Examples with DoFn org.apache.beam.sdk.transforms.DoFn used on opensource projects

Example 1 with DoFn

use of org.apache.beam.sdk.transforms.DoFn 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 2 with DoFn

use of org.apache.beam.sdk.transforms.DoFn in project beam by apache.

the class BigQueryIOTest method testValidateReadSetsDefaultProject.

@Test
public void testValidateReadSetsDefaultProject() throws Exception {
    String projectId = "someproject";
    String datasetId = "somedataset";
    String tableId = "sometable";
    BigQueryOptions bqOptions = TestPipeline.testingPipelineOptions().as(BigQueryOptions.class);
    bqOptions.setProject(projectId);
    Path baseDir = Files.createTempDirectory(tempFolder, "testValidateReadSetsDefaultProject");
    bqOptions.setTempLocation(baseDir.toString());
    FakeDatasetService fakeDatasetService = new FakeDatasetService();
    fakeDatasetService.createDataset(projectId, datasetId, "", "");
    TableReference tableReference = new TableReference().setProjectId(projectId).setDatasetId(datasetId).setTableId(tableId);
    fakeDatasetService.createTable(new Table().setTableReference(tableReference).setSchema(new TableSchema().setFields(ImmutableList.of(new TableFieldSchema().setName("name").setType("STRING"), new TableFieldSchema().setName("number").setType("INTEGER")))));
    FakeBigQueryServices fakeBqServices = new FakeBigQueryServices().withJobService(new FakeJobService()).withDatasetService(fakeDatasetService);
    List<TableRow> expected = ImmutableList.of(new TableRow().set("name", "a").set("number", 1L), new TableRow().set("name", "b").set("number", 2L), new TableRow().set("name", "c").set("number", 3L), new TableRow().set("name", "d").set("number", 4L), new TableRow().set("name", "e").set("number", 5L), new TableRow().set("name", "f").set("number", 6L));
    fakeDatasetService.insertAll(tableReference, expected, null);
    Pipeline p = TestPipeline.create(bqOptions);
    TableReference tableRef = new TableReference();
    tableRef.setDatasetId(datasetId);
    tableRef.setTableId(tableId);
    PCollection<KV<String, Long>> output = p.apply(BigQueryIO.read().from(tableRef).withTestServices(fakeBqServices)).apply(ParDo.of(new DoFn<TableRow, KV<String, Long>>() {

        @ProcessElement
        public void processElement(ProcessContext c) throws Exception {
            c.output(KV.of((String) c.element().get("name"), Long.valueOf((String) c.element().get("number"))));
        }
    }));
    PAssert.that(output).containsInAnyOrder(ImmutableList.of(KV.of("a", 1L), KV.of("b", 2L), KV.of("c", 3L), KV.of("d", 4L), KV.of("e", 5L), KV.of("f", 6L)));
    p.run();
}

Example 3 with DoFn

use of org.apache.beam.sdk.transforms.DoFn in project beam by apache.

the class DoFnOperatorTest method testLateDroppingForStatefulFn.

@Test
public void testLateDroppingForStatefulFn() throws Exception {
    WindowingStrategy<Object, IntervalWindow> windowingStrategy = WindowingStrategy.of(FixedWindows.of(new Duration(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());
        }
    };
    WindowedValue.FullWindowedValueCoder<Integer> windowedValueCoder = WindowedValue.getFullCoder(VarIntCoder.of(), windowingStrategy.getWindowFn().windowCoder());
    TupleTag<String> outputTag = new TupleTag<>("main-output");
    DoFnOperator<Integer, String, WindowedValue<String>> doFnOperator = new DoFnOperator<>(fn, "stepName", windowedValueCoder, outputTag, Collections.<TupleTag<?>>emptyList(), new DoFnOperator.DefaultOutputManagerFactory<WindowedValue<String>>(), windowingStrategy, new HashMap<Integer, PCollectionView<?>>(), /* side-input mapping */
    Collections.<PCollectionView<?>>emptyList(), /* side inputs */
    PipelineOptionsFactory.as(FlinkPipelineOptions.class), VarIntCoder.of());
    OneInputStreamOperatorTestHarness<WindowedValue<Integer>, WindowedValue<String>> testHarness = new KeyedOneInputStreamOperatorTestHarness<>(doFnOperator, new KeySelector<WindowedValue<Integer>, Integer>() {

        @Override
        public Integer getKey(WindowedValue<Integer> integerWindowedValue) throws Exception {
            return integerWindowedValue.getValue();
        }
    }, new CoderTypeInformation<>(VarIntCoder.of()));
    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(this.<String>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(this.<String>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(this.<String>stripStreamRecordFromWindowedValue(testHarness.getOutput()), emptyIterable());
    testHarness.close();
}

Example 4 with DoFn

use of org.apache.beam.sdk.transforms.DoFn in project beam by apache.

the class DoFnOperatorTest method testStateGCForStatefulFn.

@Test
public void testStateGCForStatefulFn() throws Exception {
    WindowingStrategy<Object, IntervalWindow> windowingStrategy = WindowingStrategy.of(FixedWindows.of(new Duration(10))).withAllowedLateness(Duration.ZERO);
    final String timerId = "boo";
    final String stateId = "dazzle";
    final int offset = 5000;
    final int timerOutput = 4093;
    DoFn<KV<String, Integer>, KV<String, Integer>> fn = new DoFn<KV<String, Integer>, KV<String, Integer>>() {

        @TimerId(timerId)
        private final TimerSpec spec = TimerSpecs.timer(TimeDomain.EVENT_TIME);

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

        @ProcessElement
        public void processElement(ProcessContext context, @TimerId(timerId) Timer timer, @StateId(stateId) ValueState<String> state, BoundedWindow window) {
            timer.set(window.maxTimestamp());
            state.write(context.element().getKey());
            context.output(KV.of(context.element().getKey(), context.element().getValue() + offset));
        }

        @OnTimer(timerId)
        public void onTimer(OnTimerContext context, @StateId(stateId) ValueState<String> state) {
            context.output(KV.of(state.read(), timerOutput));
        }
    };
    WindowedValue.FullWindowedValueCoder<KV<String, Integer>> windowedValueCoder = WindowedValue.getFullCoder(KvCoder.of(StringUtf8Coder.of(), VarIntCoder.of()), windowingStrategy.getWindowFn().windowCoder());
    TupleTag<KV<String, Integer>> outputTag = new TupleTag<>("main-output");
    DoFnOperator<KV<String, Integer>, KV<String, Integer>, WindowedValue<KV<String, Integer>>> doFnOperator = new DoFnOperator<>(fn, "stepName", windowedValueCoder, outputTag, Collections.<TupleTag<?>>emptyList(), new DoFnOperator.DefaultOutputManagerFactory<WindowedValue<KV<String, Integer>>>(), windowingStrategy, new HashMap<Integer, PCollectionView<?>>(), /* side-input mapping */
    Collections.<PCollectionView<?>>emptyList(), /* side inputs */
    PipelineOptionsFactory.as(FlinkPipelineOptions.class), StringUtf8Coder.of());
    KeyedOneInputStreamOperatorTestHarness<String, WindowedValue<KV<String, Integer>>, WindowedValue<KV<String, Integer>>> testHarness = new KeyedOneInputStreamOperatorTestHarness<>(doFnOperator, new KeySelector<WindowedValue<KV<String, Integer>>, String>() {

        @Override
        public String getKey(WindowedValue<KV<String, Integer>> kvWindowedValue) throws Exception {
            return kvWindowedValue.getValue().getKey();
        }
    }, new CoderTypeInformation<>(StringUtf8Coder.of()));
    testHarness.open();
    testHarness.processWatermark(0);
    assertEquals(0, testHarness.numKeyedStateEntries());
    IntervalWindow window1 = new IntervalWindow(new Instant(0), Duration.millis(10));
    testHarness.processElement(new StreamRecord<>(WindowedValue.of(KV.of("key1", 5), new Instant(1), window1, PaneInfo.NO_FIRING)));
    testHarness.processElement(new StreamRecord<>(WindowedValue.of(KV.of("key2", 7), new Instant(3), window1, PaneInfo.NO_FIRING)));
    assertThat(this.<KV<String, Integer>>stripStreamRecordFromWindowedValue(testHarness.getOutput()), contains(WindowedValue.of(KV.of("key1", 5 + offset), new Instant(1), window1, PaneInfo.NO_FIRING), WindowedValue.of(KV.of("key2", 7 + offset), new Instant(3), window1, PaneInfo.NO_FIRING)));
    assertEquals(2, testHarness.numKeyedStateEntries());
    testHarness.getOutput().clear();
    // this should trigger both the window.maxTimestamp() timer and the GC timer
    // this tests that the GC timer fires after the user timer
    testHarness.processWatermark(window1.maxTimestamp().plus(windowingStrategy.getAllowedLateness()).plus(StatefulDoFnRunner.TimeInternalsCleanupTimer.GC_DELAY_MS).getMillis());
    assertThat(this.<KV<String, Integer>>stripStreamRecordFromWindowedValue(testHarness.getOutput()), contains(WindowedValue.of(KV.of("key1", timerOutput), new Instant(9), window1, PaneInfo.NO_FIRING), WindowedValue.of(KV.of("key2", timerOutput), new Instant(9), window1, PaneInfo.NO_FIRING)));
    // ensure the state was garbage collected
    assertEquals(0, testHarness.numKeyedStateEntries());
    testHarness.close();
}

Example 5 with DoFn

use of org.apache.beam.sdk.transforms.DoFn in project beam by apache.

the class ParDoTranslator method translate.

@Override
public void translate(ParDo.MultiOutput<InputT, OutputT> transform, TranslationContext context) {
    DoFn<InputT, OutputT> doFn = transform.getFn();
    DoFnSignature signature = DoFnSignatures.getSignature(doFn.getClass());
    if (signature.processElement().isSplittable()) {
        throw new UnsupportedOperationException(String.format("%s does not support splittable DoFn: %s", ApexRunner.class.getSimpleName(), doFn));
    }
    if (signature.stateDeclarations().size() > 0) {
        throw new UnsupportedOperationException(String.format("Found %s annotations on %s, but %s cannot yet be used with state in the %s.", DoFn.StateId.class.getSimpleName(), doFn.getClass().getName(), DoFn.class.getSimpleName(), ApexRunner.class.getSimpleName()));
    }
    if (signature.timerDeclarations().size() > 0) {
        throw new UnsupportedOperationException(String.format("Found %s annotations on %s, but %s cannot yet be used with timers in the %s.", DoFn.TimerId.class.getSimpleName(), doFn.getClass().getName(), DoFn.class.getSimpleName(), ApexRunner.class.getSimpleName()));
    }
    Map<TupleTag<?>, PValue> outputs = context.getOutputs();
    PCollection<InputT> input = context.getInput();
    List<PCollectionView<?>> sideInputs = transform.getSideInputs();
    Coder<InputT> inputCoder = input.getCoder();
    WindowedValueCoder<InputT> wvInputCoder = FullWindowedValueCoder.of(inputCoder, input.getWindowingStrategy().getWindowFn().windowCoder());
    ApexParDoOperator<InputT, OutputT> operator = new ApexParDoOperator<>(context.getPipelineOptions(), doFn, transform.getMainOutputTag(), transform.getAdditionalOutputTags().getAll(), input.getWindowingStrategy(), sideInputs, wvInputCoder, context.getStateBackend());
    Map<PCollection<?>, OutputPort<?>> ports = Maps.newHashMapWithExpectedSize(outputs.size());
    for (Entry<TupleTag<?>, PValue> output : outputs.entrySet()) {
        checkArgument(output.getValue() instanceof PCollection, "%s %s outputs non-PCollection %s of type %s", ParDo.MultiOutput.class.getSimpleName(), context.getFullName(), output.getValue(), output.getValue().getClass().getSimpleName());
        PCollection<?> pc = (PCollection<?>) output.getValue();
        if (output.getKey().equals(transform.getMainOutputTag())) {
            ports.put(pc, operator.output);
        } else {
            int portIndex = 0;
            for (TupleTag<?> tag : transform.getAdditionalOutputTags().getAll()) {
                if (tag.equals(output.getKey())) {
                    ports.put(pc, operator.additionalOutputPorts[portIndex]);
                    break;
                }
                portIndex++;
            }
        }
    }
    context.addOperator(operator, ports);
    context.addStream(context.getInput(), operator.input);
    if (!sideInputs.isEmpty()) {
        addSideInputs(operator.sideInput1, sideInputs, context);
    }
}