Examples with IsmRecord org.apache.beam.runners.dataflow.internal.IsmFormat.IsmRecord used on opensource projects

Example 1 with IsmRecord

use of org.apache.beam.runners.dataflow.internal.IsmFormat.IsmRecord in project beam by apache.

the class IsmSideInputReaderTest method testMap.

@Test
public void testMap() throws Exception {
    // Note that we purposely use byte[]s as keys to force structural equality testing
    // versus using java equality testing.
    Coder<WindowedValue<Long>> valueCoder = WindowedValue.getFullCoder(VarLongCoder.of(), GLOBAL_WINDOW_CODER);
    final ListMultimap<byte[], WindowedValue<Long>> elements = ImmutableListMultimap.<byte[], WindowedValue<Long>>builder().put(new byte[] { 0x00 }, valueInGlobalWindow(12L)).put(new byte[] { 0x01 }, valueInGlobalWindow(22L)).put(new byte[] { 0x02 }, valueInGlobalWindow(32L)).put(new byte[] { 0x03 }, valueInGlobalWindow(42L)).put(new byte[] { 0x04 }, valueInGlobalWindow(52L)).put(new byte[] { 0x05 }, valueInGlobalWindow(62L)).build();
    final PCollectionView<Map<byte[], Long>> view = Pipeline.create().apply(Create.empty(KvCoder.of(ByteArrayCoder.of(), VarLongCoder.of()))).apply(View.asMap());
    IsmRecordCoder<WindowedValue<Long>> ismCoder = IsmRecordCoder.of(1, 2, ImmutableList.of(MetadataKeyCoder.of(ByteArrayCoder.of()), GLOBAL_WINDOW_CODER, BigEndianLongCoder.of()), valueCoder);
    Multimap<Integer, IsmRecord<WindowedValue<Long>>> elementsPerShard = forMap(ismCoder, elements);
    List<IsmRecord<WindowedValue<Long>>> firstElements = new ArrayList<>();
    List<IsmRecord<WindowedValue<Long>>> secondElements = new ArrayList<>();
    for (Map.Entry<Integer, Collection<IsmRecord<WindowedValue<Long>>>> entry : elementsPerShard.asMap().entrySet()) {
        if (entry.getKey() % 2 == 0) {
            firstElements.addAll(entry.getValue());
        } else {
            secondElements.addAll(entry.getValue());
        }
    }
    // Ensure that each file will have some records.
    checkState(!firstElements.isEmpty());
    checkState(!secondElements.isEmpty());
    Source sourceA = initInputFile(firstElements, ismCoder);
    Source sourceB = initInputFile(secondElements, ismCoder);
    List<IsmRecord<WindowedValue<Long>>> mapMetadata = forMapMetadata(ByteArrayCoder.of(), elements.keySet(), GlobalWindow.INSTANCE);
    Source sourceMeta = initInputFile(mapMetadata, ismCoder);
    final IsmSideInputReader reader = sideInputReader(view.getTagInternal().getId(), sourceA, sourceB, sourceMeta);
    List<Callable<Map<byte[], Long>>> tasks = new ArrayList<>();
    for (int i = 0; i < NUM_THREADS; ++i) {
        tasks.add(() -> {
            // Store a strong reference to the returned value so that the logical reference
            // cache is not cleared for this test.
            Map<byte[], Long> value = reader.get(view, GlobalWindow.INSTANCE);
            verifyMap(Maps.transformValues(elements.asMap(), new TransformForMap<Long>()), value, new ComparatorForMap<Long>());
            // Assert that the same value reference was returned showing that it was cached.
            assertSame(reader.get(view, GlobalWindow.INSTANCE), value);
            return value;
        });
    }
    List<Future<Map<byte[], Long>>> results = pipelineOptions.getExecutorService().invokeAll(tasks);
    Map<byte[], Long> value = results.get(0).get();
    // Assert that all threads got back the same reference
    for (Future<Map<byte[], Long>> result : results) {
        assertSame(value, result.get());
    }
}

Example 2 with IsmRecord

use of org.apache.beam.runners.dataflow.internal.IsmFormat.IsmRecord in project beam by apache.

the class IsmSideInputReaderTest method testMapInWindow.

@Test
public void testMapInWindow() throws Exception {
    // Note that we purposely use byte[]s as keys to force structural equality testing
    // versus using java equality testing.
    Coder<WindowedValue<Long>> valueCoder = WindowedValue.getFullCoder(VarLongCoder.of(), INTERVAL_WINDOW_CODER);
    final ListMultimap<byte[], WindowedValue<Long>> firstWindow = ImmutableListMultimap.<byte[], WindowedValue<Long>>builder().put(new byte[] { 0x00 }, valueInIntervalWindow(12L, 10)).put(new byte[] { 0x01 }, valueInIntervalWindow(22L, 10)).put(new byte[] { 0x02 }, valueInIntervalWindow(32L, 10)).build();
    final ListMultimap<byte[], WindowedValue<Long>> secondWindow = ImmutableListMultimap.<byte[], WindowedValue<Long>>builder().put(new byte[] { 0x00 }, valueInIntervalWindow(42L, 20)).put(new byte[] { 0x03 }, valueInIntervalWindow(52L, 20)).put(new byte[] { 0x02 }, valueInIntervalWindow(62L, 20)).build();
    final ListMultimap<byte[], WindowedValue<Long>> thirdWindow = ImmutableListMultimap.<byte[], WindowedValue<Long>>builder().put(new byte[] { 0x02 }, valueInIntervalWindow(72L, 30)).put(new byte[] { 0x04 }, valueInIntervalWindow(82L, 30)).put(new byte[] { 0x05 }, valueInIntervalWindow(92L, 30)).build();
    final PCollectionView<Map<byte[], Long>> view = Pipeline.create().apply(Create.empty(KvCoder.of(ByteArrayCoder.of(), VarLongCoder.of()))).apply(Window.into(FixedWindows.of(Duration.millis(10)))).apply(View.asMap());
    IsmRecordCoder<WindowedValue<Long>> ismCoder = IsmRecordCoder.of(1, 2, ImmutableList.of(MetadataKeyCoder.of(ByteArrayCoder.of()), INTERVAL_WINDOW_CODER, BigEndianLongCoder.of()), valueCoder);
    Multimap<Integer, IsmRecord<WindowedValue<Long>>> elementsPerShard = forMap(ismCoder, firstWindow);
    elementsPerShard.putAll(forMap(ismCoder, secondWindow));
    elementsPerShard.putAll(forMap(ismCoder, thirdWindow));
    List<IsmRecord<WindowedValue<Long>>> firstElements = new ArrayList<>();
    List<IsmRecord<WindowedValue<Long>>> secondElements = new ArrayList<>();
    for (Map.Entry<Integer, Collection<IsmRecord<WindowedValue<Long>>>> entry : elementsPerShard.asMap().entrySet()) {
        if (entry.getKey() % 2 == 0) {
            firstElements.addAll(entry.getValue());
        } else {
            secondElements.addAll(entry.getValue());
        }
    }
    // Ensure that each file will have some records.
    checkState(!firstElements.isEmpty());
    checkState(!secondElements.isEmpty());
    Source sourceA = initInputFile(firstElements, ismCoder);
    Source sourceB = initInputFile(secondElements, ismCoder);
    List<IsmRecord<WindowedValue<Long>>> firstWindowMapMetadata = forMapMetadata(ByteArrayCoder.of(), firstWindow.keySet(), intervalWindow(10));
    List<IsmRecord<WindowedValue<Long>>> secondWindowMapMetadata = forMapMetadata(ByteArrayCoder.of(), secondWindow.keySet(), intervalWindow(20));
    List<IsmRecord<WindowedValue<Long>>> thirdWindowMapMetadata = forMapMetadata(ByteArrayCoder.of(), thirdWindow.keySet(), intervalWindow(30));
    Source sourceMetaA = initInputFile(firstWindowMapMetadata, ismCoder);
    Source sourceMetaB = initInputFile(concat(secondWindowMapMetadata, thirdWindowMapMetadata), ismCoder);
    final IsmSideInputReader reader = sideInputReader(view.getTagInternal().getId(), sourceA, sourceB, sourceMetaA, sourceMetaB);
    List<Callable<Map<BoundedWindow, Map<byte[], Long>>>> tasks = new ArrayList<>();
    for (int i = 0; i < NUM_THREADS; ++i) {
        tasks.add(() -> {
            // Store a strong reference to the returned value so that the logical reference
            // cache is not cleared for this test.
            Map<byte[], Long> firstValues = reader.get(view, intervalWindow(10));
            Map<byte[], Long> secondValues = reader.get(view, intervalWindow(20));
            Map<byte[], Long> thirdValues = reader.get(view, intervalWindow(30));
            verifyMap(Maps.transformValues(firstWindow.asMap(), new TransformForMap<Long>()), firstValues, new ComparatorForMap<Long>());
            verifyMap(Maps.transformValues(secondWindow.asMap(), new TransformForMap<Long>()), secondValues, new ComparatorForMap<Long>());
            verifyMap(Maps.transformValues(thirdWindow.asMap(), new TransformForMap<Long>()), thirdValues, new ComparatorForMap<Long>());
            // Assert that the same value reference was returned showing that it was cached.
            assertSame(firstValues, reader.get(view, intervalWindow(10)));
            assertSame(secondValues, reader.get(view, intervalWindow(20)));
            assertSame(thirdValues, reader.get(view, intervalWindow(30)));
            // Also verify when requesting a window that is not part of the side input
            assertEquals(Collections.EMPTY_MAP, reader.get(view, intervalWindow(40)));
            return ImmutableMap.<BoundedWindow, Map<byte[], Long>>of(intervalWindow(10), firstValues, intervalWindow(20), secondValues, intervalWindow(30), thirdValues);
        });
    }
    List<Future<Map<BoundedWindow, Map<byte[], Long>>>> results = pipelineOptions.getExecutorService().invokeAll(tasks);
    Map<BoundedWindow, Map<byte[], Long>> value = results.get(0).get();
    // Assert that all threads got back the same reference
    for (Future<Map<BoundedWindow, Map<byte[], Long>>> result : results) {
        assertEquals(value, result.get());
        for (Map.Entry<BoundedWindow, Map<byte[], Long>> entry : result.get().entrySet()) {
            assertSame(value.get(entry.getKey()), entry.getValue());
        }
    }
}

Example 3 with IsmRecord

use of org.apache.beam.runners.dataflow.internal.IsmFormat.IsmRecord in project beam by apache.

the class IsmReaderTest method testInitializationForLargeFilesIsNotCached.

@Test
public void testInitializationForLargeFilesIsNotCached() throws Exception {
    File tmpFile = tmpFolder.newFile();
    List<IsmRecord<byte[]>> data = new ArrayList<>();
    // footer read buffer used by IsmReader to optimize small file handling.
    for (int i = 0; i < IsmReaderImpl.MAX_SHARD_INDEX_AND_FOOTER_SIZE / TEST_BLOCK_SIZE + 1; ++i) {
        data.add(IsmRecord.<byte[]>of(ImmutableList.of(EMPTY, Ints.toByteArray(i)), new byte[TEST_BLOCK_SIZE]));
    }
    writeElementsToFile(data, tmpFile);
    IsmReader<byte[]> reader = new IsmReaderImpl<byte[]>(FileSystems.matchSingleFileSpec(tmpFile.getAbsolutePath()).resourceId(), CODER, cache);
    assertFalse(reader.isInitialized());
    assertEquals(0, cache.size());
    reader.overKeyComponents(ImmutableList.of());
    assertTrue(reader.isInitialized());
    assertEquals(0, cache.size());
}

Example 4 with IsmRecord

use of org.apache.beam.runners.dataflow.internal.IsmFormat.IsmRecord in project beam by apache.

the class IsmReaderTest method testIsEmpty.

@Test
public void testIsEmpty() throws Exception {
    File tmpFile = tmpFolder.newFile();
    List<IsmRecord<byte[]>> data = new ArrayList<>();
    writeElementsToFile(data, tmpFile);
    IsmReader<byte[]> reader = new IsmReaderImpl<byte[]>(FileSystems.matchSingleFileSpec(tmpFile.getAbsolutePath()).resourceId(), CODER, cache);
    assertFalse(reader.isInitialized());
    assertTrue(reader.isEmpty());
    assertTrue(reader.isInitialized());
}

Example 5 with IsmRecord

use of org.apache.beam.runners.dataflow.internal.IsmFormat.IsmRecord in project beam by apache.

the class IsmReaderTest method testInitializationForSmallFilesIsCached.

@Test
public void testInitializationForSmallFilesIsCached() throws Exception {
    File tmpFile = tmpFolder.newFile();
    IsmShardKey expectedShardKey = new IsmShardKey(tmpFile.getAbsolutePath(), new RandomAccessData(0), 0, 13);
    List<IsmRecord<byte[]>> data = new ArrayList<>();
    data.add(IsmRecord.<byte[]>of(ImmutableList.of(EMPTY, new byte[] { 0x04 }), new byte[] { 0x04 }));
    data.add(IsmRecord.<byte[]>of(ImmutableList.of(EMPTY, new byte[] { 0x08 }), new byte[] { 0x08 }));
    writeElementsToFile(data, tmpFile);
    IsmReader<byte[]> reader = new IsmReaderImpl<byte[]>(FileSystems.matchSingleFileSpec(tmpFile.getAbsolutePath()).resourceId(), CODER, cache);
    // Validate that reader and cache are in initial state
    assertFalse(reader.isInitialized());
    assertEquals(0, cache.size());
    // Force initialization
    reader.overKeyComponents(ImmutableList.of());
    // Validate reader is initialized and expected entry is cached
    assertTrue(reader.isInitialized());
    WeightedValue<NavigableMap<RandomAccessData, WindowedValue<IsmRecord<byte[]>>>> block = cache.getIfPresent(expectedShardKey);
    assertNotNull(block);
    assertArrayEquals(new byte[] { 0x04 }, block.getValue().firstEntry().getValue().getValue().getValue());
    assertArrayEquals(new byte[] { 0x08 }, block.getValue().lastEntry().getValue().getValue().getValue());
}