Examples with GroupByStrategyV2 org.apache.druid.query.groupby.strategy.GroupByStrategyV2 used on opensource projects

Example 1 with GroupByStrategyV2

use of org.apache.druid.query.groupby.strategy.GroupByStrategyV2 in project druid by druid-io.

the class GroupByLimitPushDownInsufficientBufferTest method setupGroupByFactory.

private void setupGroupByFactory() {
    executorService = Execs.multiThreaded(3, "GroupByThreadPool[%d]");
    final CloseableStupidPool<ByteBuffer> bufferPool = new CloseableStupidPool<>("GroupByBenchmark-computeBufferPool", new OffheapBufferGenerator("compute", 10_000_000), 0, Integer.MAX_VALUE);
    // limit of 2 is required since we simulate both historical merge and broker merge in the same process
    final CloseableDefaultBlockingPool<ByteBuffer> mergePool = new CloseableDefaultBlockingPool<>(new OffheapBufferGenerator("merge", 10_000_000), 2);
    // limit of 2 is required since we simulate both historical merge and broker merge in the same process
    final CloseableDefaultBlockingPool<ByteBuffer> tooSmallMergePool = new CloseableDefaultBlockingPool<>(new OffheapBufferGenerator("merge", 255), 2);
    resourceCloser.register(bufferPool);
    resourceCloser.register(mergePool);
    resourceCloser.register(tooSmallMergePool);
    final GroupByQueryConfig config = new GroupByQueryConfig() {

        @Override
        public String getDefaultStrategy() {
            return "v2";
        }

        @Override
        public int getBufferGrouperInitialBuckets() {
            return -1;
        }

        @Override
        public long getMaxOnDiskStorage() {
            return 1_000_000_000L;
        }
    };
    config.setSingleThreaded(false);
    config.setMaxIntermediateRows(Integer.MAX_VALUE);
    config.setMaxResults(Integer.MAX_VALUE);
    DruidProcessingConfig druidProcessingConfig = new DruidProcessingConfig() {

        @Override
        public int getNumThreads() {
            // Used by "v2" strategy for concurrencyHint
            return 2;
        }

        @Override
        public String getFormatString() {
            return null;
        }
    };
    DruidProcessingConfig tooSmallDruidProcessingConfig = new DruidProcessingConfig() {

        @Override
        public int intermediateComputeSizeBytes() {
            return 255;
        }

        @Override
        public int getNumThreads() {
            // Used by "v2" strategy for concurrencyHint
            return 2;
        }

        @Override
        public String getFormatString() {
            return null;
        }
    };
    final Supplier<GroupByQueryConfig> configSupplier = Suppliers.ofInstance(config);
    final GroupByStrategySelector strategySelector = new GroupByStrategySelector(configSupplier, new GroupByStrategyV1(configSupplier, new GroupByQueryEngine(configSupplier, bufferPool), NOOP_QUERYWATCHER), new GroupByStrategyV2(druidProcessingConfig, configSupplier, bufferPool, mergePool, new ObjectMapper(new SmileFactory()), NOOP_QUERYWATCHER));
    final GroupByStrategySelector tooSmallStrategySelector = new GroupByStrategySelector(configSupplier, new GroupByStrategyV1(configSupplier, new GroupByQueryEngine(configSupplier, bufferPool), NOOP_QUERYWATCHER), new GroupByStrategyV2(tooSmallDruidProcessingConfig, configSupplier, bufferPool, tooSmallMergePool, new ObjectMapper(new SmileFactory()), NOOP_QUERYWATCHER));
    groupByFactory = new GroupByQueryRunnerFactory(strategySelector, new GroupByQueryQueryToolChest(strategySelector));
    tooSmallGroupByFactory = new GroupByQueryRunnerFactory(tooSmallStrategySelector, new GroupByQueryQueryToolChest(tooSmallStrategySelector));
}

Example 2 with GroupByStrategyV2

use of org.apache.druid.query.groupby.strategy.GroupByStrategyV2 in project druid by druid-io.

the class GroupByMultiSegmentTest method setupGroupByFactory.

private void setupGroupByFactory() {
    executorService = Execs.multiThreaded(2, "GroupByThreadPool[%d]");
    final CloseableStupidPool<ByteBuffer> bufferPool = new CloseableStupidPool<>("GroupByBenchmark-computeBufferPool", new OffheapBufferGenerator("compute", 10_000_000), 0, Integer.MAX_VALUE);
    // limit of 2 is required since we simulate both historical merge and broker merge in the same process
    final CloseableDefaultBlockingPool<ByteBuffer> mergePool = new CloseableDefaultBlockingPool<>(new OffheapBufferGenerator("merge", 10_000_000), 2);
    resourceCloser.register(bufferPool);
    resourceCloser.register(mergePool);
    final GroupByQueryConfig config = new GroupByQueryConfig() {

        @Override
        public String getDefaultStrategy() {
            return "v2";
        }

        @Override
        public int getBufferGrouperInitialBuckets() {
            return -1;
        }

        @Override
        public long getMaxOnDiskStorage() {
            return 1_000_000_000L;
        }
    };
    config.setSingleThreaded(false);
    config.setMaxIntermediateRows(Integer.MAX_VALUE);
    config.setMaxResults(Integer.MAX_VALUE);
    DruidProcessingConfig druidProcessingConfig = new DruidProcessingConfig() {

        @Override
        public int getNumThreads() {
            // Used by "v2" strategy for concurrencyHint
            return 2;
        }

        @Override
        public String getFormatString() {
            return null;
        }
    };
    final Supplier<GroupByQueryConfig> configSupplier = Suppliers.ofInstance(config);
    final GroupByStrategySelector strategySelector = new GroupByStrategySelector(configSupplier, new GroupByStrategyV1(configSupplier, new GroupByQueryEngine(configSupplier, bufferPool), NOOP_QUERYWATCHER), new GroupByStrategyV2(druidProcessingConfig, configSupplier, bufferPool, mergePool, new ObjectMapper(new SmileFactory()), NOOP_QUERYWATCHER));
    groupByFactory = new GroupByQueryRunnerFactory(strategySelector, new GroupByQueryQueryToolChest(strategySelector));
}

Example 3 with GroupByStrategyV2

use of org.apache.druid.query.groupby.strategy.GroupByStrategyV2 in project druid by druid-io.

the class NestedQueryPushDownTest method setupGroupByFactory.

private void setupGroupByFactory() {
    executorService = Execs.multiThreaded(3, "GroupByThreadPool[%d]");
    NonBlockingPool<ByteBuffer> bufferPool = new StupidPool<>("GroupByBenchmark-computeBufferPool", new OffheapBufferGenerator("compute", 10_000_000), 0, Integer.MAX_VALUE);
    // limit of 3 is required since we simulate running historical running nested query and broker doing the final merge
    BlockingPool<ByteBuffer> mergePool = new DefaultBlockingPool<>(new OffheapBufferGenerator("merge", 10_000_000), 10);
    // limit of 3 is required since we simulate running historical running nested query and broker doing the final merge
    BlockingPool<ByteBuffer> mergePool2 = new DefaultBlockingPool<>(new OffheapBufferGenerator("merge", 10_000_000), 10);
    final GroupByQueryConfig config = new GroupByQueryConfig() {

        @Override
        public String getDefaultStrategy() {
            return "v2";
        }

        @Override
        public int getBufferGrouperInitialBuckets() {
            return -1;
        }

        @Override
        public long getMaxOnDiskStorage() {
            return 1_000_000_000L;
        }
    };
    config.setSingleThreaded(false);
    config.setMaxIntermediateRows(Integer.MAX_VALUE);
    config.setMaxResults(Integer.MAX_VALUE);
    DruidProcessingConfig druidProcessingConfig = new DruidProcessingConfig() {

        @Override
        public int getNumThreads() {
            // Used by "v2" strategy for concurrencyHint
            return 2;
        }

        @Override
        public String getFormatString() {
            return null;
        }
    };
    final Supplier<GroupByQueryConfig> configSupplier = Suppliers.ofInstance(config);
    final GroupByStrategySelector strategySelector = new GroupByStrategySelector(configSupplier, new GroupByStrategyV1(configSupplier, new GroupByQueryEngine(configSupplier, bufferPool), NOOP_QUERYWATCHER), new GroupByStrategyV2(druidProcessingConfig, configSupplier, bufferPool, mergePool, new ObjectMapper(new SmileFactory()), NOOP_QUERYWATCHER));
    final GroupByStrategySelector strategySelector2 = new GroupByStrategySelector(configSupplier, new GroupByStrategyV1(configSupplier, new GroupByQueryEngine(configSupplier, bufferPool), NOOP_QUERYWATCHER), new GroupByStrategyV2(druidProcessingConfig, configSupplier, bufferPool, mergePool2, new ObjectMapper(new SmileFactory()), NOOP_QUERYWATCHER));
    groupByFactory = new GroupByQueryRunnerFactory(strategySelector, new GroupByQueryQueryToolChest(strategySelector));
    groupByFactory2 = new GroupByQueryRunnerFactory(strategySelector2, new GroupByQueryQueryToolChest(strategySelector2));
}

Example 4 with GroupByStrategyV2

use of org.apache.druid.query.groupby.strategy.GroupByStrategyV2 in project druid by druid-io.

the class MapVirtualColumnGroupByTest method setup.

@Before
public void setup() throws IOException {
    final IncrementalIndex incrementalIndex = MapVirtualColumnTestBase.generateIndex();
    final GroupByStrategySelector strategySelector = new GroupByStrategySelector(GroupByQueryConfig::new, null, new GroupByStrategyV2(new DruidProcessingConfig() {

        @Override
        public String getFormatString() {
            return null;
        }

        @Override
        public int intermediateComputeSizeBytes() {
            return 10 * 1024 * 1024;
        }

        @Override
        public int getNumMergeBuffers() {
            return 1;
        }

        @Override
        public int getNumThreads() {
            return 1;
        }
    }, GroupByQueryConfig::new, new StupidPool<>("map-virtual-column-groupby-test", () -> ByteBuffer.allocate(1024)), new DefaultBlockingPool<>(() -> ByteBuffer.allocate(1024), 1), new DefaultObjectMapper(), QueryRunnerTestHelper.NOOP_QUERYWATCHER));
    final GroupByQueryRunnerFactory factory = new GroupByQueryRunnerFactory(strategySelector, new GroupByQueryQueryToolChest(strategySelector));
    runner = QueryRunnerTestHelper.makeQueryRunner(factory, SegmentId.dummy("index"), new IncrementalIndexSegment(incrementalIndex, SegmentId.dummy("index")), "incremental");
}

Example 5 with GroupByStrategyV2

use of org.apache.druid.query.groupby.strategy.GroupByStrategyV2 in project druid by druid-io.

the class GroupByTypeInterfaceBenchmark method setup.

@Setup(Level.Trial)
public void setup() throws IOException {
    log.info("SETUP CALLED AT %d", System.currentTimeMillis());
    ComplexMetrics.registerSerde("hyperUnique", new HyperUniquesSerde());
    setupQueries();
    String schemaName = "basic";
    schemaInfo = GeneratorBasicSchemas.SCHEMA_MAP.get(schemaName);
    stringQuery = SCHEMA_QUERY_MAP.get(schemaName).get("string");
    longFloatQuery = SCHEMA_QUERY_MAP.get(schemaName).get("longFloat");
    longQuery = SCHEMA_QUERY_MAP.get(schemaName).get("long");
    floatQuery = SCHEMA_QUERY_MAP.get(schemaName).get("float");
    final DataGenerator dataGenerator = new DataGenerator(schemaInfo.getColumnSchemas(), RNG_SEED + 1, schemaInfo.getDataInterval(), rowsPerSegment);
    tmpDir = FileUtils.createTempDir();
    log.info("Using temp dir: %s", tmpDir.getAbsolutePath());
    // queryableIndexes   -> numSegments worth of on-disk segments
    // anIncrementalIndex -> the last incremental index
    anIncrementalIndex = null;
    queryableIndexes = new ArrayList<>(numSegments);
    for (int i = 0; i < numSegments; i++) {
        log.info("Generating rows for segment %d/%d", i + 1, numSegments);
        final IncrementalIndex index = makeIncIndex();
        for (int j = 0; j < rowsPerSegment; j++) {
            final InputRow row = dataGenerator.nextRow();
            if (j % 20000 == 0) {
                log.info("%,d/%,d rows generated.", i * rowsPerSegment + j, rowsPerSegment * numSegments);
            }
            index.add(row);
        }
        log.info("%,d/%,d rows generated, persisting segment %d/%d.", (i + 1) * rowsPerSegment, rowsPerSegment * numSegments, i + 1, numSegments);
        final File file = INDEX_MERGER_V9.persist(index, new File(tmpDir, String.valueOf(i)), new IndexSpec(), null);
        queryableIndexes.add(INDEX_IO.loadIndex(file));
        if (i == numSegments - 1) {
            anIncrementalIndex = index;
        } else {
            index.close();
        }
    }
    NonBlockingPool<ByteBuffer> bufferPool = new StupidPool<>("GroupByBenchmark-computeBufferPool", new OffheapBufferGenerator("compute", 250_000_000), 0, Integer.MAX_VALUE);
    // limit of 2 is required since we simulate both historical merge and broker merge in the same process
    BlockingPool<ByteBuffer> mergePool = new DefaultBlockingPool<>(new OffheapBufferGenerator("merge", 250_000_000), 2);
    final GroupByQueryConfig config = new GroupByQueryConfig() {

        @Override
        public String getDefaultStrategy() {
            return defaultStrategy;
        }

        @Override
        public int getBufferGrouperInitialBuckets() {
            return initialBuckets;
        }

        @Override
        public long getMaxOnDiskStorage() {
            return 1_000_000_000L;
        }
    };
    config.setSingleThreaded(false);
    config.setMaxIntermediateRows(Integer.MAX_VALUE);
    config.setMaxResults(Integer.MAX_VALUE);
    DruidProcessingConfig druidProcessingConfig = new DruidProcessingConfig() {

        @Override
        public int getNumThreads() {
            // Used by "v2" strategy for concurrencyHint
            return numProcessingThreads;
        }

        @Override
        public String getFormatString() {
            return null;
        }
    };
    final Supplier<GroupByQueryConfig> configSupplier = Suppliers.ofInstance(config);
    final GroupByStrategySelector strategySelector = new GroupByStrategySelector(configSupplier, new GroupByStrategyV1(configSupplier, new GroupByQueryEngine(configSupplier, bufferPool), QueryBenchmarkUtil.NOOP_QUERYWATCHER), new GroupByStrategyV2(druidProcessingConfig, configSupplier, bufferPool, mergePool, new ObjectMapper(new SmileFactory()), QueryBenchmarkUtil.NOOP_QUERYWATCHER));
    factory = new GroupByQueryRunnerFactory(strategySelector, new GroupByQueryQueryToolChest(strategySelector));
}