineageInfo info1 = new FieldLineageInfo(operations);
    fieldLineageWriter.write(spark1Run1, info1);
    ProgramRunId spark1Run2 = spark1.run(RunIds.generate(200));
    fieldLineageWriter.write(spark1Run2, info1);
    List<Operation> operations2 = new ArrayList<>();
    operations2.add(read);
    operations2.add(parse);
    TransformOperation normalize = new TransformOperation("normalize", "normalize address", Collections.singletonList(InputField.of("parse", "address")), "address");
    operations2.add(normalize);
    WriteOperation anotherWrite = new WriteOperation("anotherwrite", "write data", EndPoint.of("ns", "endpoint2"), Arrays.asList(InputField.of("read", "offset"), InputField.of("parse", "name"), InputField.of("normalize", "address")));
    operations2.add(anotherWrite);
    FieldLineageInfo info2 = new FieldLineageInfo(operations2);
    ProgramRunId spark1Run3 = spark1.run(RunIds.generate(300));
    fieldLineageWriter.write(spark1Run3, info2);
    // Emit some usages
    UsageWriter usageWriter = getInjector().getInstance(MessagingUsageWriter.class);
    usageWriter.register(spark1, dataset1);
    usageWriter.registerAll(Collections.singleton(spark1), dataset3);
    // Verifies lineage has been written
    Set<NamespacedEntityId> expectedLineage = new HashSet<>(Arrays.asList(run1.getParent(), dataset1, dataset2));
    Tasks.waitFor(true, () -> expectedLineage.equals(lineageReader.getEntitiesForRun(run1)), 10, TimeUnit.SECONDS, 100, TimeUnit.MILLISECONDS);
    // There shouldn't be any lineage for the "spark1" program, as only usage has been emitted.
    Assert.assertTrue(lineageReader.getRelations(spark1, 0L, Long.MAX_VALUE, x -> true).isEmpty());
    FieldLineageReader fieldLineageReader = getInjector().getInstance(FieldLineageReader.class);
    Set<Operation> expectedOperations = new HashSet<>();
    expectedOperations.add(read);
    expectedOperations.add(anotherWrite);
    List<ProgramRunOperations> expected = new ArrayList<>();
    // Descending order of program execution
    expected.add(new ProgramRunOperations(Collections.singleton(spark1Run3), expectedOperations));
    expectedOperations = new HashSet<>();
    expectedOperations.add(read);
    expectedOperations.add(write);
    expected.add(new ProgramRunOperations(new HashSet<>(Arrays.asList(spark1Run1, spark1Run2)), expectedOperations));
    EndPointField endPointField = new EndPointField(EndPoint.of("ns", "endpoint2"), "offset");
    Tasks.waitFor(expected, () -> fieldLineageReader.getIncomingOperations(endPointField, 1L, Long.MAX_VALUE - 1), 10, TimeUnit.SECONDS, 100, TimeUnit.MILLISECONDS);
    // Verifies usage has been written
    Set<EntityId> expectedUsage = new HashSet<>(Arrays.asList(dataset1, dataset3));
    UsageRegistry usageRegistry = getInjector().getInstance(UsageRegistry.class);
    Tasks.waitFor(true, () -> expectedUsage.equals(usageRegistry.getDatasets(spark1)), 10, TimeUnit.SECONDS, 100, TimeUnit.MILLISECONDS);
}

Example 2 with FieldLineageWriter

use of io.cdap.cdap.data2.metadata.writer.FieldLineageWriter in project cdap by caskdata.

the class MapReduceTaskContextProvider method createCacheLoader.

/**
 * Creates a {@link CacheLoader} for the task context cache.
 */
private CacheLoader<ContextCacheKey, BasicMapReduceTaskContext> createCacheLoader(final Injector injector) {
    DiscoveryServiceClient discoveryServiceClient = injector.getInstance(DiscoveryServiceClient.class);
    DatasetFramework datasetFramework = injector.getInstance(DatasetFramework.class);
    SecureStore secureStore = injector.getInstance(SecureStore.class);
    SecureStoreManager secureStoreManager = injector.getInstance(SecureStoreManager.class);
    MessagingService messagingService = injector.getInstance(MessagingService.class);
    // Multiple instances of BasicMapReduceTaskContext can share the same program.
    AtomicReference<Program> programRef = new AtomicReference<>();
    MetadataReader metadataReader = injector.getInstance(MetadataReader.class);
    MetadataPublisher metadataPublisher = injector.getInstance(MetadataPublisher.class);
    FieldLineageWriter fieldLineageWriter = injector.getInstance(FieldLineageWriter.class);
    RemoteClientFactory remoteClientFactory = injector.getInstance(RemoteClientFactory.class);
    return new CacheLoader<ContextCacheKey, BasicMapReduceTaskContext>() {

        @Override
        public BasicMapReduceTaskContext load(ContextCacheKey key) throws Exception {
            TaskAttemptID taskAttemptId = key.getTaskAttemptID();
            // taskAttemptId could be null if used from a org.apache.hadoop.mapreduce.Partitioner or
            // from a org.apache.hadoop.io.RawComparator, in which case we can get the JobId from the conf. Note that the
            // JobId isn't in the conf for the OutputCommitter#setupJob method, in which case we use the taskAttemptId
            Path txFile = MainOutputCommitter.getTxFile(key.getConfiguration(), taskAttemptId != null ? taskAttemptId.getJobID() : null);
            FileSystem fs = txFile.getFileSystem(key.getConfiguration());
            Transaction transaction = null;
            if (fs.exists(txFile)) {
                try (FSDataInputStream txFileInputStream = fs.open(txFile)) {
                    transaction = new TransactionCodec().decode(ByteStreams.toByteArray(txFileInputStream));
                }
            }
            MapReduceContextConfig contextConfig = new MapReduceContextConfig(key.getConfiguration());
            MapReduceClassLoader classLoader = MapReduceClassLoader.getFromConfiguration(key.getConfiguration());
            Program program = programRef.get();
            if (program == null) {
                // Creation of program is relatively cheap, so just create and do compare and set.
                programRef.compareAndSet(null, createProgram(contextConfig, classLoader.getProgramClassLoader()));
                program = programRef.get();
            }
            WorkflowProgramInfo workflowInfo = contextConfig.getWorkflowProgramInfo();
            DatasetFramework programDatasetFramework = workflowInfo == null ? datasetFramework : NameMappedDatasetFramework.createFromWorkflowProgramInfo(datasetFramework, workflowInfo, program.getApplicationSpecification());
            // Setup dataset framework context, if required
            if (programDatasetFramework instanceof ProgramContextAware) {
                ProgramRunId programRunId = program.getId().run(ProgramRunners.getRunId(contextConfig.getProgramOptions()));
                ((ProgramContextAware) programDatasetFramework).setContext(new BasicProgramContext(programRunId));
            }
            MapReduceSpecification spec = program.getApplicationSpecification().getMapReduce().get(program.getName());
            MetricsCollectionService metricsCollectionService = null;
            MapReduceMetrics.TaskType taskType = null;
            String taskId = null;
            ProgramOptions options = contextConfig.getProgramOptions();
            // from a org.apache.hadoop.io.RawComparator
            if (taskAttemptId != null) {
                taskId = taskAttemptId.getTaskID().toString();
                if (MapReduceMetrics.TaskType.hasType(taskAttemptId.getTaskType())) {
                    taskType = MapReduceMetrics.TaskType.from(taskAttemptId.getTaskType());
                    // if this is not for a mapper or a reducer, we don't need the metrics collection service
                    metricsCollectionService = injector.getInstance(MetricsCollectionService.class);
                    options = new SimpleProgramOptions(options.getProgramId(), options.getArguments(), new BasicArguments(RuntimeArguments.extractScope("task", taskType.toString().toLowerCase(), contextConfig.getProgramOptions().getUserArguments().asMap())), options.isDebug());
                }
            }
            CConfiguration cConf = injector.getInstance(CConfiguration.class);
            TransactionSystemClient txClient = injector.getInstance(TransactionSystemClient.class);
            NamespaceQueryAdmin namespaceQueryAdmin = injector.getInstance(NamespaceQueryAdmin.class);
            return new BasicMapReduceTaskContext(program, options, cConf, taskType, taskId, spec, workflowInfo, discoveryServiceClient, metricsCollectionService, txClient, transaction, programDatasetFramework, classLoader.getPluginInstantiator(), contextConfig.getLocalizedResources(), secureStore, secureStoreManager, accessEnforcer, authenticationContext, messagingService, mapReduceClassLoader, metadataReader, metadataPublisher, namespaceQueryAdmin, fieldLineageWriter, remoteClientFactory);
        }
    };
}

Example 3 with FieldLineageWriter

use of io.cdap.cdap.data2.metadata.writer.FieldLineageWriter in project cdap by caskdata.

the class SparkProgramRunner method run.

@Override
public ProgramController run(Program program, ProgramOptions options) {
    LOG.trace("Starting Spark program {} with SparkProgramRunner of ClassLoader {}", program.getId(), getClass().getClassLoader());
    // Get the RunId first. It is used for the creation of the ClassLoader closing thread.
    Arguments arguments = options.getArguments();
    RunId runId = ProgramRunners.getRunId(options);
    Deque<Closeable> closeables = new LinkedList<>();
    try {
        // Extract and verify parameters
        ApplicationSpecification appSpec = program.getApplicationSpecification();
        Preconditions.checkNotNull(appSpec, "Missing application specification.");
        ProgramType processorType = program.getType();
        Preconditions.checkNotNull(processorType, "Missing processor type.");
        Preconditions.checkArgument(processorType == ProgramType.SPARK, "Only Spark process type is supported.");
        SparkSpecification spec = appSpec.getSpark().get(program.getName());
        Preconditions.checkNotNull(spec, "Missing SparkSpecification for %s", program.getName());
        String host = options.getArguments().getOption(ProgramOptionConstants.HOST);
        Preconditions.checkArgument(host != null, "No hostname is provided");
        // Get the WorkflowProgramInfo if it is started by Workflow
        WorkflowProgramInfo workflowInfo = WorkflowProgramInfo.create(arguments);
        DatasetFramework programDatasetFramework = workflowInfo == null ? datasetFramework : NameMappedDatasetFramework.createFromWorkflowProgramInfo(datasetFramework, workflowInfo, appSpec);
        // Setup dataset framework context, if required
        if (programDatasetFramework instanceof ProgramContextAware) {
            ProgramId programId = program.getId();
            ((ProgramContextAware) programDatasetFramework).setContext(new BasicProgramContext(programId.run(runId)));
        }
        PluginInstantiator pluginInstantiator = createPluginInstantiator(options, program.getClassLoader());
        if (pluginInstantiator != null) {
            closeables.addFirst(pluginInstantiator);
        }
        SparkRuntimeContext runtimeContext = new SparkRuntimeContext(new Configuration(hConf), program, options, cConf, host, txClient, programDatasetFramework, metricsCollectionService, workflowInfo, pluginInstantiator, secureStore, secureStoreManager, accessEnforcer, authenticationContext, messagingService, serviceAnnouncer, pluginFinder, locationFactory, metadataReader, metadataPublisher, namespaceQueryAdmin, fieldLineageWriter, remoteClientFactory, () -> {
        });
        closeables.addFirst(runtimeContext);
        Spark spark;
        try {
            spark = new InstantiatorFactory(false).get(TypeToken.of(program.<Spark>getMainClass())).create();
        } catch (Exception e) {
            LOG.error("Failed to instantiate Spark class for {}", spec.getClassName(), e);
            throw Throwables.propagate(e);
        }
        boolean isLocal = SparkRuntimeContextConfig.isLocal(options);
        SparkSubmitter submitter;
        // If MasterEnvironment is not available, use non-master env spark submitters
        MasterEnvironment masterEnv = MasterEnvironments.getMasterEnvironment();
        if (masterEnv != null && cConf.getBoolean(Constants.Environment.PROGRAM_SUBMISSION_MASTER_ENV_ENABLED, true)) {
            submitter = new MasterEnvironmentSparkSubmitter(cConf, locationFactory, host, runtimeContext, masterEnv);
        } else {
            submitter = isLocal ? new LocalSparkSubmitter() : new DistributedSparkSubmitter(hConf, locationFactory, host, runtimeContext, options.getArguments().getOption(Constants.AppFabric.APP_SCHEDULER_QUEUE));
        }
        Service sparkRuntimeService = new SparkRuntimeService(cConf, spark, getPluginArchive(options), runtimeContext, submitter, locationFactory, isLocal, fieldLineageWriter, masterEnv);
        sparkRuntimeService.addListener(createRuntimeServiceListener(closeables), Threads.SAME_THREAD_EXECUTOR);
        ProgramController controller = new SparkProgramController(sparkRuntimeService, runtimeContext);
        LOG.debug("Starting Spark Job. Context: {}", runtimeContext);
        if (isLocal || UserGroupInformation.isSecurityEnabled()) {
            sparkRuntimeService.start();
        } else {
            ProgramRunners.startAsUser(cConf.get(Constants.CFG_HDFS_USER), sparkRuntimeService);
        }
        return controller;
    } catch (Throwable t) {
        closeAllQuietly(closeables);
        throw Throwables.propagate(t);
    }
}

Example 4 with FieldLineageWriter

use of io.cdap.cdap.data2.metadata.writer.FieldLineageWriter in project cdap by caskdata.

the class MapReduceProgramRunner method run.

@Override
public ProgramController run(final Program program, ProgramOptions options) {
    // Extract and verify parameters
    ApplicationSpecification appSpec = program.getApplicationSpecification();
    Preconditions.checkNotNull(appSpec, "Missing application specification.");
    ProgramType processorType = program.getType();
    Preconditions.checkNotNull(processorType, "Missing processor type.");
    Preconditions.checkArgument(processorType == ProgramType.MAPREDUCE, "Only MAPREDUCE process type is supported.");
    MapReduceSpecification spec = appSpec.getMapReduce().get(program.getName());
    Preconditions.checkNotNull(spec, "Missing MapReduceSpecification for %s", program.getName());
    Arguments arguments = options.getArguments();
    RunId runId = ProgramRunners.getRunId(options);
    WorkflowProgramInfo workflowInfo = WorkflowProgramInfo.create(arguments);
    DatasetFramework programDatasetFramework = workflowInfo == null ? datasetFramework : NameMappedDatasetFramework.createFromWorkflowProgramInfo(datasetFramework, workflowInfo, appSpec);
    // Setup dataset framework context, if required
    if (programDatasetFramework instanceof ProgramContextAware) {
        ProgramId programId = program.getId();
        ((ProgramContextAware) programDatasetFramework).setContext(new BasicProgramContext(programId.run(runId)));
    }
    MapReduce mapReduce;
    try {
        mapReduce = new InstantiatorFactory(false).get(TypeToken.of(program.<MapReduce>getMainClass())).create();
    } catch (Exception e) {
        LOG.error("Failed to instantiate MapReduce class for {}", spec.getClassName(), e);
        throw Throwables.propagate(e);
    }
    // List of all Closeable resources that needs to be cleanup
    List<Closeable> closeables = new ArrayList<>();
    try {
        PluginInstantiator pluginInstantiator = createPluginInstantiator(options, program.getClassLoader());
        if (pluginInstantiator != null) {
            closeables.add(pluginInstantiator);
        }
        final BasicMapReduceContext context = new BasicMapReduceContext(program, options, cConf, spec, workflowInfo, discoveryServiceClient, metricsCollectionService, txSystemClient, programDatasetFramework, getPluginArchive(options), pluginInstantiator, secureStore, secureStoreManager, messagingService, metadataReader, metadataPublisher, namespaceQueryAdmin, fieldLineageWriter, remoteClientFactory);
        closeables.add(context);
        Reflections.visit(mapReduce, mapReduce.getClass(), new PropertyFieldSetter(context.getSpecification().getProperties()), new MetricsFieldSetter(context.getMetrics()), new DataSetFieldSetter(context));
        // note: this sets logging context on the thread level
        LoggingContextAccessor.setLoggingContext(context.getLoggingContext());
        // Set the job queue to hConf if it is provided
        Configuration hConf = new Configuration(this.hConf);
        String schedulerQueue = options.getArguments().getOption(Constants.AppFabric.APP_SCHEDULER_QUEUE);
        if (schedulerQueue != null && !schedulerQueue.isEmpty()) {
            hConf.set(JobContext.QUEUE_NAME, schedulerQueue);
        }
        ClusterMode clusterMode = ProgramRunners.getClusterMode(options);
        Service mapReduceRuntimeService = new MapReduceRuntimeService(injector, cConf, hConf, mapReduce, spec, context, program.getJarLocation(), locationFactory, clusterMode, fieldLineageWriter);
        mapReduceRuntimeService.addListener(createRuntimeServiceListener(closeables), Threads.SAME_THREAD_EXECUTOR);
        ProgramController controller = new MapReduceProgramController(mapReduceRuntimeService, context);
        LOG.debug("Starting MapReduce Job: {}", context);
        // be running the job, but the data directory will be owned by cdap.
        if (MapReduceTaskContextProvider.isLocal(hConf) || UserGroupInformation.isSecurityEnabled()) {
            mapReduceRuntimeService.start();
        } else {
            ProgramRunners.startAsUser(cConf.get(Constants.CFG_HDFS_USER), mapReduceRuntimeService);
        }
        return controller;
    } catch (Exception e) {
        closeAllQuietly(closeables);
        throw Throwables.propagate(e);
    }
}

Example 5 with FieldLineageWriter

use of io.cdap.cdap.data2.metadata.writer.FieldLineageWriter in project cdap by caskdata.

the class LineageLimitingTest method testLineageLimiting.

@Test
public void testLineageLimiting() throws InterruptedException, ExecutionException, TimeoutException {
    LineageStoreReader lineageReader = getInjector().getInstance(LineageStoreReader.class);
    ProgramRunId run1 = service1.run(RunIds.generate());
    // Write out some lineage information
    LineageWriter lineageWriter = getInjector().getInstance(MessagingLineageWriter.class);
    lineageWriter.addAccess(run1, dataset1, AccessType.READ);
    lineageWriter.addAccess(run1, dataset2, AccessType.WRITE);
    // Write the field level lineage
    FieldLineageWriter fieldLineageWriter = getInjector().getInstance(MessagingLineageWriter.class);
    ProgramRunId spark1Run1 = spark1.run(RunIds.generate(100));
    ReadOperation read = new ReadOperation("read", "some read", EndPoint.of("ns", "endpoint1"), "offset", "body");
    TransformOperation parse = new TransformOperation("parse", "parse body", Collections.singletonList(InputField.of("read", "body")), "name", "address");
    WriteOperation write = new WriteOperation("write", "write data", EndPoint.of("ns", "endpoint2"), Arrays.asList(InputField.of("read", "offset"), InputField.of("parse", "name"), InputField.of("parse", "address")));
    List<Operation> operations = new ArrayList<>();
    operations.add(read);
    operations.add(write);
    operations.add(parse);
    FieldLineageInfo info1 = new FieldLineageInfo(operations);
    fieldLineageWriter.write(spark1Run1, info1);
    ProgramRunId spark1Run2 = spark1.run(RunIds.generate(200));
    fieldLineageWriter.write(spark1Run2, info1);
    // Verifies lineage has been written as it is smaller than maximum specified size
    Set<NamespacedEntityId> expectedLineage = new HashSet<>(Arrays.asList(run1.getParent(), dataset1, dataset2));
    Tasks.waitFor(true, () -> expectedLineage.equals(lineageReader.getEntitiesForRun(run1)), 10, TimeUnit.SECONDS, 100, TimeUnit.MILLISECONDS);
    FieldLineageReader fieldLineageReader = getInjector().getInstance(FieldLineageReader.class);
    // Verifies that empty lineage has been written
    EndPointField endPointField = new EndPointField(EndPoint.of("ns", "endpoint2"), "offset");
    List<ProgramRunOperations> incomingOperations = fieldLineageReader.getIncomingOperations(endPointField, 1L, Long.MAX_VALUE - 1);
    Assert.assertTrue(incomingOperations.isEmpty());
}