Examples with ContainerContext org.apache.samza.context.ContainerContext used on opensource projects

Example 11 with ContainerContext

use of org.apache.samza.context.ContainerContext in project samza by apache.

the class TestProjectTranslator method testTranslate.

@Test
public void testTranslate() throws IOException, ClassNotFoundException {
    // setup mock values to the constructor of FilterTranslator
    LogicalProject mockProject = PowerMockito.mock(LogicalProject.class);
    Context mockContext = mock(Context.class);
    ContainerContext mockContainerContext = mock(ContainerContext.class);
    TranslatorContext mockTranslatorContext = mock(TranslatorContext.class);
    TestMetricsRegistryImpl testMetricsRegistryImpl = new TestMetricsRegistryImpl();
    RelNode mockInput = mock(RelNode.class);
    List<RelNode> inputs = new ArrayList<>();
    inputs.add(mockInput);
    when(mockInput.getId()).thenReturn(1);
    when(mockProject.getId()).thenReturn(2);
    when(mockProject.getInputs()).thenReturn(inputs);
    when(mockProject.getInput()).thenReturn(mockInput);
    RelDataType mockRowType = mock(RelDataType.class);
    when(mockRowType.getFieldCount()).thenReturn(1);
    when(mockProject.getRowType()).thenReturn(mockRowType);
    RexNode mockRexField = mock(RexNode.class);
    List<Pair<RexNode, String>> namedProjects = new ArrayList<>();
    namedProjects.add(Pair.of(mockRexField, "test_field"));
    when(mockProject.getNamedProjects()).thenReturn(namedProjects);
    StreamApplicationDescriptorImpl mockAppDesc = mock(StreamApplicationDescriptorImpl.class);
    OperatorSpec<Object, SamzaSqlRelMessage> mockInputOp = mock(OperatorSpec.class);
    MessageStream<SamzaSqlRelMessage> mockStream = new MessageStreamImpl<>(mockAppDesc, mockInputOp);
    when(mockTranslatorContext.getMessageStream(eq(1))).thenReturn(mockStream);
    doAnswer(this.getRegisterMessageStreamAnswer()).when(mockTranslatorContext).registerMessageStream(eq(2), any(MessageStream.class));
    RexToJavaCompiler mockCompiler = mock(RexToJavaCompiler.class);
    when(mockTranslatorContext.getExpressionCompiler()).thenReturn(mockCompiler);
    Expression mockExpr = mock(Expression.class);
    when(mockCompiler.compile(any(), any())).thenReturn(mockExpr);
    when(mockContext.getContainerContext()).thenReturn(mockContainerContext);
    when(mockContainerContext.getContainerMetricsRegistry()).thenReturn(testMetricsRegistryImpl);
    // Apply translate() method to verify that we are getting the correct map operator constructed
    ProjectTranslator projectTranslator = new ProjectTranslator(1);
    projectTranslator.translate(mockProject, LOGICAL_OP_ID, mockTranslatorContext);
    // make sure that context has been registered with LogicFilter and output message streams
    verify(mockTranslatorContext, times(1)).registerRelNode(2, mockProject);
    verify(mockTranslatorContext, times(1)).registerMessageStream(2, this.getRegisteredMessageStream(2));
    when(mockTranslatorContext.getRelNode(2)).thenReturn(mockProject);
    when(mockTranslatorContext.getMessageStream(2)).thenReturn(this.getRegisteredMessageStream(2));
    StreamOperatorSpec projectSpec = (StreamOperatorSpec) Whitebox.getInternalState(this.getRegisteredMessageStream(2), "operatorSpec");
    assertNotNull(projectSpec);
    assertEquals(projectSpec.getOpCode(), OperatorSpec.OpCode.MAP);
    // Verify that the bootstrap() method will establish the context for the map function
    Map<Integer, TranslatorContext> mockContexts = new HashMap<>();
    mockContexts.put(1, mockTranslatorContext);
    when(mockContext.getApplicationTaskContext()).thenReturn(new SamzaSqlApplicationContext(mockContexts));
    projectSpec.getTransformFn().init(mockContext);
    MapFunction mapFn = (MapFunction) Whitebox.getInternalState(projectSpec, "mapFn");
    assertNotNull(mapFn);
    assertEquals(mockTranslatorContext, Whitebox.getInternalState(mapFn, "translatorContext"));
    assertEquals(mockProject, Whitebox.getInternalState(mapFn, "project"));
    assertEquals(mockExpr, Whitebox.getInternalState(mapFn, "expr"));
    // Verify TestMetricsRegistryImpl works with Project
    assertEquals(1, testMetricsRegistryImpl.getGauges().size());
    assertEquals(2, testMetricsRegistryImpl.getGauges().get(LOGICAL_OP_ID).size());
    assertEquals(1, testMetricsRegistryImpl.getCounters().size());
    assertEquals(2, testMetricsRegistryImpl.getCounters().get(LOGICAL_OP_ID).size());
    assertEquals(0, testMetricsRegistryImpl.getCounters().get(LOGICAL_OP_ID).get(0).getCount());
    assertEquals(0, testMetricsRegistryImpl.getCounters().get(LOGICAL_OP_ID).get(1).getCount());
    // Calling mapFn.apply() to verify the filter function is correctly applied to the input message
    SamzaSqlRelMessage mockInputMsg = new SamzaSqlRelMessage(new ArrayList<>(), new ArrayList<>(), new SamzaSqlRelMsgMetadata(0L, 0L));
    SamzaSqlExecutionContext executionContext = mock(SamzaSqlExecutionContext.class);
    DataContext dataContext = mock(DataContext.class);
    when(mockTranslatorContext.getExecutionContext()).thenReturn(executionContext);
    when(mockTranslatorContext.getDataContext()).thenReturn(dataContext);
    Object[] result = new Object[1];
    final Object mockFieldObj = new Object();
    doAnswer(invocation -> {
        Object[] retValue = invocation.getArgumentAt(4, Object[].class);
        retValue[0] = mockFieldObj;
        return null;
    }).when(mockExpr).execute(eq(executionContext), eq(mockContext), eq(dataContext), eq(mockInputMsg.getSamzaSqlRelRecord().getFieldValues().toArray()), eq(result));
    SamzaSqlRelMessage retMsg = (SamzaSqlRelMessage) mapFn.apply(mockInputMsg);
    assertEquals(retMsg.getSamzaSqlRelRecord().getFieldNames(), Collections.singletonList("test_field"));
    assertEquals(retMsg.getSamzaSqlRelRecord().getFieldValues(), Collections.singletonList(mockFieldObj));
    // Verify mapFn.apply() updates the TestMetricsRegistryImpl metrics
    assertEquals(1, testMetricsRegistryImpl.getCounters().get(LOGICAL_OP_ID).get(0).getCount());
    assertEquals(1, testMetricsRegistryImpl.getCounters().get(LOGICAL_OP_ID).get(1).getCount());
}

Example 12 with ContainerContext

use of org.apache.samza.context.ContainerContext in project samza by apache.

the class SamzaSqlApplication method describe.

@Override
public void describe(StreamApplicationDescriptor appDescriptor) {
    try {
        // TODO: Introduce an API to return a dsl string containing one or more sql statements.
        List<String> dslStmts = SamzaSqlDslConverter.fetchSqlFromConfig(appDescriptor.getConfig());
        Map<Integer, TranslatorContext> translatorContextMap = new HashMap<>();
        // 1. Get Calcite plan
        List<String> inputSystemStreams = new LinkedList<>();
        List<String> outputSystemStreams = new LinkedList<>();
        Collection<RelRoot> relRoots = SamzaSqlApplicationConfig.populateSystemStreamsAndGetRelRoots(dslStmts, appDescriptor.getConfig(), inputSystemStreams, outputSystemStreams);
        // 2. Populate configs
        SamzaSqlApplicationConfig sqlConfig = new SamzaSqlApplicationConfig(appDescriptor.getConfig(), inputSystemStreams, outputSystemStreams);
        // 3. Translate Calcite plan to Samza stream operators
        QueryTranslator queryTranslator = new QueryTranslator(appDescriptor, sqlConfig);
        SamzaSqlExecutionContext executionContext = new SamzaSqlExecutionContext(sqlConfig);
        // QueryId implies the index of the query in multiple query statements scenario. It should always start with 0.
        int queryId = 0;
        for (RelRoot relRoot : relRoots) {
            LOG.info("Translating relRoot {} to samza stream graph with queryId {}", relRoot, queryId);
            TranslatorContext translatorContext = new TranslatorContext(appDescriptor, relRoot, executionContext);
            translatorContextMap.put(queryId, translatorContext);
            queryTranslator.translate(relRoot, sqlConfig.getOutputSystemStreams().get(queryId), translatorContext, queryId);
            queryId++;
        }
        // 4. Set all translator contexts
        /*
       * TODO When serialization of ApplicationDescriptor is actually needed, then something will need to be updated here,
       * since translatorContext is not Serializable. Currently, a new ApplicationDescriptor instance is created in each
       * container, so it does not need to be serialized. Therefore, the translatorContext is recreated in each container
       * and does not need to be serialized.
       */
        appDescriptor.withApplicationTaskContextFactory(new ApplicationTaskContextFactory<SamzaSqlApplicationContext>() {

            @Override
            public SamzaSqlApplicationContext create(ExternalContext externalContext, JobContext jobContext, ContainerContext containerContext, TaskContext taskContext, ApplicationContainerContext applicationContainerContext) {
                return new SamzaSqlApplicationContext(translatorContextMap);
            }
        });
    } catch (RuntimeException e) {
        LOG.error("SamzaSqlApplication threw exception.", e);
        throw e;
    }
}

Example 13 with ContainerContext

use of org.apache.samza.context.ContainerContext in project samza by apache.

the class TranslatorInputMetricsMapFunction method init.

/**
 * initializes the TranslatorOutputMetricsMapFunction before any message is processed
 * @param context the {@link Context} for this task
 */
@Override
public void init(Context context) {
    ContainerContext containerContext = context.getContainerContext();
    metricsRegistry = containerContext.getContainerMetricsRegistry();
    inputEvents = metricsRegistry.newCounter(logicalOpId, TranslatorConstants.INPUT_EVENTS_NAME);
    inputEvents.clear();
}

Example 14 with ContainerContext

use of org.apache.samza.context.ContainerContext in project samza by apache.

the class TranslatorOutputMetricsMapFunction method init.

/**
 * initializes the TranslatorOutputMetricsMapFunction before any message is processed
 * @param context the {@link Context} for this task
 */
@Override
public void init(Context context) {
    ContainerContext containerContext = context.getContainerContext();
    metricsRegistry = containerContext.getContainerMetricsRegistry();
    processingTime = new SamzaHistogram(metricsRegistry, logicalOpId, TranslatorConstants.PROCESSING_TIME_NAME);
    outputEvents = metricsRegistry.newCounter(logicalOpId, TranslatorConstants.OUTPUT_EVENTS_NAME);
    outputEvents.clear();
}

Example 15 with ContainerContext

use of org.apache.samza.context.ContainerContext in project samza by apache.

the class ContainerStorageManager method restoreStores.

// Restoration of all stores, in parallel across tasks
private void restoreStores() throws InterruptedException {
    LOG.info("Store Restore started");
    Set<TaskName> activeTasks = getTasks(containerModel, TaskMode.Active).keySet();
    // TODO HIGH dchen verify davinci lifecycle
    // Find all non-side input stores
    Set<String> nonSideInputStoreNames = storageEngineFactories.keySet().stream().filter(storeName -> !sideInputStoreNames.contains(storeName)).collect(Collectors.toSet());
    // Obtain the checkpoints for each task
    Map<TaskName, Map<String, TaskRestoreManager>> taskRestoreManagers = new HashMap<>();
    Map<TaskName, Checkpoint> taskCheckpoints = new HashMap<>();
    containerModel.getTasks().forEach((taskName, taskModel) -> {
        Checkpoint taskCheckpoint = null;
        if (checkpointManager != null && activeTasks.contains(taskName)) {
            // only pass in checkpoints for active tasks
            taskCheckpoint = checkpointManager.readLastCheckpoint(taskName);
            LOG.info("Obtained checkpoint: {} for state restore for taskName: {}", taskCheckpoint, taskName);
        }
        taskCheckpoints.put(taskName, taskCheckpoint);
        Map<String, Set<String>> backendFactoryStoreNames = getBackendFactoryStoreNames(taskCheckpoint, nonSideInputStoreNames, new StorageConfig(config));
        Map<String, TaskRestoreManager> taskStoreRestoreManagers = createTaskRestoreManagers(restoreStateBackendFactories, backendFactoryStoreNames, clock, samzaContainerMetrics, taskName, taskModel);
        taskRestoreManagers.put(taskName, taskStoreRestoreManagers);
    });
    // Initialize each TaskStorageManager
    taskRestoreManagers.forEach((taskName, restoreManagers) -> restoreManagers.forEach((factoryName, taskRestoreManager) -> taskRestoreManager.init(taskCheckpoints.get(taskName))));
    // Start each store consumer once.
    // Note: These consumers are per system and only changelog system store consumers will be started.
    // Some TaskRestoreManagers may not require the consumer to to be started, but due to the agnostic nature of
    // ContainerStorageManager we always start the changelog consumer here in case it is required
    this.storeConsumers.values().stream().distinct().forEach(SystemConsumer::start);
    List<Future<Void>> taskRestoreFutures = new ArrayList<>();
    // Submit restore callable for each taskInstance
    taskRestoreManagers.forEach((taskInstance, restoreManagersMap) -> {
        // Submit for each restore factory
        restoreManagersMap.forEach((factoryName, taskRestoreManager) -> {
            long startTime = System.currentTimeMillis();
            String taskName = taskInstance.getTaskName();
            LOG.info("Starting restore for state for task: {}", taskName);
            CompletableFuture<Void> restoreFuture = taskRestoreManager.restore().handle((res, ex) -> {
                // on stop, so paralleling stop() also parallelizes their compaction (a time-intensive operation).
                try {
                    taskRestoreManager.close();
                } catch (Exception e) {
                    LOG.error("Error closing restore manager for task: {} after {} restore", taskName, ex != null ? "unsuccessful" : "successful", e);
                // ignore exception from close. container may still be be able to continue processing/backups
                // if restore manager close fails.
                }
                long timeToRestore = System.currentTimeMillis() - startTime;
                if (samzaContainerMetrics != null) {
                    Gauge taskGauge = samzaContainerMetrics.taskStoreRestorationMetrics().getOrDefault(taskInstance, null);
                    if (taskGauge != null) {
                        taskGauge.set(timeToRestore);
                    }
                }
                if (ex != null) {
                    // log and rethrow exception to communicate restore failure
                    String msg = String.format("Error restoring state for task: %s", taskName);
                    LOG.error(msg, ex);
                    // wrap in unchecked exception to throw from lambda
                    throw new SamzaException(msg, ex);
                } else {
                    return null;
                }
            });
            taskRestoreFutures.add(restoreFuture);
        });
    });
    // as samza exceptions
    for (Future<Void> future : taskRestoreFutures) {
        try {
            future.get();
        } catch (InterruptedException e) {
            LOG.warn("Received an interrupt during store restoration. Interrupting the restore executor to exit " + "prematurely without restoring full state.");
            restoreExecutor.shutdownNow();
            throw e;
        } catch (Exception e) {
            LOG.error("Exception when restoring state.", e);
            throw new SamzaException("Exception when restoring state.", e);
        }
    }
    // Stop each store consumer once
    this.storeConsumers.values().stream().distinct().forEach(SystemConsumer::stop);
    // Now create persistent non side input stores in read-write mode, leave non-persistent stores as-is
    this.taskStores = createTaskStores(nonSideInputStoreNames, this.containerModel, jobContext, containerContext, storageEngineFactories, serdes, taskInstanceMetrics, taskInstanceCollectors);
    // Add in memory stores
    this.inMemoryStores.forEach((taskName, stores) -> {
        if (!this.taskStores.containsKey(taskName)) {
            taskStores.put(taskName, new HashMap<>());
        }
        taskStores.get(taskName).putAll(stores);
    });
    // Add side input stores
    this.sideInputStores.forEach((taskName, stores) -> {
        if (!this.taskStores.containsKey(taskName)) {
            taskStores.put(taskName, new HashMap<>());
        }
        taskStores.get(taskName).putAll(stores);
    });
    LOG.info("Store Restore complete");
}