Examples with WindowingStrategy org.apache.beam.sdk.values.WindowingStrategy used on opensource projects

Example 16 with WindowingStrategy

use of org.apache.beam.sdk.values.WindowingStrategy in project beam by apache.

the class DoFnOperatorTest method testTimersRestore.

@Test
public void testTimersRestore() throws Exception {
    final Instant timerTimestamp = new Instant(1000);
    final String outputMessage = "Timer fired";
    WindowingStrategy<Object, IntervalWindow> windowingStrategy = WindowingStrategy.of(FixedWindows.of(Duration.millis(10_000)));
    DoFn<Integer, String> fn = new DoFn<Integer, String>() {

        private static final String EVENT_TIMER_ID = "eventTimer";

        @TimerId(EVENT_TIMER_ID)
        private final TimerSpec eventTimer = TimerSpecs.timer(TimeDomain.EVENT_TIME);

        @ProcessElement
        public void processElement(ProcessContext context, @TimerId(EVENT_TIMER_ID) Timer timer) {
            timer.set(timerTimestamp);
        }

        @OnTimer(EVENT_TIMER_ID)
        public void onEventTime(OnTimerContext context) {
            assertEquals("Timer timestamp must match set timestamp.", timerTimestamp, context.timestamp());
            context.outputWithTimestamp(outputMessage, context.timestamp());
        }
    };
    VarIntCoder keyCoder = VarIntCoder.of();
    WindowedValue.FullWindowedValueCoder<Integer> inputCoder = WindowedValue.getFullCoder(keyCoder, windowingStrategy.getWindowFn().windowCoder());
    WindowedValue.FullWindowedValueCoder<String> outputCoder = WindowedValue.getFullCoder(StringUtf8Coder.of(), windowingStrategy.getWindowFn().windowCoder());
    TupleTag<String> outputTag = new TupleTag<>("main-output");
    final CoderTypeSerializer<WindowedValue<String>> outputSerializer = new CoderTypeSerializer<>(outputCoder, new SerializablePipelineOptions(FlinkPipelineOptions.defaults()));
    CoderTypeInformation<ByteBuffer> keyCoderInfo = new CoderTypeInformation<>(FlinkKeyUtils.ByteBufferCoder.of(), FlinkPipelineOptions.defaults());
    KeySelector<WindowedValue<Integer>, ByteBuffer> keySelector = e -> FlinkKeyUtils.encodeKey(e.getValue(), keyCoder);
    OneInputStreamOperatorTestHarness<WindowedValue<Integer>, WindowedValue<String>> testHarness = createTestHarness(windowingStrategy, fn, inputCoder, outputCoder, keyCoder, outputTag, keyCoderInfo, keySelector);
    testHarness.setup(outputSerializer);
    testHarness.open();
    testHarness.processWatermark(0);
    IntervalWindow window1 = new IntervalWindow(new Instant(0), Duration.millis(10_000));
    // this should register a timer
    testHarness.processElement(new StreamRecord<>(WindowedValue.of(13, new Instant(0), window1, PaneInfo.NO_FIRING)));
    assertThat(stripStreamRecordFromWindowedValue(testHarness.getOutput()), emptyIterable());
    // snapshot and restore
    final OperatorSubtaskState snapshot = testHarness.snapshot(0, 0);
    testHarness.close();
    testHarness = createTestHarness(windowingStrategy, fn, inputCoder, outputCoder, VarIntCoder.of(), outputTag, keyCoderInfo, keySelector);
    testHarness.setup(outputSerializer);
    testHarness.initializeState(snapshot);
    testHarness.open();
    // this must fire the timer
    testHarness.processWatermark(timerTimestamp.getMillis() + 1);
    assertThat(stripStreamRecordFromWindowedValue(testHarness.getOutput()), contains(WindowedValue.of(outputMessage, timerTimestamp, window1, PaneInfo.NO_FIRING)));
    testHarness.close();
}

Example 17 with WindowingStrategy

use of org.apache.beam.sdk.values.WindowingStrategy in project beam by apache.

the class WindowingStrategyTranslation method toProto.

/**
 * Converts a {@link WindowingStrategy} into a {@link RunnerApi.WindowingStrategy}, registering
 * any components in the provided {@link SdkComponents}.
 */
public static RunnerApi.WindowingStrategy toProto(WindowingStrategy<?, ?> windowingStrategy, SdkComponents components) throws IOException {
    WindowFn<?, ?> windowFn = windowingStrategy.getWindowFn();
    FunctionSpec windowFnSpec = toProto(windowFn, components);
    String environmentId = Strings.isNullOrEmpty(windowingStrategy.getEnvironmentId()) ? components.getEnvironmentIdFor(ResourceHints.create()) : windowingStrategy.getEnvironmentId();
    RunnerApi.WindowingStrategy.Builder windowingStrategyProto = RunnerApi.WindowingStrategy.newBuilder().setOutputTime(toProto(windowingStrategy.getTimestampCombiner())).setAccumulationMode(toProto(windowingStrategy.getMode())).setClosingBehavior(toProto(windowingStrategy.getClosingBehavior())).setAllowedLateness(windowingStrategy.getAllowedLateness().getMillis()).setTrigger(TriggerTranslation.toProto(windowingStrategy.getTrigger())).setWindowFn(windowFnSpec).setAssignsToOneWindow(windowFn.assignsToOneWindow()).setMergeStatus(windowFn.isNonMerging() ? MergeStatus.Enum.NON_MERGING : (windowingStrategy.isAlreadyMerged() ? MergeStatus.Enum.ALREADY_MERGED : MergeStatus.Enum.NEEDS_MERGE)).setOnTimeBehavior(toProto(windowingStrategy.getOnTimeBehavior())).setWindowCoderId(components.registerCoder(windowFn.windowCoder())).setEnvironmentId(environmentId);
    return windowingStrategyProto.build();
}

Example 18 with WindowingStrategy

use of org.apache.beam.sdk.values.WindowingStrategy in project beam by apache.

the class LateDataUtils method dropExpiredWindows.

/**
 * Returns an {@code Iterable<WindowedValue<InputT>>} that only contains non-late input elements.
 */
public static <K, V> Iterable<WindowedValue<V>> dropExpiredWindows(final K key, Iterable<WindowedValue<V>> elements, final TimerInternals timerInternals, final WindowingStrategy<?, ?> windowingStrategy, final CounterCell droppedDueToLateness) {
    return FluentIterable.from(elements).transformAndConcat(// Explode windows to filter out expired ones
    input -> {
        if (input == null) {
            return null;
        }
        return input.explodeWindows();
    }).filter(input -> {
        if (input == null) {
            // drop null elements.
            return false;
        }
        BoundedWindow window = Iterables.getOnlyElement(input.getWindows());
        boolean expired = window.maxTimestamp().plus(windowingStrategy.getAllowedLateness()).isBefore(timerInternals.currentInputWatermarkTime());
        if (expired) {
            // The element is too late for this window.
            droppedDueToLateness.inc();
            WindowTracing.debug("GroupAlsoByWindow: Dropping element at {} for key: {}; " + "window: {} since it is too far behind inputWatermark: {}", input.getTimestamp(), key, window, timerInternals.currentInputWatermarkTime());
        }
        // Keep the element if the window is not expired.
        return !expired;
    });
}

Example 19 with WindowingStrategy

use of org.apache.beam.sdk.values.WindowingStrategy in project beam by apache.

the class DoFnOperatorTest method testLateDroppingForStatefulFn.

@Test
public void testLateDroppingForStatefulFn() throws Exception {
    WindowingStrategy<Object, IntervalWindow> windowingStrategy = WindowingStrategy.of(FixedWindows.of(Duration.millis(10)));
    DoFn<Integer, String> fn = new DoFn<Integer, String>() {

        @StateId("state")
        private final StateSpec<ValueState<String>> stateSpec = StateSpecs.value(StringUtf8Coder.of());

        @ProcessElement
        public void processElement(ProcessContext context) {
            context.output(context.element().toString());
        }
    };
    VarIntCoder keyCoder = VarIntCoder.of();
    Coder<WindowedValue<Integer>> inputCoder = WindowedValue.getFullCoder(keyCoder, windowingStrategy.getWindowFn().windowCoder());
    Coder<WindowedValue<String>> outputCoder = WindowedValue.getFullCoder(StringUtf8Coder.of(), windowingStrategy.getWindowFn().windowCoder());
    KeySelector<WindowedValue<Integer>, ByteBuffer> keySelector = e -> FlinkKeyUtils.encodeKey(e.getValue(), keyCoder);
    TupleTag<String> outputTag = new TupleTag<>("main-output");
    DoFnOperator<Integer, String> doFnOperator = new DoFnOperator<>(fn, "stepName", inputCoder, Collections.emptyMap(), outputTag, Collections.emptyList(), new DoFnOperator.MultiOutputOutputManagerFactory<>(outputTag, outputCoder, new SerializablePipelineOptions(FlinkPipelineOptions.defaults())), windowingStrategy, new HashMap<>(), /* side-input mapping */
    Collections.emptyList(), /* side inputs */
    FlinkPipelineOptions.defaults(), keyCoder, /* key coder */
    keySelector, DoFnSchemaInformation.create(), Collections.emptyMap());
    OneInputStreamOperatorTestHarness<WindowedValue<Integer>, WindowedValue<String>> testHarness = new KeyedOneInputStreamOperatorTestHarness<>(doFnOperator, keySelector, new CoderTypeInformation<>(FlinkKeyUtils.ByteBufferCoder.of(), FlinkPipelineOptions.defaults()));
    testHarness.open();
    testHarness.processWatermark(0);
    IntervalWindow window1 = new IntervalWindow(new Instant(0), Duration.millis(10));
    // this should not be late
    testHarness.processElement(new StreamRecord<>(WindowedValue.of(13, new Instant(0), window1, PaneInfo.NO_FIRING)));
    assertThat(stripStreamRecordFromWindowedValue(testHarness.getOutput()), contains(WindowedValue.of("13", new Instant(0), window1, PaneInfo.NO_FIRING)));
    testHarness.getOutput().clear();
    testHarness.processWatermark(9);
    // this should still not be considered late
    testHarness.processElement(new StreamRecord<>(WindowedValue.of(17, new Instant(0), window1, PaneInfo.NO_FIRING)));
    assertThat(stripStreamRecordFromWindowedValue(testHarness.getOutput()), contains(WindowedValue.of("17", new Instant(0), window1, PaneInfo.NO_FIRING)));
    testHarness.getOutput().clear();
    testHarness.processWatermark(10);
    // this should now be considered late
    testHarness.processElement(new StreamRecord<>(WindowedValue.of(17, new Instant(0), window1, PaneInfo.NO_FIRING)));
    assertThat(stripStreamRecordFromWindowedValue(testHarness.getOutput()), emptyIterable());
    testHarness.close();
}

Example 20 with WindowingStrategy

use of org.apache.beam.sdk.values.WindowingStrategy in project beam by apache.

the class DoFnOperatorTest method testWatermarkContract.

/**
 * This test specifically verifies that we correctly map Flink watermarks to Beam watermarks. In
 * Beam, a watermark {@code T} guarantees there will not be elements with a timestamp {@code < T}
 * in the future. In Flink, a watermark {@code T} guarantees there will not be elements with a
 * timestamp {@code <= T} in the future. We have to make sure to take this into account when
 * firing timers.
 *
 * <p>This does not test the timer API in general or processing-time timers because there are
 * generic tests for this in {@code ParDoTest}.
 */
@Test
public void testWatermarkContract() throws Exception {
    final Instant timerTimestamp = new Instant(1000);
    final Instant timerOutputTimestamp = timerTimestamp.minus(Duration.millis(1));
    final String eventTimeMessage = "Event timer fired: ";
    final String processingTimeMessage = "Processing timer fired";
    WindowingStrategy<Object, IntervalWindow> windowingStrategy = WindowingStrategy.of(FixedWindows.of(Duration.millis(10_000)));
    final String eventTimerId = "eventTimer";
    final String eventTimerId2 = "eventTimer2";
    final String processingTimerId = "processingTimer";
    DoFn<Integer, String> fn = new DoFn<Integer, String>() {

        @TimerId(eventTimerId)
        private final TimerSpec eventTimer = TimerSpecs.timer(TimeDomain.EVENT_TIME);

        @TimerId(eventTimerId2)
        private final TimerSpec eventTimer2 = TimerSpecs.timer(TimeDomain.EVENT_TIME);

        @TimerId(processingTimerId)
        private final TimerSpec processingTimer = TimerSpecs.timer(TimeDomain.PROCESSING_TIME);

        @ProcessElement
        public void processElement(ProcessContext context, @TimerId(eventTimerId) Timer eventTimer, @TimerId(eventTimerId2) Timer eventTimerWithOutputTimestamp, @TimerId(processingTimerId) Timer processingTimer) {
            eventTimer.set(timerTimestamp);
            eventTimerWithOutputTimestamp.withOutputTimestamp(timerOutputTimestamp).set(timerTimestamp);
            processingTimer.offset(Duration.millis(timerTimestamp.getMillis())).setRelative();
        }

        @OnTimer(eventTimerId)
        public void onEventTime(OnTimerContext context) {
            assertEquals("Timer timestamp must match set timestamp.", timerTimestamp, context.timestamp());
            context.outputWithTimestamp(eventTimeMessage + eventTimerId, context.timestamp());
        }

        @OnTimer(eventTimerId2)
        public void onEventTime2(OnTimerContext context) {
            assertEquals("Timer timestamp must match set timestamp.", timerTimestamp, context.fireTimestamp());
            context.output(eventTimeMessage + eventTimerId2);
        }

        @OnTimer(processingTimerId)
        public void onProcessingTime(OnTimerContext context) {
            assertEquals(// See SimpleDoFnRunner#onTimer
            "Timer timestamp must match current input watermark", timerTimestamp.plus(Duration.millis(1)), context.timestamp());
            context.outputWithTimestamp(processingTimeMessage, context.timestamp());
        }
    };
    VarIntCoder keyCoder = VarIntCoder.of();
    WindowedValue.FullWindowedValueCoder<Integer> inputCoder = WindowedValue.getFullCoder(keyCoder, windowingStrategy.getWindowFn().windowCoder());
    WindowedValue.FullWindowedValueCoder<String> outputCoder = WindowedValue.getFullCoder(StringUtf8Coder.of(), windowingStrategy.getWindowFn().windowCoder());
    KeySelector<WindowedValue<Integer>, ByteBuffer> keySelector = e -> FlinkKeyUtils.encodeKey(e.getValue(), keyCoder);
    TupleTag<String> outputTag = new TupleTag<>("main-output");
    DoFnOperator<Integer, String> doFnOperator = new DoFnOperator<>(fn, "stepName", inputCoder, Collections.emptyMap(), outputTag, Collections.emptyList(), new DoFnOperator.MultiOutputOutputManagerFactory<>(outputTag, outputCoder, new SerializablePipelineOptions(FlinkPipelineOptions.defaults())), windowingStrategy, new HashMap<>(), /* side-input mapping */
    Collections.emptyList(), /* side inputs */
    FlinkPipelineOptions.defaults(), keyCoder, /* key coder */
    keySelector, DoFnSchemaInformation.create(), Collections.emptyMap());
    OneInputStreamOperatorTestHarness<WindowedValue<Integer>, WindowedValue<String>> testHarness = new KeyedOneInputStreamOperatorTestHarness<>(doFnOperator, keySelector, new CoderTypeInformation<>(FlinkKeyUtils.ByteBufferCoder.of(), FlinkPipelineOptions.defaults()));
    testHarness.setup(new CoderTypeSerializer<>(outputCoder, new SerializablePipelineOptions(FlinkPipelineOptions.defaults())));
    testHarness.open();
    testHarness.processWatermark(0);
    testHarness.setProcessingTime(0);
    IntervalWindow window1 = new IntervalWindow(new Instant(0), Duration.millis(10_000));
    // this should register the two timers above
    testHarness.processElement(new StreamRecord<>(WindowedValue.of(13, new Instant(0), window1, PaneInfo.NO_FIRING)));
    assertThat(stripStreamRecordFromWindowedValue(testHarness.getOutput()), emptyIterable());
    // this does not yet fire the timers (in vanilla Flink it would)
    testHarness.processWatermark(timerTimestamp.getMillis());
    testHarness.setProcessingTime(timerTimestamp.getMillis());
    assertThat(stripStreamRecordFromWindowedValue(testHarness.getOutput()), emptyIterable());
    assertThat(doFnOperator.keyedStateInternals.minWatermarkHoldMs(), is(timerOutputTimestamp.getMillis()));
    // this must fire the event timers
    testHarness.processWatermark(timerTimestamp.getMillis() + 1);
    assertThat(stripStreamRecordFromWindowedValue(testHarness.getOutput()), containsInAnyOrder(WindowedValue.of(eventTimeMessage + eventTimerId, timerTimestamp, window1, PaneInfo.NO_FIRING), WindowedValue.of(eventTimeMessage + eventTimerId2, timerTimestamp.minus(Duration.millis(1)), window1, PaneInfo.NO_FIRING)));
    testHarness.getOutput().clear();
    // this must fire the processing timer
    testHarness.setProcessingTime(timerTimestamp.getMillis() + 1);
    assertThat(stripStreamRecordFromWindowedValue(testHarness.getOutput()), contains(WindowedValue.of(// See SimpleDoFnRunner#onTimer
    processingTimeMessage, timerTimestamp.plus(Duration.millis(1)), window1, PaneInfo.NO_FIRING)));
    testHarness.close();
}