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

Example 51 with KV

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

the class HourlyTeamScore method configureOutput.

/**
   * Create a map of information that describes how to write pipeline output to text. This map
   * is passed to the {@link WriteToText} constructor to write team score sums and
   * includes information about window start time.
   */
protected static Map<String, WriteToText.FieldFn<KV<String, Integer>>> configureOutput() {
    Map<String, WriteToText.FieldFn<KV<String, Integer>>> config = new HashMap<String, WriteToText.FieldFn<KV<String, Integer>>>();
    config.put("team", (c, w) -> c.element().getKey());
    config.put("total_score", (c, w) -> c.element().getValue());
    config.put("window_start", (c, w) -> {
        IntervalWindow window = (IntervalWindow) w;
        return fmt.print(window.start());
    });
    return config;
}

Example 52 with KV

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

the class StatefulParDoEvaluatorFactoryTest method windowCleanupScheduled.

@Test
public void windowCleanupScheduled() throws Exception {
    // To test the factory, first we set up a pipeline and then we use the constructed
    // pipeline to create the right parameters to pass to the factory
    final String stateId = "my-state-id";
    // For consistency, window it into FixedWindows. Actually we will fabricate an input bundle.
    PCollection<KV<String, Integer>> input = pipeline.apply(Create.of(KV.of("hello", 1), KV.of("hello", 2))).apply(Window.<KV<String, Integer>>into(FixedWindows.of(Duration.millis(10))));
    TupleTag<Integer> mainOutput = new TupleTag<>();
    PCollection<Integer> produced = input.apply(new ParDoMultiOverrideFactory.GbkThenStatefulParDo<>(ParDo.of(new DoFn<KV<String, Integer>, Integer>() {

        @StateId(stateId)
        private final StateSpec<ValueState<String>> spec = StateSpecs.value(StringUtf8Coder.of());

        @ProcessElement
        public void process(ProcessContext c) {
        }
    }).withOutputTags(mainOutput, TupleTagList.empty()))).get(mainOutput).setCoder(VarIntCoder.of());
    StatefulParDoEvaluatorFactory<String, Integer, Integer> factory = new StatefulParDoEvaluatorFactory(mockEvaluationContext);
    AppliedPTransform<PCollection<? extends KeyedWorkItem<String, KV<String, Integer>>>, PCollectionTuple, StatefulParDo<String, Integer, Integer>> producingTransform = (AppliedPTransform) DirectGraphs.getProducer(produced);
    // Then there will be a digging down to the step context to get the state internals
    when(mockEvaluationContext.getExecutionContext(eq(producingTransform), Mockito.<StructuralKey>any())).thenReturn(mockExecutionContext);
    when(mockExecutionContext.getStepContext(anyString())).thenReturn(mockStepContext);
    IntervalWindow firstWindow = new IntervalWindow(new Instant(0), new Instant(9));
    IntervalWindow secondWindow = new IntervalWindow(new Instant(10), new Instant(19));
    StateNamespace firstWindowNamespace = StateNamespaces.window(IntervalWindow.getCoder(), firstWindow);
    StateNamespace secondWindowNamespace = StateNamespaces.window(IntervalWindow.getCoder(), secondWindow);
    StateTag<ValueState<String>> tag = StateTags.tagForSpec(stateId, StateSpecs.value(StringUtf8Coder.of()));
    // Set up non-empty state. We don't mock + verify calls to clear() but instead
    // check that state is actually empty. We musn't care how it is accomplished.
    stateInternals.state(firstWindowNamespace, tag).write("first");
    stateInternals.state(secondWindowNamespace, tag).write("second");
    // A single bundle with some elements in the global window; it should register cleanup for the
    // global window state merely by having the evaluator created. The cleanup logic does not
    // depend on the window.
    CommittedBundle<KV<String, Integer>> inputBundle = BUNDLE_FACTORY.createBundle(input).add(WindowedValue.of(KV.of("hello", 1), new Instant(3), firstWindow, PaneInfo.NO_FIRING)).add(WindowedValue.of(KV.of("hello", 2), new Instant(11), secondWindow, PaneInfo.NO_FIRING)).commit(Instant.now());
    // Merely creating the evaluator should suffice to register the cleanup callback
    factory.forApplication(producingTransform, inputBundle);
    ArgumentCaptor<Runnable> argumentCaptor = ArgumentCaptor.forClass(Runnable.class);
    verify(mockEvaluationContext).scheduleAfterWindowExpiration(eq(producingTransform), eq(firstWindow), Mockito.<WindowingStrategy<?, ?>>any(), argumentCaptor.capture());
    // Should actually clear the state for the first window
    argumentCaptor.getValue().run();
    assertThat(stateInternals.state(firstWindowNamespace, tag).read(), nullValue());
    assertThat(stateInternals.state(secondWindowNamespace, tag).read(), equalTo("second"));
    verify(mockEvaluationContext).scheduleAfterWindowExpiration(eq(producingTransform), eq(secondWindow), Mockito.<WindowingStrategy<?, ?>>any(), argumentCaptor.capture());
    // Should actually clear the state for the second window
    argumentCaptor.getValue().run();
    assertThat(stateInternals.state(secondWindowNamespace, tag).read(), nullValue());
}

Example 53 with KV

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

the class StatefulParDoEvaluatorFactoryTest method testUnprocessedElements.

/**
   * A test that explicitly delays a side input so that the main input will have to be reprocessed,
   * testing that {@code finishBundle()} re-assembles the GBK outputs correctly.
   */
@Test
public void testUnprocessedElements() throws Exception {
    // To test the factory, first we set up a pipeline and then we use the constructed
    // pipeline to create the right parameters to pass to the factory
    final String stateId = "my-state-id";
    // For consistency, window it into FixedWindows. Actually we will fabricate an input bundle.
    PCollection<KV<String, Integer>> mainInput = pipeline.apply(Create.of(KV.of("hello", 1), KV.of("hello", 2))).apply(Window.<KV<String, Integer>>into(FixedWindows.of(Duration.millis(10))));
    final PCollectionView<List<Integer>> sideInput = pipeline.apply("Create side input", Create.of(42)).apply("Window side input", Window.<Integer>into(FixedWindows.of(Duration.millis(10)))).apply("View side input", View.<Integer>asList());
    TupleTag<Integer> mainOutput = new TupleTag<>();
    PCollection<Integer> produced = mainInput.apply(new ParDoMultiOverrideFactory.GbkThenStatefulParDo<>(ParDo.of(new DoFn<KV<String, Integer>, Integer>() {

        @StateId(stateId)
        private final StateSpec<ValueState<String>> spec = StateSpecs.value(StringUtf8Coder.of());

        @ProcessElement
        public void process(ProcessContext c) {
        }
    }).withSideInputs(sideInput).withOutputTags(mainOutput, TupleTagList.empty()))).get(mainOutput).setCoder(VarIntCoder.of());
    StatefulParDoEvaluatorFactory<String, Integer, Integer> factory = new StatefulParDoEvaluatorFactory(mockEvaluationContext);
    // This will be the stateful ParDo from the expansion
    AppliedPTransform<PCollection<KeyedWorkItem<String, KV<String, Integer>>>, PCollectionTuple, StatefulParDo<String, Integer, Integer>> producingTransform = (AppliedPTransform) DirectGraphs.getProducer(produced);
    // Then there will be a digging down to the step context to get the state internals
    when(mockEvaluationContext.getExecutionContext(eq(producingTransform), Mockito.<StructuralKey>any())).thenReturn(mockExecutionContext);
    when(mockExecutionContext.getStepContext(anyString())).thenReturn(mockStepContext);
    when(mockEvaluationContext.createBundle(Matchers.<PCollection<Integer>>any())).thenReturn(mockUncommittedBundle);
    when(mockStepContext.getTimerUpdate()).thenReturn(TimerUpdate.empty());
    // And digging to check whether the window is ready
    when(mockEvaluationContext.createSideInputReader(anyList())).thenReturn(mockSideInputReader);
    when(mockSideInputReader.isReady(Matchers.<PCollectionView<?>>any(), Matchers.<BoundedWindow>any())).thenReturn(false);
    IntervalWindow firstWindow = new IntervalWindow(new Instant(0), new Instant(9));
    // A single bundle with some elements in the global window; it should register cleanup for the
    // global window state merely by having the evaluator created. The cleanup logic does not
    // depend on the window.
    String key = "hello";
    WindowedValue<KV<String, Integer>> firstKv = WindowedValue.of(KV.of(key, 1), new Instant(3), firstWindow, PaneInfo.NO_FIRING);
    WindowedValue<KeyedWorkItem<String, KV<String, Integer>>> gbkOutputElement = firstKv.withValue(KeyedWorkItems.elementsWorkItem("hello", ImmutableList.of(firstKv, firstKv.withValue(KV.of(key, 13)), firstKv.withValue(KV.of(key, 15)))));
    CommittedBundle<KeyedWorkItem<String, KV<String, Integer>>> inputBundle = BUNDLE_FACTORY.createBundle((PCollection<KeyedWorkItem<String, KV<String, Integer>>>) Iterables.getOnlyElement(producingTransform.getInputs().values())).add(gbkOutputElement).commit(Instant.now());
    TransformEvaluator<KeyedWorkItem<String, KV<String, Integer>>> evaluator = factory.forApplication(producingTransform, inputBundle);
    evaluator.processElement(gbkOutputElement);
    // This should push back every element as a KV<String, Iterable<Integer>>
    // in the appropriate window. Since the keys are equal they are single-threaded
    TransformResult<KeyedWorkItem<String, KV<String, Integer>>> result = evaluator.finishBundle();
    List<Integer> pushedBackInts = new ArrayList<>();
    for (WindowedValue<? extends KeyedWorkItem<String, KV<String, Integer>>> unprocessedElement : result.getUnprocessedElements()) {
        assertThat(Iterables.getOnlyElement(unprocessedElement.getWindows()), equalTo((BoundedWindow) firstWindow));
        assertThat(unprocessedElement.getValue().key(), equalTo("hello"));
        for (WindowedValue<KV<String, Integer>> windowedKv : unprocessedElement.getValue().elementsIterable()) {
            pushedBackInts.add(windowedKv.getValue().getValue());
        }
    }
    assertThat(pushedBackInts, containsInAnyOrder(1, 13, 15));
}

Example 54 with KV

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

the class SideInputContainerTest method getAfterWriteReturnsPaneInWindow.

@Test
public void getAfterWriteReturnsPaneInWindow() throws Exception {
    WindowedValue<KV<String, Integer>> one = WindowedValue.of(KV.of("one", 1), new Instant(1L), FIRST_WINDOW, PaneInfo.ON_TIME_AND_ONLY_FIRING);
    WindowedValue<KV<String, Integer>> two = WindowedValue.of(KV.of("two", 2), new Instant(20L), FIRST_WINDOW, PaneInfo.ON_TIME_AND_ONLY_FIRING);
    container.write(mapView, ImmutableList.<WindowedValue<?>>of(one, two));
    Map<String, Integer> viewContents = container.createReaderForViews(ImmutableList.<PCollectionView<?>>of(mapView)).get(mapView, FIRST_WINDOW);
    assertThat(viewContents, hasEntry("one", 1));
    assertThat(viewContents, hasEntry("two", 2));
    assertThat(viewContents.size(), is(2));
}

Example 55 with KV

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

the class SideInputContainerTest method finishDoesNotOverwriteWrittenElements.

@Test
public void finishDoesNotOverwriteWrittenElements() throws Exception {
    WindowedValue<KV<String, Integer>> one = WindowedValue.of(KV.of("one", 1), new Instant(1L), SECOND_WINDOW, PaneInfo.createPane(true, false, Timing.EARLY));
    WindowedValue<KV<String, Integer>> two = WindowedValue.of(KV.of("two", 2), new Instant(20L), SECOND_WINDOW, PaneInfo.createPane(true, false, Timing.EARLY));
    container.write(mapView, ImmutableList.<WindowedValue<?>>of(one, two));
    immediatelyInvokeCallback(mapView, SECOND_WINDOW);
    Map<String, Integer> viewContents = container.createReaderForViews(ImmutableList.<PCollectionView<?>>of(mapView)).get(mapView, SECOND_WINDOW);
    assertThat(viewContents, hasEntry("one", 1));
    assertThat(viewContents, hasEntry("two", 2));
    assertThat(viewContents.size(), is(2));
}