Examples with CheckpointableKeyedStateBackend org.apache.flink.runtime.state.CheckpointableKeyedStateBackend used on opensource projects

Example 1 with CheckpointableKeyedStateBackend

use of org.apache.flink.runtime.state.CheckpointableKeyedStateBackend in project flink by apache.

the class StateInitializationContextImplTest method setUp.

@Before
public void setUp() throws Exception {
    this.writtenKeyGroups = 0;
    this.writtenOperatorStates = new HashSet<>();
    this.closableRegistry = new CloseableRegistry();
    ByteArrayOutputStreamWithPos out = new ByteArrayOutputStreamWithPos(64);
    List<KeyedStateHandle> keyedStateHandles = new ArrayList<>(NUM_HANDLES);
    int prev = 0;
    for (int i = 0; i < NUM_HANDLES; ++i) {
        out.reset();
        int size = i % 4;
        int end = prev + size;
        DataOutputView dov = new DataOutputViewStreamWrapper(out);
        KeyGroupRangeOffsets offsets = new KeyGroupRangeOffsets(i == 9 ? KeyGroupRange.EMPTY_KEY_GROUP_RANGE : new KeyGroupRange(prev, end));
        prev = end + 1;
        for (int kg : offsets.getKeyGroupRange()) {
            offsets.setKeyGroupOffset(kg, out.getPosition());
            dov.writeInt(kg);
            ++writtenKeyGroups;
        }
        KeyedStateHandle handle = new KeyGroupsStateHandle(offsets, new ByteStateHandleCloseChecking("kg-" + i, out.toByteArray()));
        keyedStateHandles.add(handle);
    }
    List<OperatorStateHandle> operatorStateHandles = new ArrayList<>(NUM_HANDLES);
    for (int i = 0; i < NUM_HANDLES; ++i) {
        int size = i % 4;
        out.reset();
        DataOutputView dov = new DataOutputViewStreamWrapper(out);
        LongArrayList offsets = new LongArrayList(size);
        for (int s = 0; s < size; ++s) {
            offsets.add(out.getPosition());
            int val = i * NUM_HANDLES + s;
            dov.writeInt(val);
            writtenOperatorStates.add(val);
        }
        Map<String, OperatorStateHandle.StateMetaInfo> offsetsMap = new HashMap<>();
        offsetsMap.put(DefaultOperatorStateBackend.DEFAULT_OPERATOR_STATE_NAME, new OperatorStateHandle.StateMetaInfo(offsets.toArray(), OperatorStateHandle.Mode.SPLIT_DISTRIBUTE));
        OperatorStateHandle operatorStateHandle = new OperatorStreamStateHandle(offsetsMap, new ByteStateHandleCloseChecking("os-" + i, out.toByteArray()));
        operatorStateHandles.add(operatorStateHandle);
    }
    OperatorSubtaskState operatorSubtaskState = OperatorSubtaskState.builder().setRawOperatorState(new StateObjectCollection<>(operatorStateHandles)).setRawKeyedState(new StateObjectCollection<>(keyedStateHandles)).build();
    OperatorID operatorID = new OperatorID();
    TaskStateSnapshot taskStateSnapshot = new TaskStateSnapshot();
    taskStateSnapshot.putSubtaskStateByOperatorID(operatorID, operatorSubtaskState);
    JobManagerTaskRestore jobManagerTaskRestore = new JobManagerTaskRestore(0L, taskStateSnapshot);
    TaskStateManager manager = new TaskStateManagerImpl(new JobID(), new ExecutionAttemptID(), new TestTaskLocalStateStore(), new InMemoryStateChangelogStorage(), jobManagerTaskRestore, mock(CheckpointResponder.class));
    DummyEnvironment environment = new DummyEnvironment("test", 1, 0, prev);
    environment.setTaskStateManager(manager);
    StateBackend stateBackend = new MemoryStateBackend(1024);
    StreamTaskStateInitializer streamTaskStateManager = new StreamTaskStateInitializerImpl(environment, stateBackend, TtlTimeProvider.DEFAULT, new InternalTimeServiceManager.Provider() {

        @Override
        public <K> InternalTimeServiceManager<K> create(CheckpointableKeyedStateBackend<K> keyedStatedBackend, ClassLoader userClassloader, KeyContext keyContext, ProcessingTimeService processingTimeService, Iterable<KeyGroupStatePartitionStreamProvider> rawKeyedStates) throws Exception {
            // stream.
            return null;
        }
    });
    AbstractStreamOperator<?> mockOperator = mock(AbstractStreamOperator.class);
    when(mockOperator.getOperatorID()).thenReturn(operatorID);
    StreamOperatorStateContext stateContext = streamTaskStateManager.streamOperatorStateContext(operatorID, "TestOperatorClass", mock(ProcessingTimeService.class), mockOperator, // consumed by the timer service.
    IntSerializer.INSTANCE, closableRegistry, new UnregisteredMetricsGroup(), 1.0, false);
    OptionalLong restoredCheckpointId = stateContext.getRestoredCheckpointId();
    this.initializationContext = new StateInitializationContextImpl(restoredCheckpointId.isPresent() ? restoredCheckpointId.getAsLong() : null, stateContext.operatorStateBackend(), mock(KeyedStateStore.class), stateContext.rawKeyedStateInputs(), stateContext.rawOperatorStateInputs());
}

Example 2 with CheckpointableKeyedStateBackend

use of org.apache.flink.runtime.state.CheckpointableKeyedStateBackend in project flink by apache.

the class StreamTaskStateInitializerImpl method keyedStatedBackend.

protected <K> CheckpointableKeyedStateBackend<K> keyedStatedBackend(TypeSerializer<K> keySerializer, String operatorIdentifierText, PrioritizedOperatorSubtaskState prioritizedOperatorSubtaskStates, CloseableRegistry backendCloseableRegistry, MetricGroup metricGroup, double managedMemoryFraction) throws Exception {
    if (keySerializer == null) {
        return null;
    }
    String logDescription = "keyed state backend for " + operatorIdentifierText;
    TaskInfo taskInfo = environment.getTaskInfo();
    final KeyGroupRange keyGroupRange = KeyGroupRangeAssignment.computeKeyGroupRangeForOperatorIndex(taskInfo.getMaxNumberOfParallelSubtasks(), taskInfo.getNumberOfParallelSubtasks(), taskInfo.getIndexOfThisSubtask());
    // Now restore processing is included in backend building/constructing process, so we need
    // to make sure
    // each stream constructed in restore could also be closed in case of task cancel, for
    // example the data
    // input stream opened for serDe during restore.
    CloseableRegistry cancelStreamRegistryForRestore = new CloseableRegistry();
    backendCloseableRegistry.registerCloseable(cancelStreamRegistryForRestore);
    BackendRestorerProcedure<CheckpointableKeyedStateBackend<K>, KeyedStateHandle> backendRestorer = new BackendRestorerProcedure<>((stateHandles) -> stateBackend.createKeyedStateBackend(environment, environment.getJobID(), operatorIdentifierText, keySerializer, taskInfo.getMaxNumberOfParallelSubtasks(), keyGroupRange, environment.getTaskKvStateRegistry(), ttlTimeProvider, metricGroup, stateHandles, cancelStreamRegistryForRestore, managedMemoryFraction), backendCloseableRegistry, logDescription);
    try {
        return backendRestorer.createAndRestore(prioritizedOperatorSubtaskStates.getPrioritizedManagedKeyedState());
    } finally {
        if (backendCloseableRegistry.unregisterCloseable(cancelStreamRegistryForRestore)) {
            IOUtils.closeQuietly(cancelStreamRegistryForRestore);
        }
    }
}

Example 3 with CheckpointableKeyedStateBackend

use of org.apache.flink.runtime.state.CheckpointableKeyedStateBackend in project flink by apache.

the class StreamOperatorStateHandler method setCurrentKey.

@SuppressWarnings({ "unchecked" })
public void setCurrentKey(Object key) {
    if (keyedStateBackend != null) {
        try {
            // need to work around type restrictions
            @SuppressWarnings("rawtypes") CheckpointableKeyedStateBackend rawBackend = keyedStateBackend;
            rawBackend.setCurrentKey(key);
        } catch (Exception e) {
            throw new RuntimeException("Exception occurred while setting the current key context.", e);
        }
    }
}

Example 4 with CheckpointableKeyedStateBackend

use of org.apache.flink.runtime.state.CheckpointableKeyedStateBackend in project flink by apache.

the class StreamTaskStateInitializerImplTest method testWithRestore.

@SuppressWarnings("unchecked")
@Test
public void testWithRestore() throws Exception {
    StateBackend mockingBackend = spy(new StateBackend() {

        @Override
        public <K> AbstractKeyedStateBackend<K> createKeyedStateBackend(Environment env, JobID jobID, String operatorIdentifier, TypeSerializer<K> keySerializer, int numberOfKeyGroups, KeyGroupRange keyGroupRange, TaskKvStateRegistry kvStateRegistry, TtlTimeProvider ttlTimeProvider, MetricGroup metricGroup, @Nonnull Collection<KeyedStateHandle> stateHandles, CloseableRegistry cancelStreamRegistry) throws Exception {
            return mock(AbstractKeyedStateBackend.class);
        }

        @Override
        public OperatorStateBackend createOperatorStateBackend(Environment env, String operatorIdentifier, @Nonnull Collection<OperatorStateHandle> stateHandles, CloseableRegistry cancelStreamRegistry) throws Exception {
            return mock(OperatorStateBackend.class);
        }
    });
    OperatorID operatorID = new OperatorID(47L, 11L);
    TaskStateSnapshot taskStateSnapshot = new TaskStateSnapshot();
    Random random = new Random(0x42);
    OperatorSubtaskState operatorSubtaskState = OperatorSubtaskState.builder().setManagedOperatorState(new OperatorStreamStateHandle(Collections.singletonMap("a", new OperatorStateHandle.StateMetaInfo(new long[] { 0, 10 }, SPLIT_DISTRIBUTE)), CheckpointTestUtils.createDummyStreamStateHandle(random, null))).setRawOperatorState(new OperatorStreamStateHandle(Collections.singletonMap("_default_", new OperatorStateHandle.StateMetaInfo(new long[] { 0, 20, 30 }, SPLIT_DISTRIBUTE)), CheckpointTestUtils.createDummyStreamStateHandle(random, null))).setManagedKeyedState(CheckpointTestUtils.createDummyKeyGroupStateHandle(random, null)).setRawKeyedState(CheckpointTestUtils.createDummyKeyGroupStateHandle(random, null)).setInputChannelState(singleton(createNewInputChannelStateHandle(10, random))).setResultSubpartitionState(singleton(createNewResultSubpartitionStateHandle(10, random))).build();
    taskStateSnapshot.putSubtaskStateByOperatorID(operatorID, operatorSubtaskState);
    JobManagerTaskRestore jobManagerTaskRestore = new JobManagerTaskRestore(42L, taskStateSnapshot);
    StreamTaskStateInitializer streamTaskStateManager = streamTaskStateManager(mockingBackend, jobManagerTaskRestore, false);
    AbstractStreamOperator<?> streamOperator = mock(AbstractStreamOperator.class);
    when(streamOperator.getOperatorID()).thenReturn(operatorID);
    TypeSerializer<?> typeSerializer = new IntSerializer();
    CloseableRegistry closeableRegistry = new CloseableRegistry();
    StreamOperatorStateContext stateContext = streamTaskStateManager.streamOperatorStateContext(streamOperator.getOperatorID(), streamOperator.getClass().getSimpleName(), new TestProcessingTimeService(), streamOperator, typeSerializer, closeableRegistry, new UnregisteredMetricsGroup(), 1.0, false);
    OperatorStateBackend operatorStateBackend = stateContext.operatorStateBackend();
    CheckpointableKeyedStateBackend<?> keyedStateBackend = stateContext.keyedStateBackend();
    InternalTimeServiceManager<?> timeServiceManager = stateContext.internalTimerServiceManager();
    CloseableIterable<KeyGroupStatePartitionStreamProvider> keyedStateInputs = stateContext.rawKeyedStateInputs();
    CloseableIterable<StatePartitionStreamProvider> operatorStateInputs = stateContext.rawOperatorStateInputs();
    Assert.assertTrue("Expected the context to be restored", stateContext.isRestored());
    Assert.assertEquals(OptionalLong.of(42L), stateContext.getRestoredCheckpointId());
    Assert.assertNotNull(operatorStateBackend);
    Assert.assertNotNull(keyedStateBackend);
    // this is deactivated on purpose so that it does not attempt to consume the raw keyed
    // state.
    Assert.assertNull(timeServiceManager);
    Assert.assertNotNull(keyedStateInputs);
    Assert.assertNotNull(operatorStateInputs);
    int count = 0;
    for (KeyGroupStatePartitionStreamProvider keyedStateInput : keyedStateInputs) {
        ++count;
    }
    Assert.assertEquals(1, count);
    count = 0;
    for (StatePartitionStreamProvider operatorStateInput : operatorStateInputs) {
        ++count;
    }
    Assert.assertEquals(3, count);
    checkCloseablesRegistered(closeableRegistry, operatorStateBackend, keyedStateBackend, keyedStateInputs, operatorStateInputs);
}

Example 5 with CheckpointableKeyedStateBackend

use of org.apache.flink.runtime.state.CheckpointableKeyedStateBackend in project flink by apache.

the class StreamOperatorSnapshotRestoreTest method testOperatorStatesSnapshotRestoreInternal.

private void testOperatorStatesSnapshotRestoreInternal(final int mode) throws Exception {
    // -------------------------------------------------------------------------- snapshot
    StateBackend stateBackend;
    FsStateBackend fsstateBackend = createStateBackendInternal();
    switch(stateBackendEnum) {
        case FILE:
            stateBackend = fsstateBackend;
            break;
        case ROCKSDB_FULLY_ASYNC:
            stateBackend = new RocksDBStateBackend(fsstateBackend, TernaryBoolean.FALSE);
            break;
        case ROCKSDB_INCREMENTAL:
            stateBackend = new RocksDBStateBackend(fsstateBackend, TernaryBoolean.TRUE);
            break;
        default:
            throw new IllegalStateException(String.format("Do not support statebackend type %s", stateBackendEnum));
    }
    TestOneInputStreamOperator op = new TestOneInputStreamOperator(false);
    JobID jobID = new JobID();
    JobVertexID jobVertexID = new JobVertexID();
    int subtaskIdx = 0;
    LocalRecoveryDirectoryProvider directoryProvider = mode == ONLY_JM_RECOVERY ? null : new LocalRecoveryDirectoryProviderImpl(temporaryFolder.newFolder(), jobID, jobVertexID, subtaskIdx);
    LocalRecoveryConfig localRecoveryConfig = new LocalRecoveryConfig(directoryProvider);
    MockEnvironment mockEnvironment = new MockEnvironmentBuilder().setJobID(jobID).setJobVertexID(jobVertexID).setTaskName("test").setManagedMemorySize(1024L * 1024L).setInputSplitProvider(new MockInputSplitProvider()).setBufferSize(1024 * 1024).setTaskStateManager(new TestTaskStateManager(localRecoveryConfig)).setMaxParallelism(MAX_PARALLELISM).setSubtaskIndex(subtaskIdx).setUserCodeClassLoader(getClass().getClassLoader()).build();
    KeyedOneInputStreamOperatorTestHarness<Integer, Integer, Integer> testHarness = new KeyedOneInputStreamOperatorTestHarness<>(op, (KeySelector<Integer, Integer>) value -> value, TypeInformation.of(Integer.class), mockEnvironment);
    testHarness.setStateBackend(stateBackend);
    testHarness.open();
    for (int i = 0; i < 10; ++i) {
        testHarness.processElement(new StreamRecord<>(i));
    }
    OperatorSnapshotFinalizer snapshotWithLocalState = testHarness.snapshotWithLocalState(1L, 1L);
    testHarness.close();
    // -------------------------------------------------------------------------- restore
    op = new TestOneInputStreamOperator(true);
    testHarness = new KeyedOneInputStreamOperatorTestHarness<>(op, (KeySelector<Integer, Integer>) value -> value, TypeInformation.of(Integer.class), MAX_PARALLELISM, 1, /* num subtasks */
    0);
    testHarness.setTimeServiceManagerProvider(new InternalTimeServiceManager.Provider() {

        @Override
        public <K> InternalTimeServiceManager<K> create(CheckpointableKeyedStateBackend<K> keyedStatedBackend, ClassLoader userClassloader, KeyContext keyContext, ProcessingTimeService processingTimeService, Iterable<KeyGroupStatePartitionStreamProvider> rawKeyedStates) throws IOException {
            return null;
        }
    });
    testHarness.setStateBackend(stateBackend);
    OperatorSubtaskState jobManagerOwnedState = snapshotWithLocalState.getJobManagerOwnedState();
    OperatorSubtaskState taskLocalState = snapshotWithLocalState.getTaskLocalState();
    // We check if local state was created when we enabled local recovery
    Assert.assertTrue(mode > ONLY_JM_RECOVERY == (taskLocalState != null && taskLocalState.hasState()));
    if (mode == TM_REMOVE_JM_RECOVERY) {
        jobManagerOwnedState.getManagedKeyedState().discardState();
    } else if (mode == JM_REMOVE_TM_RECOVERY) {
        taskLocalState.getManagedKeyedState().discardState();
    }
    testHarness.initializeState(jobManagerOwnedState, taskLocalState);
    testHarness.open();
    for (int i = 0; i < 10; ++i) {
        testHarness.processElement(new StreamRecord<>(i));
    }
    testHarness.close();
}