Examples with SystemConsumer org.apache.samza.system.SystemConsumer used on opensource projects

Example 11 with SystemConsumer

use of org.apache.samza.system.SystemConsumer in project samza by apache.

the class NonTransactionalStateTaskRestoreManager method registerStartingOffsets.

/**
 * Determines the starting offset for each store SSP (based on {@link #getStartingOffset(SystemStreamPartition, SystemAdmin)}) and
 * registers it with the respective SystemConsumer for starting consumption.
 */
private void registerStartingOffsets() {
    for (Map.Entry<String, SystemStream> changelogSystemStreamEntry : storeChangelogs.entrySet()) {
        SystemStreamPartition systemStreamPartition = new SystemStreamPartition(changelogSystemStreamEntry.getValue(), taskModel.getChangelogPartition());
        SystemAdmin systemAdmin = systemAdmins.getSystemAdmin(changelogSystemStreamEntry.getValue().getSystem());
        SystemConsumer systemConsumer = storeConsumers.get(changelogSystemStreamEntry.getKey());
        String offset = getStartingOffset(systemStreamPartition, systemAdmin);
        if (offset != null) {
            LOG.info("Registering change log consumer with offset " + offset + " for %" + systemStreamPartition);
            systemConsumer.register(systemStreamPartition, offset);
        } else {
            LOG.info("Skipping change log restoration for {} because stream appears to be empty (offset was null).", systemStreamPartition);
            taskStoresToRestore.remove(changelogSystemStreamEntry.getKey());
        }
    }
}

Example 12 with SystemConsumer

use of org.apache.samza.system.SystemConsumer in project samza by apache.

the class NonTransactionalStateTaskRestoreManager method restore.

/**
 * Restore each store in taskStoresToRestore sequentially
 */
@Override
public CompletableFuture<Void> restore() {
    return CompletableFuture.runAsync(() -> {
        for (String storeName : taskStoresToRestore) {
            LOG.info("Restoring store: {} for task: {}", storeName, taskModel.getTaskName());
            SystemConsumer systemConsumer = storeConsumers.get(storeName);
            SystemStream systemStream = storeChangelogs.get(storeName);
            SystemAdmin systemAdmin = systemAdmins.getSystemAdmin(systemStream.getSystem());
            ChangelogSSPIterator changelogSSPIterator = new ChangelogSSPIterator(systemConsumer, new SystemStreamPartition(systemStream, taskModel.getChangelogPartition()), null, systemAdmin, false);
            try {
                taskStores.get(storeName).restore(changelogSSPIterator);
            } catch (InterruptedException e) {
                String msg = String.format("Interrupted while restoring store: %s for task: %s", storeName, taskModel.getTaskName().getTaskName());
                // wrap in unchecked exception to throw from lambda
                throw new SamzaException(msg, e);
            }
        }
    }, restoreExecutor);
}

Example 13 with SystemConsumer

use of org.apache.samza.system.SystemConsumer in project samza by apache.

the class TestContainerStorageManager method setUp.

/**
 * Method to create a containerStorageManager with mocked dependencies
 */
@Before
public void setUp() throws InterruptedException {
    taskRestoreMetricGauges = new HashMap<>();
    this.tasks = new HashMap<>();
    this.taskInstanceMetrics = new HashMap<>();
    // Add two mocked tasks
    addMockedTask("task 0", 0);
    addMockedTask("task 1", 1);
    // Mock container metrics
    samzaContainerMetrics = mock(SamzaContainerMetrics.class);
    when(samzaContainerMetrics.taskStoreRestorationMetrics()).thenReturn(taskRestoreMetricGauges);
    // Create a map of test changeLogSSPs
    Map<String, SystemStream> changelogSystemStreams = new HashMap<>();
    changelogSystemStreams.put(STORE_NAME, new SystemStream(SYSTEM_NAME, STREAM_NAME));
    // Create mocked storage engine factories
    Map<String, StorageEngineFactory<Object, Object>> storageEngineFactories = new HashMap<>();
    StorageEngineFactory mockStorageEngineFactory = (StorageEngineFactory<Object, Object>) mock(StorageEngineFactory.class);
    StorageEngine mockStorageEngine = mock(StorageEngine.class);
    when(mockStorageEngine.getStoreProperties()).thenReturn(new StoreProperties.StorePropertiesBuilder().setLoggedStore(true).setPersistedToDisk(true).build());
    doAnswer(invocation -> {
        return mockStorageEngine;
    }).when(mockStorageEngineFactory).getStorageEngine(anyString(), any(), any(), any(), any(), any(), any(), any(), any(), any());
    storageEngineFactories.put(STORE_NAME, mockStorageEngineFactory);
    // Add instrumentation to mocked storage engine, to record the number of store.restore() calls
    doAnswer(invocation -> {
        storeRestoreCallCount++;
        return CompletableFuture.completedFuture(null);
    }).when(mockStorageEngine).restore(any());
    // Set the mocked stores' properties to be persistent
    doAnswer(invocation -> {
        return new StoreProperties.StorePropertiesBuilder().setLoggedStore(true).build();
    }).when(mockStorageEngine).getStoreProperties();
    // Mock and setup sysconsumers
    SystemConsumer mockSystemConsumer = mock(SystemConsumer.class);
    doAnswer(invocation -> {
        systemConsumerStartCount++;
        return null;
    }).when(mockSystemConsumer).start();
    doAnswer(invocation -> {
        systemConsumerStopCount++;
        return null;
    }).when(mockSystemConsumer).stop();
    // Create mocked system factories
    Map<String, SystemFactory> systemFactories = new HashMap<>();
    // Count the number of sysConsumers created
    SystemFactory mockSystemFactory = mock(SystemFactory.class);
    doAnswer(invocation -> {
        this.systemConsumerCreationCount++;
        return mockSystemConsumer;
    }).when(mockSystemFactory).getConsumer(anyString(), any(), any());
    systemFactories.put(SYSTEM_NAME, mockSystemFactory);
    // Create mocked configs for specifying serdes
    Map<String, String> configMap = new HashMap<>();
    configMap.put("stores." + STORE_NAME + ".key.serde", "stringserde");
    configMap.put("stores." + STORE_NAME + ".msg.serde", "stringserde");
    configMap.put("stores." + STORE_NAME + ".factory", mockStorageEngineFactory.getClass().getName());
    configMap.put("stores." + STORE_NAME + ".changelog", SYSTEM_NAME + "." + STREAM_NAME);
    configMap.put("serializers.registry.stringserde.class", StringSerdeFactory.class.getName());
    configMap.put(TaskConfig.TRANSACTIONAL_STATE_RETAIN_EXISTING_STATE, "true");
    Config config = new MapConfig(configMap);
    Map<String, Serde<Object>> serdes = new HashMap<>();
    serdes.put("stringserde", mock(Serde.class));
    // Create mocked system admins
    SystemAdmin mockSystemAdmin = mock(SystemAdmin.class);
    doAnswer(new Answer<Void>() {

        public Void answer(InvocationOnMock invocation) {
            Object[] args = invocation.getArguments();
            System.out.println("called with arguments: " + Arrays.toString(args));
            return null;
        }
    }).when(mockSystemAdmin).validateStream(any());
    SystemAdmins mockSystemAdmins = mock(SystemAdmins.class);
    when(mockSystemAdmins.getSystemAdmin("kafka")).thenReturn(mockSystemAdmin);
    // Create a mocked mockStreamMetadataCache
    SystemStreamMetadata.SystemStreamPartitionMetadata sspMetadata = new SystemStreamMetadata.SystemStreamPartitionMetadata("0", "50", "51");
    Map<Partition, SystemStreamMetadata.SystemStreamPartitionMetadata> partitionMetadata = new HashMap<>();
    partitionMetadata.put(new Partition(0), sspMetadata);
    partitionMetadata.put(new Partition(1), sspMetadata);
    SystemStreamMetadata systemStreamMetadata = new SystemStreamMetadata(STREAM_NAME, partitionMetadata);
    StreamMetadataCache mockStreamMetadataCache = mock(StreamMetadataCache.class);
    when(mockStreamMetadataCache.getStreamMetadata(JavaConverters.asScalaSetConverter(new HashSet<SystemStream>(changelogSystemStreams.values())).asScala().toSet(), false)).thenReturn(new scala.collection.immutable.Map.Map1(new SystemStream(SYSTEM_NAME, STREAM_NAME), systemStreamMetadata));
    CheckpointManager checkpointManager = mock(CheckpointManager.class);
    when(checkpointManager.readLastCheckpoint(any(TaskName.class))).thenReturn(new CheckpointV1(new HashMap<>()));
    SSPMetadataCache mockSSPMetadataCache = mock(SSPMetadataCache.class);
    when(mockSSPMetadataCache.getMetadata(any(SystemStreamPartition.class))).thenReturn(new SystemStreamMetadata.SystemStreamPartitionMetadata("0", "10", "11"));
    ContainerContext mockContainerContext = mock(ContainerContext.class);
    ContainerModel mockContainerModel = new ContainerModel("samza-container-test", tasks);
    when(mockContainerContext.getContainerModel()).thenReturn(mockContainerModel);
    // Reset the expected number of sysConsumer create, start and stop calls, and store.restore() calls
    this.systemConsumerCreationCount = 0;
    this.systemConsumerStartCount = 0;
    this.systemConsumerStopCount = 0;
    this.storeRestoreCallCount = 0;
    StateBackendFactory backendFactory = mock(StateBackendFactory.class);
    TaskRestoreManager restoreManager = mock(TaskRestoreManager.class);
    ArgumentCaptor<ExecutorService> restoreExecutorCaptor = ArgumentCaptor.forClass(ExecutorService.class);
    when(backendFactory.getRestoreManager(any(), any(), any(), restoreExecutorCaptor.capture(), any(), any(), any(), any(), any(), any(), any())).thenReturn(restoreManager);
    doAnswer(invocation -> {
        storeRestoreCallCount++;
        return CompletableFuture.completedFuture(null);
    }).when(restoreManager).restore();
    // Create the container storage manager
    this.containerStorageManager = new ContainerStorageManager(checkpointManager, mockContainerModel, mockStreamMetadataCache, mockSystemAdmins, changelogSystemStreams, new HashMap<>(), storageEngineFactories, systemFactories, serdes, config, taskInstanceMetrics, samzaContainerMetrics, mock(JobContext.class), mockContainerContext, ImmutableMap.of(StorageConfig.KAFKA_STATE_BACKEND_FACTORY, backendFactory), mock(Map.class), DEFAULT_LOGGED_STORE_BASE_DIR, DEFAULT_STORE_BASE_DIR, null, new SystemClock());
}

Example 14 with SystemConsumer

use of org.apache.samza.system.SystemConsumer 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");
}

Example 15 with SystemConsumer

use of org.apache.samza.system.SystemConsumer in project samza by apache.

the class SystemConsumerBench method start.

public void start() throws IOException, InterruptedException {
    super.start();
    SystemAdmin systemAdmin = factory.getAdmin(systemName, config);
    SystemStreamMetadata ssm = systemAdmin.getSystemStreamMetadata(Collections.singleton(physicalStreamName)).get(physicalStreamName);
    NoOpMetricsRegistry metricsRegistry = new NoOpMetricsRegistry();
    Set<SystemStreamPartition> ssps = createSSPs(systemName, physicalStreamName, startPartition, endPartition);
    SystemConsumer consumer = factory.getConsumer(systemName, config, metricsRegistry);
    for (SystemStreamPartition ssp : ssps) {
        consumer.register(ssp, ssm.getSystemStreamPartitionMetadata().get(ssp.getPartition()).getOldestOffset());
    }
    consumer.start();
    System.out.println("starting consumption at " + Instant.now());
    Instant startTime = Instant.now();
    int numEvents = 0;
    while (numEvents < totalEvents) {
        Map<SystemStreamPartition, List<IncomingMessageEnvelope>> pollResult = consumer.poll(ssps, 2000);
        numEvents += pollResult.values().stream().mapToInt(List::size).sum();
    }
    System.out.println("Ending consumption at " + Instant.now());
    System.out.println(String.format("Event Rate is %s Messages/Sec ", numEvents * 1000 / Duration.between(startTime, Instant.now()).toMillis()));
    consumer.stop();
    System.exit(0);
}