Examples with Watermark io.pravega.shared.watermarks.Watermark used on opensource projects

Example 1 with Watermark

use of io.pravega.shared.watermarks.Watermark in project pravega by pravega.

the class WatermarkSerializerTest method testWatermark.

@Test
public void testWatermark() {
    SegmentWithRange segmentWithRange1 = new SegmentWithRange(0L, 0.0, 0.5);
    SegmentWithRange segmentWithRange2 = new SegmentWithRange(1L, 0.5, 1.0);
    ImmutableMap<SegmentWithRange, Long> map = ImmutableMap.of(segmentWithRange1, 1L, segmentWithRange2, 1L);
    Watermark watermark = new Watermark(0L, 1L, map);
    WatermarkSerializer serializer = new WatermarkSerializer();
    ByteBuffer serialized = serializer.serialize(watermark);
    Watermark deserialized = serializer.deserialize(serialized);
    assertEquals(watermark, deserialized);
}

Example 2 with Watermark

use of io.pravega.shared.watermarks.Watermark in project pravega by pravega.

the class PeriodicWatermarking method filterWritersAndComputeWatermark.

private CompletionStage<Void> filterWritersAndComputeWatermark(String scope, String streamName, OperationContext context, WatermarkClient watermarkClient, Map<String, WriterMark> writers, StreamConfiguration config) {
    // 1. filter writers that are active.
    List<Entry<String, WriterMark>> activeWriters = new ArrayList<>();
    List<Entry<String, WriterMark>> inactiveWriters = new ArrayList<>();
    AtomicBoolean allActiveAreParticipating = new AtomicBoolean(true);
    writers.entrySet().forEach(x -> {
        if (watermarkClient.isWriterActive(x, config.getTimestampAggregationTimeout())) {
            activeWriters.add(x);
            if (!watermarkClient.isWriterParticipating(x.getValue().getTimestamp())) {
                allActiveAreParticipating.set(false);
            }
        } else {
            inactiveWriters.add(x);
        }
    });
    // Stop all inactive writers that have been shutdown.
    CompletableFuture<List<Void>> removeInactiveWriters = Futures.allOfWithResults(inactiveWriters.stream().map(x -> Futures.exceptionallyExpecting(streamMetadataStore.removeWriter(scope, streamName, x.getKey(), x.getValue(), context, executor).thenAccept(v -> watermarkClient.untrackWriterInactivity(x.getKey())), e -> Exceptions.unwrap(e) instanceof StoreException.WriteConflictException, null)).collect(Collectors.toList()));
    if (activeWriters.isEmpty()) {
        // periodic processing will resume.
        return removeInactiveWriters.thenCompose(v -> bucketStore.removeStreamFromBucketStore(BucketStore.ServiceType.WatermarkingService, scope, streamName, executor));
    }
    CompletableFuture<Watermark> watermarkFuture;
    if (!allActiveAreParticipating.get()) {
        // there are active writers that have not reported their marks. We should wait
        // until they either report or become inactive. So we will complete this iteration without
        // emitting any watermark (null) and in subsequent iterations if these writers have made progress
        // we will emit watermark or evict writers from watermark computation.
        watermarkFuture = CompletableFuture.completedFuture(null);
    } else {
        // compute new mark
        watermarkFuture = computeWatermark(scope, streamName, context, activeWriters, watermarkClient.getPreviousWatermark());
    }
    // we will compute watermark and remove inactive writers concurrently
    return CompletableFuture.allOf(removeInactiveWriters, watermarkFuture.thenAccept(watermarkClient::completeIteration));
}

Example 3 with Watermark

use of io.pravega.shared.watermarks.Watermark in project pravega by pravega.

the class WatermarkWorkflowTest method testWatermarkClient.

@Test(timeout = 10000L)
public void testWatermarkClient() {
    Stream stream = new StreamImpl("scope", "stream");
    SynchronizerClientFactory clientFactory = spy(SynchronizerClientFactory.class);
    @Cleanup MockRevisionedStreamClient revisionedClient = new MockRevisionedStreamClient();
    doAnswer(x -> revisionedClient).when(clientFactory).createRevisionedStreamClient(anyString(), any(), any());
    @Cleanup PeriodicWatermarking.WatermarkClient client = new PeriodicWatermarking.WatermarkClient(stream, clientFactory);
    // iteration 1 ==> null -> w1
    client.reinitialize();
    // There is no watermark in the stream. All values should be null and all writers active and participating.
    assertEquals(revisionedClient.getMark(), MockRevision.EMPTY);
    assertTrue(revisionedClient.watermarks.isEmpty());
    assertEquals(client.getPreviousWatermark(), Watermark.EMPTY);
    Map.Entry<String, WriterMark> entry0 = new AbstractMap.SimpleEntry<>("writerId", new WriterMark(0L, ImmutableMap.of()));
    Map.Entry<String, WriterMark> entry1 = new AbstractMap.SimpleEntry<>("writerId", new WriterMark(1L, ImmutableMap.of()));
    Map.Entry<String, WriterMark> entry2 = new AbstractMap.SimpleEntry<>("writerId", new WriterMark(2L, ImmutableMap.of()));
    Map.Entry<String, WriterMark> entry3 = new AbstractMap.SimpleEntry<>("writerId", new WriterMark(3L, ImmutableMap.of()));
    Map.Entry<String, WriterMark> entry4 = new AbstractMap.SimpleEntry<>("writerId", new WriterMark(4L, ImmutableMap.of()));
    Map.Entry<String, WriterMark> entry5 = new AbstractMap.SimpleEntry<>("writerId", new WriterMark(5L, ImmutableMap.of()));
    assertTrue(client.isWriterActive(entry0, 0L));
    assertTrue(client.isWriterParticipating(0L));
    Watermark first = new Watermark(1L, 2L, ImmutableMap.of());
    client.completeIteration(first);
    // iteration 2 : do not emit ==> w1 -> w1
    client.reinitialize();
    // There is one watermark. All writers should be active and writers greater than last watermark should be participating
    assertEquals(revisionedClient.getMark(), MockRevision.EMPTY);
    assertEquals(revisionedClient.watermarks.size(), 1);
    assertEquals(client.getPreviousWatermark(), first);
    assertTrue(client.isWriterActive(entry2, 0L));
    assertFalse(client.isWriterActive(entry1, 0L));
    assertTrue(client.isWriterTracked(entry1.getKey()));
    assertFalse(client.isWriterParticipating(1L));
    assertTrue(client.isWriterParticipating(2L));
    // dont emit a watermark. Everything stays same as before.
    client.completeIteration(null);
    // iteration 3 : emit ==> w1 -> w1 w2
    client.reinitialize();
    // There is one watermark. All writers should be active and writers greater than last watermark should be participating
    assertEquals(revisionedClient.getMark(), MockRevision.EMPTY);
    assertEquals(revisionedClient.watermarks.size(), 1);
    assertEquals(client.getPreviousWatermark(), first);
    assertTrue(client.isWriterActive(entry2, 0L));
    assertFalse(client.isWriterParticipating(1L));
    assertTrue(client.isWriterParticipating(2L));
    // emit second watermark
    Watermark second = new Watermark(2L, 3L, ImmutableMap.of());
    client.completeIteration(second);
    // iteration 4: do not emit ==> w1 w2 -> w1 w2
    client.reinitialize();
    assertEquals(revisionedClient.getMark(), revisionedClient.watermarks.get(0).getKey());
    assertEquals(2, revisionedClient.watermarks.size());
    assertEquals(client.getPreviousWatermark(), second);
    assertFalse(client.isWriterActive(entry2, 0L));
    assertTrue(client.isWriterTracked(entry2.getKey()));
    assertTrue(client.isWriterActive(entry3, 0L));
    assertFalse(client.isWriterParticipating(2L));
    assertTrue(client.isWriterParticipating(3L));
    assertTrue(client.isWriterActive(entry0, 1000L));
    assertTrue(client.isWriterTracked(entry0.getKey()));
    // dont emit a watermark but complete this iteration.
    client.completeIteration(null);
    // iteration 6: emit ==> w1 w2 -> w1 w2 w3
    client.reinitialize();
    assertEquals(revisionedClient.getMark(), revisionedClient.watermarks.get(0).getKey());
    assertEquals(2, revisionedClient.watermarks.size());
    assertEquals(client.getPreviousWatermark(), second);
    assertTrue(client.isWriterActive(entry3, 0L));
    assertFalse(client.isWriterTracked(entry3.getKey()));
    assertFalse(client.isWriterParticipating(2L));
    assertTrue(client.isWriterParticipating(3L));
    // emit third watermark
    Watermark third = new Watermark(3L, 4L, ImmutableMap.of());
    client.completeIteration(third);
    // iteration 7: do not emit ==> w1 w2 w3 -> w1 w2 w3
    client.reinitialize();
    // active writers should be ahead of first watermark. participating writers should be ahead of second watermark
    assertEquals(revisionedClient.getMark(), revisionedClient.watermarks.get(1).getKey());
    assertEquals(3, revisionedClient.watermarks.size());
    assertEquals(client.getPreviousWatermark(), third);
    assertFalse(client.isWriterActive(entry3, 0L));
    assertTrue(client.isWriterActive(entry4, 0L));
    assertFalse(client.isWriterParticipating(3L));
    assertTrue(client.isWriterParticipating(4L));
    client.completeIteration(null);
    // iteration 8 : emit ==> w2 w3 -> w2 w3 w4
    client.reinitialize();
    assertEquals(revisionedClient.getMark(), revisionedClient.watermarks.get(1).getKey());
    // window = w2 w3
    assertEquals(revisionedClient.watermarks.size(), 3);
    assertEquals(client.getPreviousWatermark(), third);
    assertFalse(client.isWriterActive(entry3, 0L));
    assertTrue(client.isWriterActive(entry4, 0L));
    assertFalse(client.isWriterParticipating(3L));
    assertTrue(client.isWriterParticipating(4L));
    // emit fourth watermark
    Watermark fourth = new Watermark(4L, 5L, ImmutableMap.of());
    client.completeIteration(fourth);
    // iteration 9: do not emit ==> w1 w2 w3 w4 -> w1 w2 w3 w4.. check writer timeout
    client.reinitialize();
    assertEquals(revisionedClient.getMark(), revisionedClient.watermarks.get(2).getKey());
    assertEquals(revisionedClient.watermarks.size(), 4);
    assertEquals(client.getPreviousWatermark(), fourth);
    assertFalse(client.isWriterActive(entry3, 0L));
    assertTrue(client.isWriterTracked(entry4.getKey()));
    assertFalse(client.isWriterParticipating(4L));
    assertTrue(client.isWriterParticipating(5L));
    // verify that writer is active if we specify a higher timeout
    assertTrue(client.isWriterActive(entry1, 1000L));
    assertTrue(client.isWriterTracked(entry1.getKey()));
    // now that the writer is being tracked
    assertFalse(Futures.delayedTask(() -> client.isWriterActive(entry1, 1L), Duration.ofSeconds(1), executor).join());
    assertTrue(client.isWriterTracked(entry1.getKey()));
    // dont emit a watermark but complete this iteration. This should shrink the window again.
    client.completeIteration(null);
    // iteration 10
    client.reinitialize();
    assertEquals(revisionedClient.getMark(), revisionedClient.watermarks.get(2).getKey());
    assertEquals(revisionedClient.watermarks.size(), 4);
    assertEquals(client.getPreviousWatermark(), fourth);
    assertFalse(client.isWriterActive(entry4, 0L));
    assertTrue(client.isWriterActive(entry5, 0L));
    assertFalse(client.isWriterParticipating(4L));
    assertTrue(client.isWriterParticipating(5L));
}

Example 4 with Watermark

use of io.pravega.shared.watermarks.Watermark in project pravega by pravega.

the class WatermarkWorkflowTest method testWatermarkingWorkflow.

@Test(timeout = 30000L)
public void testWatermarkingWorkflow() {
    SynchronizerClientFactory clientFactory = spy(SynchronizerClientFactory.class);
    ConcurrentHashMap<String, MockRevisionedStreamClient> revisionedStreamClientMap = new ConcurrentHashMap<>();
    doAnswer(x -> {
        String streamName = x.getArgument(0);
        return revisionedStreamClientMap.compute(streamName, (s, rsc) -> {
            if (rsc != null) {
                return rsc;
            } else {
                return new MockRevisionedStreamClient();
            }
        });
    }).when(clientFactory).createRevisionedStreamClient(anyString(), any(), any());
    @Cleanup PeriodicWatermarking periodicWatermarking = new PeriodicWatermarking(streamMetadataStore, bucketStore, sp -> clientFactory, executor, new RequestTracker(false));
    String streamName = "stream";
    String scope = "scope";
    streamMetadataStore.createScope(scope, null, executor).join();
    streamMetadataStore.createStream(scope, streamName, StreamConfiguration.builder().scalingPolicy(ScalingPolicy.fixed(3)).timestampAggregationTimeout(10000L).build(), System.currentTimeMillis(), null, executor).join();
    streamMetadataStore.setState(scope, streamName, State.ACTIVE, null, executor).join();
    // set minimum number of segments to 1
    StreamConfiguration config = StreamConfiguration.builder().scalingPolicy(ScalingPolicy.fixed(1)).timestampAggregationTimeout(10000L).build();
    streamMetadataStore.startUpdateConfiguration(scope, streamName, config, null, executor).join();
    VersionedMetadata<StreamConfigurationRecord> configRecord = streamMetadataStore.getConfigurationRecord(scope, streamName, null, executor).join();
    streamMetadataStore.completeUpdateConfiguration(scope, streamName, configRecord, null, executor).join();
    // 2. note writer1, writer2, writer3 marks
    // writer 1 reports segments 0, 1.
    // writer 2 reports segments 1, 2,
    // writer 3 reports segment 0, 2
    String writer1 = "writer1";
    Map<Long, Long> map1 = ImmutableMap.of(0L, 100L, 1L, 200L);
    streamMetadataStore.noteWriterMark(scope, streamName, writer1, 100L, map1, null, executor).join();
    String writer2 = "writer2";
    Map<Long, Long> map2 = ImmutableMap.of(1L, 100L, 2L, 200L);
    streamMetadataStore.noteWriterMark(scope, streamName, writer2, 101L, map2, null, executor).join();
    String writer3 = "writer3";
    Map<Long, Long> map3 = ImmutableMap.of(2L, 100L, 0L, 200L);
    streamMetadataStore.noteWriterMark(scope, streamName, writer3, 102L, map3, null, executor).join();
    // 3. run watermarking workflow.
    StreamImpl stream = new StreamImpl(scope, streamName);
    periodicWatermarking.watermark(stream).join();
    // verify that a watermark has been emitted.
    // this should emit a watermark that contains all three segments with offsets = 200L
    // and timestamp = 100L
    MockRevisionedStreamClient revisionedClient = revisionedStreamClientMap.get(NameUtils.getMarkStreamForStream(streamName));
    assertEquals(revisionedClient.watermarks.size(), 1);
    Watermark watermark = revisionedClient.watermarks.get(0).getValue();
    assertEquals(watermark.getLowerTimeBound(), 100L);
    assertEquals(watermark.getStreamCut().size(), 3);
    assertEquals(getSegmentOffset(watermark, 0L), 200L);
    assertEquals(getSegmentOffset(watermark, 1L), 200L);
    assertEquals(getSegmentOffset(watermark, 2L), 200L);
    // send positions only on segment 1 and segment 2. nothing on segment 0.
    map1 = ImmutableMap.of(1L, 300L);
    streamMetadataStore.noteWriterMark(scope, streamName, writer1, 200L, map1, null, executor).join();
    map2 = ImmutableMap.of(1L, 100L, 2L, 300L);
    streamMetadataStore.noteWriterMark(scope, streamName, writer2, 201L, map2, null, executor).join();
    map3 = ImmutableMap.of(2L, 300L);
    streamMetadataStore.noteWriterMark(scope, streamName, writer3, 202L, map3, null, executor).join();
    // run watermark workflow. this will emit a watermark with time = 200L and streamcut = 0 -> 200L, 1 -> 300L, 2 -> 300L
    periodicWatermarking.watermark(stream).join();
    assertEquals(revisionedClient.watermarks.size(), 2);
    watermark = revisionedClient.watermarks.get(1).getValue();
    assertEquals(watermark.getLowerTimeBound(), 200L);
    assertEquals(watermark.getStreamCut().size(), 3);
    assertEquals(getSegmentOffset(watermark, 0L), 200L);
    assertEquals(getSegmentOffset(watermark, 1L), 300L);
    assertEquals(getSegmentOffset(watermark, 2L), 300L);
    // scale stream 0, 1, 2 -> 3, 4
    scaleStream(streamName, scope);
    // writer 1 reports segments 0, 1.
    // writer 2 reports segments 1, 2
    // writer 3 reports segment 3
    map1 = ImmutableMap.of(0L, 300L, 1L, 400L);
    streamMetadataStore.noteWriterMark(scope, streamName, writer1, 302L, map1, null, executor).join();
    map2 = ImmutableMap.of(1L, 100L, 2L, 400L);
    streamMetadataStore.noteWriterMark(scope, streamName, writer2, 301L, map2, null, executor).join();
    long segment3 = NameUtils.computeSegmentId(3, 1);
    long segment4 = NameUtils.computeSegmentId(4, 1);
    map3 = ImmutableMap.of(segment3, 100L);
    // writer 3 has lowest reported time.
    streamMetadataStore.noteWriterMark(scope, streamName, writer3, 300L, map3, null, executor).join();
    // run watermark workflow. this will emit a watermark with time = 300L and streamcut = 3 -> 100L, 4 -> 0L
    periodicWatermarking.watermark(stream).join();
    assertEquals(revisionedClient.watermarks.size(), 3);
    watermark = revisionedClient.watermarks.get(2).getValue();
    assertEquals(watermark.getLowerTimeBound(), 300L);
    assertEquals(watermark.getStreamCut().size(), 2);
    assertEquals(getSegmentOffset(watermark, segment3), 100L);
    assertEquals(getSegmentOffset(watermark, segment4), 0L);
    // report complete positions from writers.
    // writer 1 reports 0, 1, 2
    // writer 2 reports 0, 1, 2
    // writer 3 doesnt report.
    map1 = ImmutableMap.of(0L, 400L, 1L, 400L);
    streamMetadataStore.noteWriterMark(scope, streamName, writer1, 400L, map1, null, executor).join();
    map2 = ImmutableMap.of(1L, 100L, 2L, 400L);
    streamMetadataStore.noteWriterMark(scope, streamName, writer2, 401L, map2, null, executor).join();
    // run watermark workflow. there shouldn't be a watermark emitted because writer 3 is active and has not reported a time.
    periodicWatermarking.watermark(stream).join();
    assertEquals(revisionedClient.watermarks.size(), 3);
    // even though writer3 is excluded from computation, its mark is still not removed because it is still active
    WriterMark writer3Mark = streamMetadataStore.getWriterMark(scope, streamName, writer3, null, executor).join();
    assertTrue(writer3Mark.isAlive());
    assertEquals(writer3Mark.getTimestamp(), 300L);
    // report shutdown of writer 3
    streamMetadataStore.shutdownWriter(scope, streamName, writer3, null, executor).join();
    writer3Mark = streamMetadataStore.getWriterMark(scope, streamName, writer3, null, executor).join();
    assertFalse(writer3Mark.isAlive());
    assertEquals(writer3Mark.getTimestamp(), 300L);
    // now a watermark should be generated. Time should be advanced. But watermark's stream cut is already ahead of writer's
    // positions so stream cut should not advance.
    // Also writer 3 being inactive and shutdown, should be removed.
    periodicWatermarking.watermark(stream).join();
    assertEquals(revisionedClient.watermarks.size(), 4);
    watermark = revisionedClient.watermarks.get(3).getValue();
    assertEquals(watermark.getLowerTimeBound(), 400L);
    assertEquals(watermark.getStreamCut().size(), 2);
    assertEquals(getSegmentOffset(watermark, segment3), 100L);
    assertEquals(getSegmentOffset(watermark, segment4), 0L);
    AssertExtensions.assertFutureThrows("Writer 3 should have been removed from store", streamMetadataStore.getWriterMark(scope, streamName, writer3, null, executor), e -> Exceptions.unwrap(e) instanceof StoreException.DataNotFoundException);
    // writer 1, 2 and 3 report marks. With writer 3 reporting mark on segment 4. Writer3 will get added again
    map1 = ImmutableMap.of(0L, 500L, 1L, 500L);
    streamMetadataStore.noteWriterMark(scope, streamName, writer1, 500L, map1, null, executor).join();
    map2 = ImmutableMap.of(1L, 100L, 2L, 500L);
    streamMetadataStore.noteWriterMark(scope, streamName, writer2, 501L, map2, null, executor).join();
    map3 = ImmutableMap.of(segment4, 500L);
    streamMetadataStore.noteWriterMark(scope, streamName, writer3, 502L, map3, null, executor).join();
    // run watermarking workflow. It should generate watermark that includes segments 3 -> 100L and 4 -> 500L with time 500L
    periodicWatermarking.watermark(stream).join();
    assertEquals(revisionedClient.watermarks.size(), 5);
    watermark = revisionedClient.watermarks.get(4).getValue();
    assertEquals(watermark.getLowerTimeBound(), 500L);
    assertEquals(watermark.getStreamCut().size(), 2);
    assertEquals(getSegmentOffset(watermark, segment3), 100L);
    assertEquals(getSegmentOffset(watermark, segment4), 500L);
}

Example 5 with Watermark

use of io.pravega.shared.watermarks.Watermark in project pravega by pravega.

the class RestoreBackUpDataRecoveryTest method getStreamCutsFromWaterMarks.

private List<Map<Stream, StreamCut>> getStreamCutsFromWaterMarks(Stream streamObj, LinkedBlockingQueue<Watermark> watermarks) throws InterruptedException {
    Watermark watermark0 = watermarks.take();
    Watermark watermark1 = watermarks.take();
    assertTrue(watermark0.getLowerTimeBound() <= watermark0.getUpperTimeBound());
    assertTrue(watermark1.getLowerTimeBound() <= watermark1.getUpperTimeBound());
    assertTrue(watermark0.getLowerTimeBound() < watermark1.getLowerTimeBound());
    Map<Segment, Long> positionMap0 = watermark0.getStreamCut().entrySet().stream().collect(Collectors.toMap(x -> new Segment(SCOPE, STREAM1, x.getKey().getSegmentId()), Map.Entry::getValue));
    Map<Segment, Long> positionMap1 = watermark1.getStreamCut().entrySet().stream().collect(Collectors.toMap(x -> new Segment(SCOPE, STREAM1, x.getKey().getSegmentId()), Map.Entry::getValue));
    StreamCut streamCutFirst = new StreamCutImpl(streamObj, positionMap0);
    StreamCut streamCutSecond = new StreamCutImpl(streamObj, positionMap1);
    Map<Stream, StreamCut> firstMarkStreamCut = Collections.singletonMap(streamObj, streamCutFirst);
    Map<Stream, StreamCut> secondMarkStreamCut = Collections.singletonMap(streamObj, streamCutSecond);
    return Arrays.asList(firstMarkStreamCut, secondMarkStreamCut);
}