Search in sources :

Example 1 with SourceSplitResponse

use of com.google.api.services.dataflow.model.SourceSplitResponse in project beam by apache.

the class BatchDataflowWorkerTest method testIsSplitResponseTooLarge.

@Test
public void testIsSplitResponseTooLarge() throws IOException {
    SourceSplitResponse splitResponse = new SourceSplitResponse();
    splitResponse.setShards(ImmutableList.<SourceSplitShard>of(new SourceSplitShard(), new SourceSplitShard()));
    assertThat(DataflowApiUtils.computeSerializedSizeBytes(splitResponse), greaterThan(0L));
}
Also used : SourceSplitResponse(com.google.api.services.dataflow.model.SourceSplitResponse) SourceSplitShard(com.google.api.services.dataflow.model.SourceSplitShard) Test(org.junit.Test)

Example 2 with SourceSplitResponse

use of com.google.api.services.dataflow.model.SourceSplitResponse in project beam by apache.

the class WorkerCustomSourcesSplitOnlySourceTest method testAllSplitsAreReturned.

@Test
public void testAllSplitsAreReturned() throws Exception {
    final long apiSizeLimitForTest = 500 * 1024;
    DataflowPipelineOptions options = PipelineOptionsFactory.as(DataflowPipelineOptions.class);
    options.setAppName("TestAppName");
    options.setProject("test-project");
    options.setRegion("some-region1");
    options.setTempLocation("gs://test/temp/location");
    options.setGcpCredential(new TestCredential());
    options.setRunner(DataflowRunner.class);
    options.setPathValidatorClass(NoopPathValidator.class);
    // Generate a CountingSource and split it into the desired number of splits
    // (desired size = 1 byte), triggering the re-split with a larger bundle size.
    // Thus below we expect to produce 'numberOfSplits' splits.
    com.google.api.services.dataflow.model.Source source = WorkerCustomSourcesTest.translateIOToCloudSource(CountingSource.upTo(numberOfSplits), options);
    SourceSplitResponse split = WorkerCustomSourcesTest.performSplit(source, options, 1L, null, /* numBundles limit */
    apiSizeLimitForTest);
    assertThat(split.getBundles().size(), lessThanOrEqualTo(WorkerCustomSources.DEFAULT_NUM_BUNDLES_LIMIT));
    List<OffsetBasedSource<?>> originalSplits = new ArrayList<>(numberOfSplits);
    // Collect all the splits
    for (DerivedSource derivedSource : split.getBundles()) {
        Object deserializedSource = WorkerCustomSources.deserializeFromCloudSource(derivedSource.getSource().getSpec());
        if (deserializedSource instanceof SplittableOnlyBoundedSource) {
            SplittableOnlyBoundedSource<?> splittableOnlySource = (SplittableOnlyBoundedSource<?>) deserializedSource;
            originalSplits.addAll((List) splittableOnlySource.split(1L, options));
        } else {
            originalSplits.add((OffsetBasedSource<?>) deserializedSource);
        }
    }
    assertEquals(numberOfSplits, originalSplits.size());
    for (int i = 0; i < originalSplits.size(); i++) {
        OffsetBasedSource<?> offsetBasedSource = (OffsetBasedSource<?>) originalSplits.get(i);
        assertEquals(i, offsetBasedSource.getStartOffset());
        assertEquals(i + 1, offsetBasedSource.getEndOffset());
    }
}
Also used : DataflowPipelineOptions(org.apache.beam.runners.dataflow.options.DataflowPipelineOptions) ArrayList(java.util.ArrayList) OffsetBasedSource(org.apache.beam.sdk.io.OffsetBasedSource) TestCredential(org.apache.beam.sdk.extensions.gcp.auth.TestCredential) SourceSplitResponse(com.google.api.services.dataflow.model.SourceSplitResponse) DerivedSource(com.google.api.services.dataflow.model.DerivedSource) SplittableOnlyBoundedSource(org.apache.beam.runners.dataflow.worker.WorkerCustomSources.SplittableOnlyBoundedSource) Test(org.junit.Test)

Example 3 with SourceSplitResponse

use of com.google.api.services.dataflow.model.SourceSplitResponse in project beam by apache.

the class WorkerCustomSourcesTest method testLargeNumberOfSplitsReturnsSplittableOnlyBoundedSources.

@Test
public void testLargeNumberOfSplitsReturnsSplittableOnlyBoundedSources() throws Exception {
    final long apiSizeLimitForTest = 500 * 1024;
    // Generate a CountingSource and split it into the desired number of splits
    // (desired size = 1 byte), triggering the re-split with a larger bundle size.
    // Thus below we expect to produce 451 splits.
    com.google.api.services.dataflow.model.Source source = translateIOToCloudSource(CountingSource.upTo(451), options);
    SourceSplitResponse split = performSplit(source, options, 1L, null, /* numBundles limit */
    apiSizeLimitForTest);
    assertEquals(WorkerCustomSources.DEFAULT_NUM_BUNDLES_LIMIT, split.getBundles().size());
    // 0-99, 100-199, 200-299, 300-355, 356, 357, ... 451
    for (int i = 0; i <= 3; ++i) {
        DerivedSource derivedSource = split.getBundles().get(i);
        // Make sure that we are setting the flag telling Dataflow that we need further splits.
        assertFalse(derivedSource.getSource().getDoesNotNeedSplitting());
        Object deserializedSource = WorkerCustomSources.deserializeFromCloudSource(derivedSource.getSource().getSpec());
        assertTrue(deserializedSource instanceof SplittableOnlyBoundedSource);
        SplittableOnlyBoundedSource<?> splittableOnlySource = (SplittableOnlyBoundedSource<?>) deserializedSource;
        List<? extends BoundedSource<?>> splitSources = splittableOnlySource.split(1L, options);
        int expectedNumSplits = i < 3 ? 100 : 55;
        assertEquals(expectedNumSplits, splitSources.size());
        for (int j = 0; j < splitSources.size(); ++j) {
            assertTrue(splitSources.get(j) instanceof OffsetBasedSource);
            OffsetBasedSource<?> offsetBasedSource = (OffsetBasedSource<?>) splitSources.get(j);
            assertEquals(i * 100 + j, offsetBasedSource.getStartOffset());
            assertEquals(i * 100 + j + 1, offsetBasedSource.getEndOffset());
        }
    }
    for (int i = 4; i < WorkerCustomSources.DEFAULT_NUM_BUNDLES_LIMIT; ++i) {
        DerivedSource derivedSource = split.getBundles().get(i);
        // Make sure that we are not setting the flag telling Dataflow that we need further splits
        // for the individual counting sources
        assertTrue(derivedSource.getSource().getDoesNotNeedSplitting());
        Object deserializedSource = WorkerCustomSources.deserializeFromCloudSource(derivedSource.getSource().getSpec());
        assertTrue(deserializedSource instanceof OffsetBasedSource);
        OffsetBasedSource<?> offsetBasedSource = (OffsetBasedSource<?>) deserializedSource;
        assertEquals(351 + i, offsetBasedSource.getStartOffset());
        assertEquals(351 + i + 1, offsetBasedSource.getEndOffset());
    }
}
Also used : Source(com.google.api.services.dataflow.model.Source) OffsetBasedSource(org.apache.beam.sdk.io.OffsetBasedSource) SourceSplitResponse(com.google.api.services.dataflow.model.SourceSplitResponse) DerivedSource(com.google.api.services.dataflow.model.DerivedSource) CloudObject(org.apache.beam.runners.dataflow.util.CloudObject) Structs.getObject(org.apache.beam.runners.dataflow.util.Structs.getObject) SplittableOnlyBoundedSource(org.apache.beam.runners.dataflow.worker.WorkerCustomSources.SplittableOnlyBoundedSource) Test(org.junit.Test)

Example 4 with SourceSplitResponse

use of com.google.api.services.dataflow.model.SourceSplitResponse in project beam by apache.

the class WorkerCustomSourcesTest method testOversplittingDesiredBundleSizeScaledFirst.

@Test
public void testOversplittingDesiredBundleSizeScaledFirst() throws Exception {
    // Create a source that greatly oversplits but with coalescing/compression it would still fit
    // under the API limit. Test that the API limit gets applied first, so oversplitting is
    // reduced.
    com.google.api.services.dataflow.model.Source source = translateIOToCloudSource(CountingSource.upTo(8000), options);
    // Without either limit, produces 1000 bundles, total size ~500kb.
    // With only numBundles limit 100, produces 100 bundles, total size ~72kb.
    // With only apiSize limit = 10kb, 72 bundles, total size ~40kb (over the limit but oh well).
    // With numBundles limit 100 and apiSize limit 10kb, should produce 72 bundles.
    // On the other hand, if the numBundles limit of 100 was applied first, we'd get 100 bundles.
    SourceSplitResponse bundledWithOnlyNumBundlesLimit = performSplit(source, options, 8L, 100, /* numBundles limit */
    10000 * 1024L);
    assertEquals(100, bundledWithOnlyNumBundlesLimit.getBundles().size());
    assertThat(DataflowApiUtils.computeSerializedSizeBytes(bundledWithOnlyNumBundlesLimit), greaterThan(10 * 1024L));
    SourceSplitResponse bundledWithOnlySizeLimit = performSplit(source, options, 8L, 1000000, /* numBundles limit */
    10 * 1024L);
    int numBundlesWithOnlySizeLimit = bundledWithOnlySizeLimit.getBundles().size();
    assertThat(numBundlesWithOnlySizeLimit, lessThan(100));
    SourceSplitResponse bundledWithSizeLimit = performSplit(source, options, 8L, 100, 10 * 1024L);
    assertEquals(numBundlesWithOnlySizeLimit, bundledWithSizeLimit.getBundles().size());
}
Also used : SourceSplitResponse(com.google.api.services.dataflow.model.SourceSplitResponse) Source(com.google.api.services.dataflow.model.Source) Test(org.junit.Test)

Example 5 with SourceSplitResponse

use of com.google.api.services.dataflow.model.SourceSplitResponse in project beam by apache.

the class WorkerCustomSourcesTest method testSplitAndReadBundlesBack.

@Test
public void testSplitAndReadBundlesBack() throws Exception {
    com.google.api.services.dataflow.model.Source source = translateIOToCloudSource(CountingSource.upTo(10L), options);
    List<WindowedValue<Integer>> elems = readElemsFromSource(options, source);
    assertEquals(10L, elems.size());
    for (long i = 0; i < 10L; i++) {
        assertEquals(valueInGlobalWindow(i), elems.get((int) i));
    }
    SourceSplitResponse response = performSplit(source, options, 16L, /*desiredBundleSizeBytes for two longs*/
    null, /* numBundles limit */
    null);
    assertEquals("SOURCE_SPLIT_OUTCOME_SPLITTING_HAPPENED", response.getOutcome());
    List<DerivedSource> bundles = response.getBundles();
    assertEquals(5, bundles.size());
    for (int i = 0; i < 5; ++i) {
        DerivedSource bundle = bundles.get(i);
        assertEquals("SOURCE_DERIVATION_MODE_INDEPENDENT", bundle.getDerivationMode());
        com.google.api.services.dataflow.model.Source bundleSource = bundle.getSource();
        assertTrue(bundleSource.getDoesNotNeedSplitting());
        bundleSource.setCodec(source.getCodec());
        List<WindowedValue<Integer>> xs = readElemsFromSource(options, bundleSource);
        assertThat("Failed on bundle " + i, xs, contains(valueInGlobalWindow(0L + 2 * i), valueInGlobalWindow(1L + 2 * i)));
        assertTrue(bundle.getSource().getMetadata().getEstimatedSizeBytes() > 0);
    }
}
Also used : SourceSplitResponse(com.google.api.services.dataflow.model.SourceSplitResponse) WindowedValue(org.apache.beam.sdk.util.WindowedValue) Source(com.google.api.services.dataflow.model.Source) DerivedSource(com.google.api.services.dataflow.model.DerivedSource) Test(org.junit.Test)

Aggregations

SourceSplitResponse (com.google.api.services.dataflow.model.SourceSplitResponse)6 Test (org.junit.Test)6 Source (com.google.api.services.dataflow.model.Source)4 DerivedSource (com.google.api.services.dataflow.model.DerivedSource)3 SplittableOnlyBoundedSource (org.apache.beam.runners.dataflow.worker.WorkerCustomSources.SplittableOnlyBoundedSource)2 OffsetBasedSource (org.apache.beam.sdk.io.OffsetBasedSource)2 SourceSplitShard (com.google.api.services.dataflow.model.SourceSplitShard)1 ArrayList (java.util.ArrayList)1 DataflowPipelineOptions (org.apache.beam.runners.dataflow.options.DataflowPipelineOptions)1 CloudObject (org.apache.beam.runners.dataflow.util.CloudObject)1 Structs.getObject (org.apache.beam.runners.dataflow.util.Structs.getObject)1 TestCredential (org.apache.beam.sdk.extensions.gcp.auth.TestCredential)1 WindowedValue (org.apache.beam.sdk.util.WindowedValue)1