Search in sources :

Example 1 with DEFAULT_WIRE_CODER_SETTINGS

use of org.apache.beam.runners.core.construction.graph.ExecutableStage.DEFAULT_WIRE_CODER_SETTINGS in project beam by apache.

the class OutputDeduplicatorTest method multipleDuplicatesInStages.

@Test
public void multipleDuplicatesInStages() {
    /* A stage that produces multiple duplicates should have them all synthesized.
     *
     * Original Pipeline:
     * red -> .out ---> one -> .out -----\
     *             \                      -> shared.out
     *              \--> two -> .out ----|
     *               \                    -> otherShared -> .out
     *                \-> three --> .out /
     *
     * Fused Pipeline:
     *      -> .out [-> one -> .out -> shared -> .out] \
     *     /                                            -> blue -> .out
     *     |                        -> shared -> .out] /
     * red -> .out [-> two -> .out |
     *     |                        -> otherShared -> .out]
     *     \
     *      -> .out [-> three -> .out -> otherShared -> .out]
     *
     * Deduplicated Pipeline:
     *           [-> one -> .out -> shared -> .out:0] --\
     *           |                                       -> shared -> .out -> blue -> .out
     *           |                 -> shared -> .out:1] /
     * red -> .out [-> two -> .out |
     *           |                  -> otherShared -> .out:0] --\
     *           |                                               -> otherShared -> .out
     *           [-> three -> .out -> otherShared -> .out:1] ---/
     */
    PCollection redOut = PCollection.newBuilder().setUniqueName("red.out").build();
    PTransform red = PTransform.newBuilder().setSpec(FunctionSpec.newBuilder().setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).build()).putOutputs("out", redOut.getUniqueName()).build();
    PCollection threeOut = PCollection.newBuilder().setUniqueName("three.out").build();
    PTransform three = PTransform.newBuilder().setSpec(FunctionSpec.newBuilder().setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).build()).putInputs("in", redOut.getUniqueName()).putOutputs("out", threeOut.getUniqueName()).build();
    PCollection oneOut = PCollection.newBuilder().setUniqueName("one.out").build();
    PTransform one = PTransform.newBuilder().setSpec(FunctionSpec.newBuilder().setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).build()).putInputs("in", redOut.getUniqueName()).putOutputs("out", oneOut.getUniqueName()).build();
    PCollection twoOut = PCollection.newBuilder().setUniqueName("two.out").build();
    PTransform two = PTransform.newBuilder().setSpec(FunctionSpec.newBuilder().setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).build()).putInputs("in", redOut.getUniqueName()).putOutputs("out", twoOut.getUniqueName()).build();
    PCollection sharedOut = PCollection.newBuilder().setUniqueName("shared.out").build();
    PTransform shared = PTransform.newBuilder().setSpec(FunctionSpec.newBuilder().setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).build()).putInputs("one", oneOut.getUniqueName()).putInputs("two", twoOut.getUniqueName()).putOutputs("shared", sharedOut.getUniqueName()).build();
    PCollection otherSharedOut = PCollection.newBuilder().setUniqueName("shared.out2").build();
    PTransform otherShared = PTransform.newBuilder().setSpec(FunctionSpec.newBuilder().setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).build()).putInputs("multi", threeOut.getUniqueName()).putInputs("two", twoOut.getUniqueName()).putOutputs("out", otherSharedOut.getUniqueName()).build();
    PCollection blueOut = PCollection.newBuilder().setUniqueName("blue.out").build();
    PTransform blue = PTransform.newBuilder().setSpec(FunctionSpec.newBuilder().setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).build()).putInputs("in", sharedOut.getUniqueName()).putOutputs("out", blueOut.getUniqueName()).build();
    RunnerApi.Components components = Components.newBuilder().putTransforms("one", one).putPcollections(oneOut.getUniqueName(), oneOut).putTransforms("two", two).putPcollections(twoOut.getUniqueName(), twoOut).putTransforms("multi", three).putPcollections(threeOut.getUniqueName(), threeOut).putTransforms("shared", shared).putPcollections(sharedOut.getUniqueName(), sharedOut).putTransforms("otherShared", otherShared).putPcollections(otherSharedOut.getUniqueName(), otherSharedOut).putTransforms("red", red).putPcollections(redOut.getUniqueName(), redOut).putTransforms("blue", blue).putPcollections(blueOut.getUniqueName(), blueOut).build();
    ExecutableStage multiStage = ImmutableExecutableStage.of(components, Environment.getDefaultInstance(), PipelineNode.pCollection(redOut.getUniqueName(), redOut), ImmutableList.of(), ImmutableList.of(), ImmutableList.of(), ImmutableList.of(PipelineNode.pTransform("multi", three), PipelineNode.pTransform("shared", shared), PipelineNode.pTransform("otherShared", otherShared)), ImmutableList.of(PipelineNode.pCollection(sharedOut.getUniqueName(), sharedOut), PipelineNode.pCollection(otherSharedOut.getUniqueName(), otherSharedOut)), DEFAULT_WIRE_CODER_SETTINGS);
    ExecutableStage oneStage = ImmutableExecutableStage.of(components, Environment.getDefaultInstance(), PipelineNode.pCollection(redOut.getUniqueName(), redOut), ImmutableList.of(), ImmutableList.of(), ImmutableList.of(), ImmutableList.of(PipelineNode.pTransform("one", one), PipelineNode.pTransform("shared", shared)), ImmutableList.of(PipelineNode.pCollection(sharedOut.getUniqueName(), sharedOut)), DEFAULT_WIRE_CODER_SETTINGS);
    ExecutableStage twoStage = ImmutableExecutableStage.of(components, Environment.getDefaultInstance(), PipelineNode.pCollection(redOut.getUniqueName(), redOut), ImmutableList.of(), ImmutableList.of(), ImmutableList.of(), ImmutableList.of(PipelineNode.pTransform("two", two), PipelineNode.pTransform("otherShared", otherShared)), ImmutableList.of(PipelineNode.pCollection(otherSharedOut.getUniqueName(), otherSharedOut)), DEFAULT_WIRE_CODER_SETTINGS);
    PTransformNode redTransform = PipelineNode.pTransform("red", red);
    PTransformNode blueTransform = PipelineNode.pTransform("blue", blue);
    QueryablePipeline pipeline = QueryablePipeline.forPrimitivesIn(components);
    DeduplicationResult result = OutputDeduplicator.ensureSingleProducer(pipeline, ImmutableList.of(oneStage, twoStage, multiStage), ImmutableList.of(redTransform, blueTransform));
    assertThat(result.getIntroducedTransforms(), hasSize(2));
    assertThat(result.getDeduplicatedStages().keySet(), containsInAnyOrder(multiStage, oneStage, twoStage));
    assertThat(result.getDeduplicatedTransforms().keySet(), empty());
    Collection<String> introducedIds = result.getIntroducedTransforms().stream().flatMap(pt -> pt.getTransform().getInputsMap().values().stream()).collect(Collectors.toList());
    String[] stageOutputs = result.getDeduplicatedStages().values().stream().flatMap(s -> s.getOutputPCollections().stream().map(PCollectionNode::getId)).toArray(String[]::new);
    assertThat(introducedIds, containsInAnyOrder(stageOutputs));
    assertThat(result.getDeduplicatedComponents().getPcollectionsMap().keySet(), hasItems(introducedIds.toArray(new String[0])));
    assertThat(result.getDeduplicatedComponents().getTransformsMap().entrySet(), hasItems(result.getIntroducedTransforms().stream().collect(Collectors.toMap(PTransformNode::getId, PTransformNode::getTransform)).entrySet().toArray(new Map.Entry[0])));
}
Also used : PTransform(org.apache.beam.model.pipeline.v1.RunnerApi.PTransform) RunWith(org.junit.runner.RunWith) Matchers.hasItems(org.hamcrest.Matchers.hasItems) ArrayList(java.util.ArrayList) FunctionSpec(org.apache.beam.model.pipeline.v1.RunnerApi.FunctionSpec) PCollection(org.apache.beam.model.pipeline.v1.RunnerApi.PCollection) Map(java.util.Map) Matchers.hasSize(org.hamcrest.Matchers.hasSize) Components(org.apache.beam.model.pipeline.v1.RunnerApi.Components) MatcherAssert.assertThat(org.hamcrest.MatcherAssert.assertThat) PTransformNode(org.apache.beam.runners.core.construction.graph.PipelineNode.PTransformNode) Matchers.hasEntry(org.hamcrest.Matchers.hasEntry) RunnerApi(org.apache.beam.model.pipeline.v1.RunnerApi) Matchers.empty(org.hamcrest.Matchers.empty) PTransformTranslation(org.apache.beam.runners.core.construction.PTransformTranslation) Collection(java.util.Collection) DeduplicationResult(org.apache.beam.runners.core.construction.graph.OutputDeduplicator.DeduplicationResult) Test(org.junit.Test) JUnit4(org.junit.runners.JUnit4) Collectors(java.util.stream.Collectors) List(java.util.List) Matchers.containsInAnyOrder(org.hamcrest.Matchers.containsInAnyOrder) Matchers.equalTo(org.hamcrest.Matchers.equalTo) Environment(org.apache.beam.model.pipeline.v1.RunnerApi.Environment) ImmutableList(org.apache.beam.vendor.guava.v26_0_jre.com.google.common.collect.ImmutableList) Iterables.getOnlyElement(org.apache.beam.vendor.guava.v26_0_jre.com.google.common.collect.Iterables.getOnlyElement) PCollectionNode(org.apache.beam.runners.core.construction.graph.PipelineNode.PCollectionNode) DEFAULT_WIRE_CODER_SETTINGS(org.apache.beam.runners.core.construction.graph.ExecutableStage.DEFAULT_WIRE_CODER_SETTINGS) PTransformNode(org.apache.beam.runners.core.construction.graph.PipelineNode.PTransformNode) PCollectionNode(org.apache.beam.runners.core.construction.graph.PipelineNode.PCollectionNode) PCollection(org.apache.beam.model.pipeline.v1.RunnerApi.PCollection) RunnerApi(org.apache.beam.model.pipeline.v1.RunnerApi) DeduplicationResult(org.apache.beam.runners.core.construction.graph.OutputDeduplicator.DeduplicationResult) Components(org.apache.beam.model.pipeline.v1.RunnerApi.Components) Map(java.util.Map) PTransform(org.apache.beam.model.pipeline.v1.RunnerApi.PTransform) Test(org.junit.Test)

Example 2 with DEFAULT_WIRE_CODER_SETTINGS

use of org.apache.beam.runners.core.construction.graph.ExecutableStage.DEFAULT_WIRE_CODER_SETTINGS in project beam by apache.

the class OutputDeduplicatorTest method duplicateOverStages.

@Test
public void duplicateOverStages() {
    /* When multiple stages and a runner-executed transform produce a PCollection, all should be
     * replaced with synthetic flattens.
     * original graph:
     *             --> one -> .out \
     * red -> .out |                -> shared -> .out -> blue -> .out
     *             --> two -> .out /
     *
     * fused graph:
     *             --> [one -> .out -> shared ->] .out
     * red -> .out |                                   (shared.out) -> blue -> .out
     *             --> [two -> .out -> shared ->] .out
     *
     * deduplicated graph:
     *             --> [one -> .out -> shared ->] .out:0 \
     * red -> .out |                                      -> shared -> .out -> blue ->.out
     *             --> [two -> .out -> shared ->] .out:1 /
     */
    PCollection redOut = PCollection.newBuilder().setUniqueName("red.out").build();
    PTransform red = PTransform.newBuilder().setSpec(FunctionSpec.newBuilder().setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).build()).putOutputs("out", redOut.getUniqueName()).build();
    PCollection oneOut = PCollection.newBuilder().setUniqueName("one.out").build();
    PTransform one = PTransform.newBuilder().setSpec(FunctionSpec.newBuilder().setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).build()).putInputs("in", redOut.getUniqueName()).putOutputs("out", oneOut.getUniqueName()).build();
    PCollection twoOut = PCollection.newBuilder().setUniqueName("two.out").build();
    PTransform two = PTransform.newBuilder().setSpec(FunctionSpec.newBuilder().setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).build()).putInputs("in", redOut.getUniqueName()).putOutputs("out", twoOut.getUniqueName()).build();
    PCollection sharedOut = PCollection.newBuilder().setUniqueName("shared.out").build();
    PTransform shared = PTransform.newBuilder().setSpec(FunctionSpec.newBuilder().setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).build()).putInputs("one", oneOut.getUniqueName()).putInputs("two", twoOut.getUniqueName()).putOutputs("shared", sharedOut.getUniqueName()).build();
    PCollection blueOut = PCollection.newBuilder().setUniqueName("blue.out").build();
    PTransform blue = PTransform.newBuilder().setSpec(FunctionSpec.newBuilder().setUrn(PTransformTranslation.PAR_DO_TRANSFORM_URN).build()).putInputs("in", sharedOut.getUniqueName()).putOutputs("out", blueOut.getUniqueName()).build();
    RunnerApi.Components components = Components.newBuilder().putTransforms("one", one).putPcollections(oneOut.getUniqueName(), oneOut).putTransforms("two", two).putPcollections(twoOut.getUniqueName(), twoOut).putTransforms("shared", shared).putPcollections(sharedOut.getUniqueName(), sharedOut).putTransforms("red", red).putPcollections(redOut.getUniqueName(), redOut).putTransforms("blue", blue).putPcollections(blueOut.getUniqueName(), blueOut).build();
    ExecutableStage oneStage = ImmutableExecutableStage.of(components, Environment.getDefaultInstance(), PipelineNode.pCollection(redOut.getUniqueName(), redOut), ImmutableList.of(), ImmutableList.of(), ImmutableList.of(), ImmutableList.of(PipelineNode.pTransform("one", one), PipelineNode.pTransform("shared", shared)), ImmutableList.of(PipelineNode.pCollection(sharedOut.getUniqueName(), sharedOut)), DEFAULT_WIRE_CODER_SETTINGS);
    ExecutableStage twoStage = ImmutableExecutableStage.of(components, Environment.getDefaultInstance(), PipelineNode.pCollection(redOut.getUniqueName(), redOut), ImmutableList.of(), ImmutableList.of(), ImmutableList.of(), ImmutableList.of(PipelineNode.pTransform("two", two), PipelineNode.pTransform("shared", shared)), ImmutableList.of(PipelineNode.pCollection(sharedOut.getUniqueName(), sharedOut)), DEFAULT_WIRE_CODER_SETTINGS);
    PTransformNode redTransform = PipelineNode.pTransform("red", red);
    PTransformNode blueTransform = PipelineNode.pTransform("blue", blue);
    QueryablePipeline pipeline = QueryablePipeline.forPrimitivesIn(components);
    DeduplicationResult result = OutputDeduplicator.ensureSingleProducer(pipeline, ImmutableList.of(oneStage, twoStage), ImmutableList.of(redTransform, blueTransform));
    assertThat(result.getIntroducedTransforms(), hasSize(1));
    PTransformNode introduced = getOnlyElement(result.getIntroducedTransforms());
    assertThat(introduced.getTransform().getOutputsMap().size(), equalTo(1));
    assertThat(getOnlyElement(introduced.getTransform().getOutputsMap().values()), equalTo(sharedOut.getUniqueName()));
    assertThat(result.getDeduplicatedComponents().getPcollectionsMap().keySet(), hasItems(introduced.getTransform().getInputsMap().values().toArray(new String[0])));
    assertThat(result.getDeduplicatedStages().keySet(), hasSize(2));
    List<String> stageOutputs = result.getDeduplicatedStages().values().stream().flatMap(stage -> stage.getOutputPCollections().stream().map(PCollectionNode::getId)).collect(Collectors.toList());
    assertThat(stageOutputs, containsInAnyOrder(introduced.getTransform().getInputsMap().values().toArray()));
    assertThat(result.getDeduplicatedTransforms().keySet(), empty());
    assertThat(result.getDeduplicatedComponents().getPcollectionsMap().keySet(), hasItems(stageOutputs.toArray(new String[0])));
    assertThat(result.getDeduplicatedComponents().getTransformsMap(), hasEntry(introduced.getId(), introduced.getTransform()));
}
Also used : PTransform(org.apache.beam.model.pipeline.v1.RunnerApi.PTransform) RunWith(org.junit.runner.RunWith) Matchers.hasItems(org.hamcrest.Matchers.hasItems) ArrayList(java.util.ArrayList) FunctionSpec(org.apache.beam.model.pipeline.v1.RunnerApi.FunctionSpec) PCollection(org.apache.beam.model.pipeline.v1.RunnerApi.PCollection) Map(java.util.Map) Matchers.hasSize(org.hamcrest.Matchers.hasSize) Components(org.apache.beam.model.pipeline.v1.RunnerApi.Components) MatcherAssert.assertThat(org.hamcrest.MatcherAssert.assertThat) PTransformNode(org.apache.beam.runners.core.construction.graph.PipelineNode.PTransformNode) Matchers.hasEntry(org.hamcrest.Matchers.hasEntry) RunnerApi(org.apache.beam.model.pipeline.v1.RunnerApi) Matchers.empty(org.hamcrest.Matchers.empty) PTransformTranslation(org.apache.beam.runners.core.construction.PTransformTranslation) Collection(java.util.Collection) DeduplicationResult(org.apache.beam.runners.core.construction.graph.OutputDeduplicator.DeduplicationResult) Test(org.junit.Test) JUnit4(org.junit.runners.JUnit4) Collectors(java.util.stream.Collectors) List(java.util.List) Matchers.containsInAnyOrder(org.hamcrest.Matchers.containsInAnyOrder) Matchers.equalTo(org.hamcrest.Matchers.equalTo) Environment(org.apache.beam.model.pipeline.v1.RunnerApi.Environment) ImmutableList(org.apache.beam.vendor.guava.v26_0_jre.com.google.common.collect.ImmutableList) Iterables.getOnlyElement(org.apache.beam.vendor.guava.v26_0_jre.com.google.common.collect.Iterables.getOnlyElement) PCollectionNode(org.apache.beam.runners.core.construction.graph.PipelineNode.PCollectionNode) DEFAULT_WIRE_CODER_SETTINGS(org.apache.beam.runners.core.construction.graph.ExecutableStage.DEFAULT_WIRE_CODER_SETTINGS) PCollection(org.apache.beam.model.pipeline.v1.RunnerApi.PCollection) RunnerApi(org.apache.beam.model.pipeline.v1.RunnerApi) DeduplicationResult(org.apache.beam.runners.core.construction.graph.OutputDeduplicator.DeduplicationResult) Components(org.apache.beam.model.pipeline.v1.RunnerApi.Components) PTransformNode(org.apache.beam.runners.core.construction.graph.PipelineNode.PTransformNode) PCollectionNode(org.apache.beam.runners.core.construction.graph.PipelineNode.PCollectionNode) PTransform(org.apache.beam.model.pipeline.v1.RunnerApi.PTransform) Test(org.junit.Test)

Aggregations

ArrayList (java.util.ArrayList)2 Collection (java.util.Collection)2 List (java.util.List)2 Map (java.util.Map)2 Collectors (java.util.stream.Collectors)2 RunnerApi (org.apache.beam.model.pipeline.v1.RunnerApi)2 Components (org.apache.beam.model.pipeline.v1.RunnerApi.Components)2 Environment (org.apache.beam.model.pipeline.v1.RunnerApi.Environment)2 FunctionSpec (org.apache.beam.model.pipeline.v1.RunnerApi.FunctionSpec)2 PCollection (org.apache.beam.model.pipeline.v1.RunnerApi.PCollection)2 PTransform (org.apache.beam.model.pipeline.v1.RunnerApi.PTransform)2 PTransformTranslation (org.apache.beam.runners.core.construction.PTransformTranslation)2 DEFAULT_WIRE_CODER_SETTINGS (org.apache.beam.runners.core.construction.graph.ExecutableStage.DEFAULT_WIRE_CODER_SETTINGS)2 DeduplicationResult (org.apache.beam.runners.core.construction.graph.OutputDeduplicator.DeduplicationResult)2 PCollectionNode (org.apache.beam.runners.core.construction.graph.PipelineNode.PCollectionNode)2 PTransformNode (org.apache.beam.runners.core.construction.graph.PipelineNode.PTransformNode)2 ImmutableList (org.apache.beam.vendor.guava.v26_0_jre.com.google.common.collect.ImmutableList)2 Iterables.getOnlyElement (org.apache.beam.vendor.guava.v26_0_jre.com.google.common.collect.Iterables.getOnlyElement)2 MatcherAssert.assertThat (org.hamcrest.MatcherAssert.assertThat)2 Matchers.containsInAnyOrder (org.hamcrest.Matchers.containsInAnyOrder)2