use of org.apache.druid.segment.IncrementalIndexSegment in project druid by druid-io.
the class TopNQueryRunnerBenchmark method setUp.
@BeforeClass
public static void setUp() {
QueryRunnerFactory factory = new TopNQueryRunnerFactory(new StupidPool<ByteBuffer>("TopNQueryRunnerFactory-directBufferPool", new Supplier<ByteBuffer>() {
@Override
public ByteBuffer get() {
// Instead of causing a circular dependency, we simply mimic its behavior
return ByteBuffer.allocateDirect(2000);
}
}), new TopNQueryQueryToolChest(new TopNQueryConfig()), QueryRunnerTestHelper.NOOP_QUERYWATCHER);
TEST_CASE_MAP.put(TestCases.rtIndex, QueryRunnerTestHelper.makeQueryRunner(factory, new IncrementalIndexSegment(TestIndex.getIncrementalTestIndex(), SEGMENT_ID), null));
TEST_CASE_MAP.put(TestCases.mMappedTestIndex, QueryRunnerTestHelper.makeQueryRunner(factory, new QueryableIndexSegment(TestIndex.getMMappedTestIndex(), SEGMENT_ID), null));
TEST_CASE_MAP.put(TestCases.mergedRealtimeIndex, QueryRunnerTestHelper.makeQueryRunner(factory, new QueryableIndexSegment(TestIndex.mergedRealtimeIndex(), SEGMENT_ID), null));
// Thread.sleep(10000);
}
use of org.apache.druid.segment.IncrementalIndexSegment in project druid by druid-io.
the class TimeseriesQueryRunnerBonusTest method runTimeseriesCount.
private List<Result<TimeseriesResultValue>> runTimeseriesCount(IncrementalIndex index) {
final QueryRunnerFactory factory = new TimeseriesQueryRunnerFactory(new TimeseriesQueryQueryToolChest(), new TimeseriesQueryEngine(), QueryRunnerTestHelper.NOOP_QUERYWATCHER);
final QueryRunner<Result<TimeseriesResultValue>> runner = makeQueryRunner(factory, new IncrementalIndexSegment(index, SegmentId.dummy("ds")));
TimeseriesQuery query = Druids.newTimeseriesQueryBuilder().dataSource("xxx").granularity(Granularities.ALL).intervals(ImmutableList.of(Intervals.of("2012-01-01T00:00:00Z/P1D"))).aggregators(new CountAggregatorFactory("rows")).descending(descending).build();
return runner.run(QueryPlus.wrap(query)).toList();
}
use of org.apache.druid.segment.IncrementalIndexSegment in project druid by druid-io.
the class SearchQueryRunnerWithCaseTest method constructorFeeder.
@Parameterized.Parameters
public static Iterable<Object[]> constructorFeeder() {
final SearchQueryConfig[] configs = new SearchQueryConfig[3];
configs[0] = new SearchQueryConfig();
configs[0].setSearchStrategy(UseIndexesStrategy.NAME);
configs[1] = new SearchQueryConfig();
configs[1].setSearchStrategy(CursorOnlyStrategy.NAME);
configs[2] = new SearchQueryConfig();
configs[2].setSearchStrategy(AutoStrategy.NAME);
CharSource input = CharSource.wrap("2011-01-12T00:00:00.000Z\tspot\tAutoMotive\t1000\t10000.0\t10000.0\t100000\t10\t10.0\t10.0\tPREFERRED\ta\u0001preferred\t100.000000\n" + "2011-01-12T00:00:00.000Z\tSPot\tbusiness\t1100\t11000.0\t11000.0\t110000\t20\t20.0\t20.0\tpreferred\tb\u0001Preferred\t100.000000\n" + "2011-01-12T00:00:00.000Z\tspot\tentertainment\t1200\t12000.0\t12000.0\t120000\t\t\t\tPREFERRed\te\u0001preferred\t100.000000\n" + "2011-01-13T00:00:00.000Z\tspot\tautomotive\t1000\t10000.0\t10000.0\t100000\t10\t10.0\t10.0\tpreferred\ta\u0001preferred\t94.874713");
IncrementalIndex index1 = TestIndex.makeRealtimeIndex(input);
IncrementalIndex index2 = TestIndex.makeRealtimeIndex(input);
QueryableIndex index3 = TestIndex.persistRealtimeAndLoadMMapped(index1);
QueryableIndex index4 = TestIndex.persistRealtimeAndLoadMMapped(index2);
final List<QueryRunner<Result<SearchResultValue>>> runners = new ArrayList<>();
for (SearchQueryConfig config : configs) {
runners.addAll(Arrays.asList(QueryRunnerTestHelper.makeQueryRunner(makeRunnerFactory(config), SegmentId.dummy("index1"), new IncrementalIndexSegment(index1, SegmentId.dummy("index1")), "index1"), QueryRunnerTestHelper.makeQueryRunner(makeRunnerFactory(config), SegmentId.dummy("index2"), new IncrementalIndexSegment(index2, SegmentId.dummy("index2")), "index2"), QueryRunnerTestHelper.makeQueryRunner(makeRunnerFactory(config), SegmentId.dummy("index3"), new QueryableIndexSegment(index3, SegmentId.dummy("index3")), "index3"), QueryRunnerTestHelper.makeQueryRunner(makeRunnerFactory(config), SegmentId.dummy("index4"), new QueryableIndexSegment(index4, SegmentId.dummy("index4")), "index4")));
}
return QueryRunnerTestHelper.transformToConstructionFeeder(runners);
}
use of org.apache.druid.segment.IncrementalIndexSegment 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());
}
}
}
use of org.apache.druid.segment.IncrementalIndexSegment 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());
}
}
}
Aggregations