Examples with Segment org.apache.druid.segment.Segment used on opensource projects

Example 6 with Segment

use of org.apache.druid.segment.Segment in project druid by druid-io.

the class AbstractMultiPhaseParallelIndexingTest method querySegment.

List<ScanResultValue> querySegment(DataSegment dataSegment, List<String> columns, File tempSegmentDir) {
    Segment segment = loadSegment(dataSegment, tempSegmentDir);
    final QueryRunner<ScanResultValue> runner = SCAN_QUERY_RUNNER_FACTORY.createRunner(segment);
    return runner.run(QueryPlus.wrap(new ScanQuery(new TableDataSource("dataSource"), new SpecificSegmentSpec(new SegmentDescriptor(dataSegment.getInterval(), dataSegment.getVersion(), dataSegment.getShardSpec().getPartitionNum())), null, null, 0, 0, 0, null, null, null, columns, false, null))).toList();
}

Example 7 with Segment

use of org.apache.druid.segment.Segment in project druid by druid-io.

the class OnheapIncrementalIndexBenchmark method testConcurrentAddRead.

@Ignore
@Test
@BenchmarkOptions(callgc = true, clock = Clock.REAL_TIME, warmupRounds = 10, benchmarkRounds = 20)
public void testConcurrentAddRead() throws InterruptedException, ExecutionException, NoSuchMethodException, IllegalAccessException, InvocationTargetException, InstantiationException {
    final int taskCount = 30;
    final int concurrentThreads = 3;
    final int elementsPerThread = 1 << 15;
    final IncrementalIndex incrementalIndex = this.incrementalIndex.getConstructor(IncrementalIndexSchema.class, boolean.class, boolean.class, boolean.class, boolean.class, int.class).newInstance(new IncrementalIndexSchema.Builder().withMetrics(factories).build(), true, true, false, true, elementsPerThread * taskCount);
    final ArrayList<AggregatorFactory> queryAggregatorFactories = new ArrayList<>(DIMENSION_COUNT + 1);
    queryAggregatorFactories.add(new CountAggregatorFactory("rows"));
    for (int i = 0; i < DIMENSION_COUNT; ++i) {
        queryAggregatorFactories.add(new LongSumAggregatorFactory(StringUtils.format("sumResult%s", i), StringUtils.format("sumResult%s", i)));
        queryAggregatorFactories.add(new DoubleSumAggregatorFactory(StringUtils.format("doubleSumResult%s", i), StringUtils.format("doubleSumResult%s", i)));
    }
    final ListeningExecutorService indexExecutor = MoreExecutors.listeningDecorator(Executors.newFixedThreadPool(concurrentThreads, new ThreadFactoryBuilder().setDaemon(false).setNameFormat("index-executor-%d").setPriority(Thread.MIN_PRIORITY).build()));
    final ListeningExecutorService queryExecutor = MoreExecutors.listeningDecorator(Executors.newFixedThreadPool(concurrentThreads, new ThreadFactoryBuilder().setDaemon(false).setNameFormat("query-executor-%d").build()));
    final long timestamp = System.currentTimeMillis();
    final Interval queryInterval = Intervals.of("1900-01-01T00:00:00Z/2900-01-01T00:00:00Z");
    final List<ListenableFuture<?>> indexFutures = new ArrayList<>();
    final List<ListenableFuture<?>> queryFutures = new ArrayList<>();
    final Segment incrementalIndexSegment = new IncrementalIndexSegment(incrementalIndex, null);
    final QueryRunnerFactory factory = new TimeseriesQueryRunnerFactory(new TimeseriesQueryQueryToolChest(), new TimeseriesQueryEngine(), QueryRunnerTestHelper.NOOP_QUERYWATCHER);
    final AtomicInteger currentlyRunning = new AtomicInteger(0);
    final AtomicBoolean concurrentlyRan = new AtomicBoolean(false);
    final AtomicBoolean someoneRan = new AtomicBoolean(false);
    for (int j = 0; j < taskCount; j++) {
        indexFutures.add(indexExecutor.submit(new Runnable() {

            @Override
            public void run() {
                currentlyRunning.incrementAndGet();
                try {
                    for (int i = 0; i < elementsPerThread; i++) {
                        incrementalIndex.add(getLongRow(timestamp + i, 1, DIMENSION_COUNT));
                    }
                } catch (IndexSizeExceededException e) {
                    throw new RuntimeException(e);
                }
                currentlyRunning.decrementAndGet();
                someoneRan.set(true);
            }
        }));
        queryFutures.add(queryExecutor.submit(new Runnable() {

            @Override
            public void run() {
                QueryRunner<Result<TimeseriesResultValue>> runner = new FinalizeResultsQueryRunner<Result<TimeseriesResultValue>>(factory.createRunner(incrementalIndexSegment), factory.getToolchest());
                TimeseriesQuery query = Druids.newTimeseriesQueryBuilder().dataSource("xxx").granularity(Granularities.ALL).intervals(ImmutableList.of(queryInterval)).aggregators(queryAggregatorFactories).build();
                List<Result<TimeseriesResultValue>> results = runner.run(QueryPlus.wrap(query)).toList();
                for (Result<TimeseriesResultValue> result : results) {
                    if (someoneRan.get()) {
                        Assert.assertTrue(result.getValue().getDoubleMetric("doubleSumResult0") > 0);
                    }
                }
                if (currentlyRunning.get() > 0) {
                    concurrentlyRan.set(true);
                }
            }
        }));
    }
    List<ListenableFuture<?>> allFutures = new ArrayList<>(queryFutures.size() + indexFutures.size());
    allFutures.addAll(queryFutures);
    allFutures.addAll(indexFutures);
    Futures.allAsList(allFutures).get();
    // Assert.assertTrue("Did not hit concurrency, please try again", concurrentlyRan.get());
    queryExecutor.shutdown();
    indexExecutor.shutdown();
    QueryRunner<Result<TimeseriesResultValue>> runner = new FinalizeResultsQueryRunner<Result<TimeseriesResultValue>>(factory.createRunner(incrementalIndexSegment), factory.getToolchest());
    TimeseriesQuery query = Druids.newTimeseriesQueryBuilder().dataSource("xxx").granularity(Granularities.ALL).intervals(ImmutableList.of(queryInterval)).aggregators(queryAggregatorFactories).build();
    List<Result<TimeseriesResultValue>> results = runner.run(QueryPlus.wrap(query)).toList();
    final int expectedVal = elementsPerThread * taskCount;
    for (Result<TimeseriesResultValue> result : results) {
        Assert.assertEquals(elementsPerThread, result.getValue().getLongMetric("rows").intValue());
        for (int i = 0; i < DIMENSION_COUNT; ++i) {
            Assert.assertEquals(StringUtils.format("Failed long sum on dimension %d", i), expectedVal, result.getValue().getLongMetric(StringUtils.format("sumResult%s", i)).intValue());
            Assert.assertEquals(StringUtils.format("Failed double sum on dimension %d", i), expectedVal, result.getValue().getDoubleMetric(StringUtils.format("doubleSumResult%s", i)).intValue());
        }
    }
}

Example 8 with Segment

use of org.apache.druid.segment.Segment in project druid by druid-io.

the class AppenderatorImpl method persistHydrant.

/**
 * Persists the given hydrant and returns the number of rows persisted. Must only be called in the single-threaded
 * persistExecutor.
 *
 * @param indexToPersist hydrant to persist
 * @param identifier     the segment this hydrant is going to be part of
 *
 * @return the number of rows persisted
 */
private int persistHydrant(FireHydrant indexToPersist, SegmentIdWithShardSpec identifier) {
    synchronized (indexToPersist) {
        if (indexToPersist.hasSwapped()) {
            log.info("Segment[%s] hydrant[%s] already swapped. Ignoring request to persist.", identifier, indexToPersist);
            return 0;
        }
        log.debug("Segment[%s], persisting Hydrant[%s]", identifier, indexToPersist);
        try {
            final long startTime = System.nanoTime();
            int numRows = indexToPersist.getIndex().size();
            final File persistedFile;
            final File persistDir = createPersistDirIfNeeded(identifier);
            persistedFile = indexMerger.persist(indexToPersist.getIndex(), identifier.getInterval(), new File(persistDir, String.valueOf(indexToPersist.getCount())), tuningConfig.getIndexSpecForIntermediatePersists(), tuningConfig.getSegmentWriteOutMediumFactory());
            log.info("Flushed in-memory data for segment[%s] spill[%s] to disk in [%,d] ms (%,d rows).", indexToPersist.getSegmentId(), indexToPersist.getCount(), (System.nanoTime() - startTime) / 1000000, numRows);
            // Map only when this appenderator is being driven by a real time task:
            Segment segmentToSwap = null;
            if (isOpenSegments()) {
                segmentToSwap = new QueryableIndexSegment(indexIO.loadIndex(persistedFile), indexToPersist.getSegmentId());
            } else {
                // remember file path & segment id to rebuild the queryable index for merge:
                persistedHydrantMetadata.put(indexToPersist, new Pair<>(persistedFile, indexToPersist.getSegmentId()));
            }
            indexToPersist.swapSegment(segmentToSwap);
            return numRows;
        } catch (IOException e) {
            log.makeAlert("Incremental persist failed").addData("segment", identifier.toString()).addData("dataSource", schema.getDataSource()).addData("count", indexToPersist.getCount()).emit();
            throw new RuntimeException(e);
        }
    }
}

Example 9 with Segment

use of org.apache.druid.segment.Segment in project druid by druid-io.

the class SegmentLocalCacheLoader method getSegment.

@Override
public ReferenceCountingSegment getSegment(DataSegment segment, boolean lazy, SegmentLazyLoadFailCallback loadFailed) throws SegmentLoadingException {
    final File segmentFiles = cacheManager.getSegmentFiles(segment);
    File factoryJson = new File(segmentFiles, "factory.json");
    final SegmentizerFactory factory;
    if (factoryJson.exists()) {
        try {
            factory = jsonMapper.readValue(factoryJson, SegmentizerFactory.class);
        } catch (IOException e) {
            throw new SegmentLoadingException(e, "%s", e.getMessage());
        }
    } else {
        factory = new MMappedQueryableSegmentizerFactory(indexIO);
    }
    Segment segmentObject = factory.factorize(segment, segmentFiles, lazy, loadFailed);
    return ReferenceCountingSegment.wrapSegment(segmentObject, segment.getShardSpec());
}

Example 10 with Segment

use of org.apache.druid.segment.Segment in project druid by druid-io.

the class IncrementalIndexTest method testConcurrentAddRead.

@Test(timeout = 60_000L)
public void testConcurrentAddRead() throws InterruptedException, ExecutionException {
    final int dimensionCount = 5;
    final ArrayList<AggregatorFactory> ingestAggregatorFactories = new ArrayList<>(dimensionCount + 1);
    ingestAggregatorFactories.add(new CountAggregatorFactory("rows"));
    for (int i = 0; i < dimensionCount; ++i) {
        ingestAggregatorFactories.add(new LongSumAggregatorFactory(StringUtils.format("sumResult%s", i), StringUtils.format("Dim_%s", i)));
        ingestAggregatorFactories.add(new DoubleSumAggregatorFactory(StringUtils.format("doubleSumResult%s", i), StringUtils.format("Dim_%s", i)));
    }
    final ArrayList<AggregatorFactory> queryAggregatorFactories = new ArrayList<>(dimensionCount + 1);
    queryAggregatorFactories.add(new CountAggregatorFactory("rows"));
    for (int i = 0; i < dimensionCount; ++i) {
        queryAggregatorFactories.add(new LongSumAggregatorFactory(StringUtils.format("sumResult%s", i), StringUtils.format("sumResult%s", i)));
        queryAggregatorFactories.add(new DoubleSumAggregatorFactory(StringUtils.format("doubleSumResult%s", i), StringUtils.format("doubleSumResult%s", i)));
    }
    final IncrementalIndex index = indexCreator.createIndex((Object) ingestAggregatorFactories.toArray(new AggregatorFactory[0]));
    final int concurrentThreads = 2;
    final int elementsPerThread = 10_000;
    final ListeningExecutorService indexExecutor = MoreExecutors.listeningDecorator(Executors.newFixedThreadPool(concurrentThreads, new ThreadFactoryBuilder().setDaemon(false).setNameFormat("index-executor-%d").setPriority(Thread.MIN_PRIORITY).build()));
    final ListeningExecutorService queryExecutor = MoreExecutors.listeningDecorator(Executors.newFixedThreadPool(concurrentThreads, new ThreadFactoryBuilder().setDaemon(false).setNameFormat("query-executor-%d").build()));
    final long timestamp = System.currentTimeMillis();
    final Interval queryInterval = Intervals.of("1900-01-01T00:00:00Z/2900-01-01T00:00:00Z");
    final List<ListenableFuture<?>> indexFutures = Lists.newArrayListWithExpectedSize(concurrentThreads);
    final List<ListenableFuture<?>> queryFutures = Lists.newArrayListWithExpectedSize(concurrentThreads);
    final Segment incrementalIndexSegment = new IncrementalIndexSegment(index, null);
    final QueryRunnerFactory factory = new TimeseriesQueryRunnerFactory(new TimeseriesQueryQueryToolChest(), new TimeseriesQueryEngine(), QueryRunnerTestHelper.NOOP_QUERYWATCHER);
    final AtomicInteger currentlyRunning = new AtomicInteger(0);
    final AtomicInteger concurrentlyRan = new AtomicInteger(0);
    final AtomicInteger someoneRan = new AtomicInteger(0);
    final CountDownLatch startLatch = new CountDownLatch(1);
    final CountDownLatch readyLatch = new CountDownLatch(concurrentThreads * 2);
    final AtomicInteger queriesAccumualted = new AtomicInteger(0);
    for (int j = 0; j < concurrentThreads; j++) {
        indexFutures.add(indexExecutor.submit(new Runnable() {

            @Override
            public void run() {
                readyLatch.countDown();
                try {
                    startLatch.await();
                } catch (InterruptedException e) {
                    Thread.currentThread().interrupt();
                    throw new RuntimeException(e);
                }
                currentlyRunning.incrementAndGet();
                try {
                    for (int i = 0; i < elementsPerThread; i++) {
                        index.add(getLongRow(timestamp + i, dimensionCount));
                        someoneRan.incrementAndGet();
                    }
                } catch (IndexSizeExceededException e) {
                    throw new RuntimeException(e);
                }
                currentlyRunning.decrementAndGet();
            }
        }));
        final TimeseriesQuery query = Druids.newTimeseriesQueryBuilder().dataSource("xxx").granularity(Granularities.ALL).intervals(ImmutableList.of(queryInterval)).aggregators(queryAggregatorFactories).build();
        queryFutures.add(queryExecutor.submit(new Runnable() {

            @Override
            public void run() {
                readyLatch.countDown();
                try {
                    startLatch.await();
                } catch (InterruptedException e) {
                    Thread.currentThread().interrupt();
                    throw new RuntimeException(e);
                }
                while (concurrentlyRan.get() == 0) {
                    QueryRunner<Result<TimeseriesResultValue>> runner = new FinalizeResultsQueryRunner<Result<TimeseriesResultValue>>(factory.createRunner(incrementalIndexSegment), factory.getToolchest());
                    Sequence<Result<TimeseriesResultValue>> sequence = runner.run(QueryPlus.wrap(query));
                    Double[] results = sequence.accumulate(new Double[0], new Accumulator<Double[], Result<TimeseriesResultValue>>() {

                        @Override
                        public Double[] accumulate(Double[] accumulated, Result<TimeseriesResultValue> in) {
                            if (currentlyRunning.get() > 0) {
                                concurrentlyRan.incrementAndGet();
                            }
                            queriesAccumualted.incrementAndGet();
                            return Lists.asList(in.getValue().getDoubleMetric("doubleSumResult0"), accumulated).toArray(new Double[0]);
                        }
                    });
                    for (Double result : results) {
                        final Integer maxValueExpected = someoneRan.get() + concurrentThreads;
                        if (maxValueExpected > 0) {
                            // Eventually consistent, but should be somewhere in that range
                            // Actual result is validated after all writes are guaranteed done.
                            Assert.assertTrue(StringUtils.format("%d >= %g >= 0 violated", maxValueExpected, result), result >= 0 && result <= maxValueExpected);
                        }
                    }
                }
            }
        }));
    }
    readyLatch.await();
    startLatch.countDown();
    List<ListenableFuture<?>> allFutures = new ArrayList<>(queryFutures.size() + indexFutures.size());
    allFutures.addAll(queryFutures);
    allFutures.addAll(indexFutures);
    Futures.allAsList(allFutures).get();
    Assert.assertTrue("Queries ran too fast", queriesAccumualted.get() > 0);
    Assert.assertTrue("Did not hit concurrency, please try again", concurrentlyRan.get() > 0);
    queryExecutor.shutdown();
    indexExecutor.shutdown();
    QueryRunner<Result<TimeseriesResultValue>> runner = new FinalizeResultsQueryRunner<Result<TimeseriesResultValue>>(factory.createRunner(incrementalIndexSegment), factory.getToolchest());
    TimeseriesQuery query = Druids.newTimeseriesQueryBuilder().dataSource("xxx").granularity(Granularities.ALL).intervals(ImmutableList.of(queryInterval)).aggregators(queryAggregatorFactories).build();
    List<Result<TimeseriesResultValue>> results = runner.run(QueryPlus.wrap(query)).toList();
    boolean isRollup = index.isRollup();
    for (Result<TimeseriesResultValue> result : results) {
        Assert.assertEquals(elementsPerThread * (isRollup ? 1 : concurrentThreads), result.getValue().getLongMetric("rows").intValue());
        for (int i = 0; i < dimensionCount; ++i) {
            Assert.assertEquals(StringUtils.format("Failed long sum on dimension %d", i), elementsPerThread * concurrentThreads, result.getValue().getLongMetric(StringUtils.format("sumResult%s", i)).intValue());
            Assert.assertEquals(StringUtils.format("Failed double sum on dimension %d", i), elementsPerThread * concurrentThreads, result.getValue().getDoubleMetric(StringUtils.format("doubleSumResult%s", i)).intValue());
        }
    }
}