Search in sources :

Example 6 with TestingGcMonitor

use of io.airlift.stats.TestingGcMonitor in project trino by trinodb.

the class TestSqlTask method createInitialTask.

private SqlTask createInitialTask() {
    TaskId taskId = new TaskId(new StageId("query", 0), nextTaskId.incrementAndGet(), 0);
    URI location = URI.create("fake://task/" + taskId);
    QueryContext queryContext = new QueryContext(new QueryId("query"), DataSize.of(1, MEGABYTE), new MemoryPool(DataSize.of(1, GIGABYTE)), new TestingGcMonitor(), taskNotificationExecutor, driverYieldExecutor, DataSize.of(1, MEGABYTE), new SpillSpaceTracker(DataSize.of(1, GIGABYTE)));
    queryContext.addTaskContext(new TaskStateMachine(taskId, taskNotificationExecutor), testSessionBuilder().build(), () -> {
    }, false, false);
    return createSqlTask(taskId, location, "fake", queryContext, sqlTaskExecutionFactory, taskNotificationExecutor, sqlTask -> {
    }, DataSize.of(32, MEGABYTE), DataSize.of(200, MEGABYTE), new ExchangeManagerRegistry(new ExchangeHandleResolver()), new CounterStat());
}
Also used : SpillSpaceTracker(io.trino.spiller.SpillSpaceTracker) CounterStat(io.airlift.stats.CounterStat) QueryId(io.trino.spi.QueryId) TestingGcMonitor(io.airlift.stats.TestingGcMonitor) QueryContext(io.trino.memory.QueryContext) URI(java.net.URI) ExchangeManagerRegistry(io.trino.exchange.ExchangeManagerRegistry) ExchangeHandleResolver(io.trino.metadata.ExchangeHandleResolver) MemoryPool(io.trino.memory.MemoryPool)

Example 7 with TestingGcMonitor

use of io.airlift.stats.TestingGcMonitor in project trino by trinodb.

the class GroupByHashYieldAssertion method finishOperatorWithYieldingGroupByHash.

/**
 * @param operatorFactory creates an Operator that should directly or indirectly contain GroupByHash
 * @param getHashCapacity returns the hash table capacity for the input operator
 * @param additionalMemoryInBytes the memory used in addition to the GroupByHash in the operator (e.g., aggregator)
 */
public static GroupByHashYieldResult finishOperatorWithYieldingGroupByHash(List<Page> input, Type hashKeyType, OperatorFactory operatorFactory, Function<Operator, Integer> getHashCapacity, long additionalMemoryInBytes) {
    assertLessThan(additionalMemoryInBytes, 1L << 21, "additionalMemoryInBytes should be a relatively small number");
    List<Page> result = new LinkedList<>();
    // mock an adjustable memory pool
    QueryId queryId = new QueryId("test_query");
    MemoryPool memoryPool = new MemoryPool(DataSize.of(1, GIGABYTE));
    QueryContext queryContext = new QueryContext(queryId, DataSize.of(512, MEGABYTE), memoryPool, new TestingGcMonitor(), EXECUTOR, SCHEDULED_EXECUTOR, DataSize.of(512, MEGABYTE), new SpillSpaceTracker(DataSize.of(512, MEGABYTE)));
    DriverContext driverContext = createTaskContext(queryContext, EXECUTOR, TEST_SESSION).addPipelineContext(0, true, true, false).addDriverContext();
    Operator operator = operatorFactory.createOperator(driverContext);
    // run operator
    int yieldCount = 0;
    long expectedReservedExtraBytes = 0;
    for (Page page : input) {
        // unblocked
        assertTrue(operator.needsInput());
        // saturate the pool with a tiny memory left
        long reservedMemoryInBytes = memoryPool.getFreeBytes() - additionalMemoryInBytes;
        memoryPool.reserve(queryId, "test", reservedMemoryInBytes);
        long oldMemoryUsage = operator.getOperatorContext().getDriverContext().getMemoryUsage();
        int oldCapacity = getHashCapacity.apply(operator);
        // add a page and verify different behaviors
        operator.addInput(page);
        // get output to consume the input
        Page output = operator.getOutput();
        if (output != null) {
            result.add(output);
        }
        long newMemoryUsage = operator.getOperatorContext().getDriverContext().getMemoryUsage();
        // between rehash and memory used by aggregator
        if (newMemoryUsage < DataSize.of(4, MEGABYTE).toBytes()) {
            // free the pool for the next iteration
            memoryPool.free(queryId, "test", reservedMemoryInBytes);
            // this required in case input is blocked
            operator.getOutput();
            continue;
        }
        long actualIncreasedMemory = newMemoryUsage - oldMemoryUsage;
        if (operator.needsInput()) {
            // We have successfully added a page
            // Assert we are not blocked
            assertTrue(operator.getOperatorContext().isWaitingForMemory().isDone());
            // assert the hash capacity is not changed; otherwise, we should have yielded
            assertTrue(oldCapacity == getHashCapacity.apply(operator));
            // We are not going to rehash; therefore, assert the memory increase only comes from the aggregator
            assertLessThan(actualIncreasedMemory, additionalMemoryInBytes);
            // free the pool for the next iteration
            memoryPool.free(queryId, "test", reservedMemoryInBytes);
        } else {
            // We failed to finish the page processing i.e. we yielded
            yieldCount++;
            // Assert we are blocked
            assertFalse(operator.getOperatorContext().isWaitingForMemory().isDone());
            // Hash table capacity should not change
            assertEquals(oldCapacity, (long) getHashCapacity.apply(operator));
            // Increased memory is no smaller than the hash table size and no greater than the hash table size + the memory used by aggregator
            if (hashKeyType == BIGINT) {
                // groupIds and values double by hashCapacity; while valuesByGroupId double by maxFill = hashCapacity / 0.75
                expectedReservedExtraBytes = oldCapacity * (long) (Long.BYTES * 1.75 + Integer.BYTES) + page.getRetainedSizeInBytes();
            } else {
                // groupAddressByHash, groupIdsByHash, and rawHashByHashPosition double by hashCapacity; while groupAddressByGroupId double by maxFill = hashCapacity / 0.75
                expectedReservedExtraBytes = oldCapacity * (long) (Long.BYTES * 1.75 + Integer.BYTES + Byte.BYTES) + page.getRetainedSizeInBytes();
            }
            assertBetweenInclusive(actualIncreasedMemory, expectedReservedExtraBytes, expectedReservedExtraBytes + additionalMemoryInBytes);
            // Output should be blocked as well
            assertNull(operator.getOutput());
            // Free the pool to unblock
            memoryPool.free(queryId, "test", reservedMemoryInBytes);
            // Trigger a process through getOutput() or needsInput()
            output = operator.getOutput();
            if (output != null) {
                result.add(output);
            }
            assertTrue(operator.needsInput());
            // Hash table capacity has increased
            assertGreaterThan(getHashCapacity.apply(operator), oldCapacity);
            // Assert the estimated reserved memory before rehash is very close to the one after rehash
            long rehashedMemoryUsage = operator.getOperatorContext().getDriverContext().getMemoryUsage();
            assertBetweenInclusive(rehashedMemoryUsage * 1.0 / newMemoryUsage, 0.99, 1.01);
            // unblocked
            assertTrue(operator.needsInput());
        }
    }
    result.addAll(finishOperator(operator));
    return new GroupByHashYieldResult(yieldCount, expectedReservedExtraBytes, result);
}
Also used : OperatorAssertion.finishOperator(io.trino.operator.OperatorAssertion.finishOperator) SpillSpaceTracker(io.trino.spiller.SpillSpaceTracker) QueryId(io.trino.spi.QueryId) Page(io.trino.spi.Page) QueryContext(io.trino.memory.QueryContext) LinkedList(java.util.LinkedList) TestingGcMonitor(io.airlift.stats.TestingGcMonitor) MemoryPool(io.trino.memory.MemoryPool)

Example 8 with TestingGcMonitor

use of io.airlift.stats.TestingGcMonitor in project hetu-core by openlookeng.

the class GroupByHashYieldAssertion method finishOperatorWithYieldingGroupByHash.

/**
 * @param operatorFactory creates an Operator that should directly or indirectly contain GroupByHash
 * @param getHashCapacity returns the hash table capacity for the input operator
 * @param additionalMemoryInBytes the memory used in addition to the GroupByHash in the operator (e.g., aggregator)
 */
public static GroupByHashYieldResult finishOperatorWithYieldingGroupByHash(List<Page> input, Type hashKeyType, OperatorFactory operatorFactory, Function<Operator, Integer> getHashCapacity, long additionalMemoryInBytes) {
    assertLessThan(additionalMemoryInBytes, 1L << 21, "additionalMemoryInBytes should be a relatively small number");
    List<Page> result = new LinkedList<>();
    // mock an adjustable memory pool
    QueryId queryId = new QueryId("test_query");
    MemoryPool memoryPool = new MemoryPool(new MemoryPoolId("test"), new DataSize(1, GIGABYTE));
    QueryContext queryContext = new QueryContext(queryId, new DataSize(512, MEGABYTE), new DataSize(1024, MEGABYTE), memoryPool, new TestingGcMonitor(), EXECUTOR, SCHEDULED_EXECUTOR, new DataSize(512, MEGABYTE), new SpillSpaceTracker(new DataSize(512, MEGABYTE)), NOOP_SNAPSHOT_UTILS);
    DriverContext driverContext = createTaskContext(queryContext, EXECUTOR, TEST_SESSION).addPipelineContext(0, true, true, false).addDriverContext();
    Operator operator = operatorFactory.createOperator(driverContext);
    // run operator
    int yieldCount = 0;
    long expectedReservedExtraBytes = 0;
    for (Page page : input) {
        // unblocked
        assertTrue(operator.needsInput());
        // saturate the pool with a tiny memory left
        long reservedMemoryInBytes = memoryPool.getFreeBytes() - additionalMemoryInBytes;
        memoryPool.reserve(queryId, "test", reservedMemoryInBytes);
        long oldMemoryUsage = operator.getOperatorContext().getDriverContext().getMemoryUsage();
        int oldCapacity = getHashCapacity.apply(operator);
        // add a page and verify different behaviors
        operator.addInput(page);
        // get output to consume the input
        Page output = operator.getOutput();
        if (output != null) {
            result.add(output);
        }
        long newMemoryUsage = operator.getOperatorContext().getDriverContext().getMemoryUsage();
        // between rehash and memory used by aggregator
        if (newMemoryUsage < new DataSize(4, MEGABYTE).toBytes()) {
            // free the pool for the next iteration
            memoryPool.free(queryId, "test", reservedMemoryInBytes);
            // this required in case input is blocked
            operator.getOutput();
            continue;
        }
        long actualIncreasedMemory = newMemoryUsage - oldMemoryUsage;
        if (operator.needsInput()) {
            // We have successfully added a page
            // Assert we are not blocked
            assertTrue(operator.getOperatorContext().isWaitingForMemory().isDone());
            // assert the hash capacity is not changed; otherwise, we should have yielded
            assertTrue(oldCapacity == getHashCapacity.apply(operator));
            // We are not going to rehash; therefore, assert the memory increase only comes from the aggregator
            assertLessThan(actualIncreasedMemory, additionalMemoryInBytes);
            // free the pool for the next iteration
            memoryPool.free(queryId, "test", reservedMemoryInBytes);
        } else {
            // We failed to finish the page processing i.e. we yielded
            yieldCount++;
            // Assert we are blocked
            assertFalse(operator.getOperatorContext().isWaitingForMemory().isDone());
            // Hash table capacity should not change
            assertEquals(oldCapacity, (long) getHashCapacity.apply(operator));
            // Increased memory is no smaller than the hash table size and no greater than the hash table size + the memory used by aggregator
            if (hashKeyType == BIGINT) {
                // groupIds and values double by hashCapacity; while valuesByGroupId double by maxFill = hashCapacity / 0.75
                expectedReservedExtraBytes = oldCapacity * (long) (Long.BYTES * 1.75 + Integer.BYTES) + page.getRetainedSizeInBytes();
            } else {
                // groupAddressByHash, groupIdsByHash, and rawHashByHashPosition double by hashCapacity; while groupAddressByGroupId double by maxFill = hashCapacity / 0.75
                expectedReservedExtraBytes = oldCapacity * (long) (Long.BYTES * 1.75 + Integer.BYTES + Byte.BYTES) + page.getRetainedSizeInBytes();
            }
            assertBetweenInclusive(actualIncreasedMemory, expectedReservedExtraBytes, expectedReservedExtraBytes + additionalMemoryInBytes);
            // Output should be blocked as well
            assertNull(operator.getOutput());
            // Free the pool to unblock
            memoryPool.free(queryId, "test", reservedMemoryInBytes);
            // Trigger a process through getOutput() or needsInput()
            output = operator.getOutput();
            if (output != null) {
                result.add(output);
            }
            assertTrue(operator.needsInput());
            // Hash table capacity has increased
            assertGreaterThan(getHashCapacity.apply(operator), oldCapacity);
            // Assert the estimated reserved memory before rehash is very close to the one after rehash
            long rehashedMemoryUsage = operator.getOperatorContext().getDriverContext().getMemoryUsage();
            assertBetweenInclusive(rehashedMemoryUsage * 1.0 / newMemoryUsage, 0.99, 1.01);
            // unblocked
            assertTrue(operator.needsInput());
        }
    }
    result.addAll(finishOperator(operator));
    return new GroupByHashYieldResult(yieldCount, expectedReservedExtraBytes, result);
}
Also used : OperatorAssertion.finishOperator(io.prestosql.operator.OperatorAssertion.finishOperator) SpillSpaceTracker(io.prestosql.spiller.SpillSpaceTracker) QueryId(io.prestosql.spi.QueryId) Page(io.prestosql.spi.Page) QueryContext(io.prestosql.memory.QueryContext) LinkedList(java.util.LinkedList) DataSize(io.airlift.units.DataSize) TestingGcMonitor(io.airlift.stats.TestingGcMonitor) MemoryPoolId(io.prestosql.spi.memory.MemoryPoolId) MemoryPool(io.prestosql.memory.MemoryPool)

Example 9 with TestingGcMonitor

use of io.airlift.stats.TestingGcMonitor in project hetu-core by openlookeng.

the class AbstractOperatorBenchmark method runOnce.

@Override
protected Map<String, Long> runOnce() {
    Session setSession = testSessionBuilder().setSystemProperty("optimizer.optimize-hash-generation", "true").build();
    MemoryPool memoryPool = new MemoryPool(new MemoryPoolId("test"), new DataSize(1, GIGABYTE));
    SpillSpaceTracker spillSpaceTracker = new SpillSpaceTracker(new DataSize(1, GIGABYTE));
    TaskContext taskContext = new QueryContext(new QueryId("test"), new DataSize(256, MEGABYTE), new DataSize(512, MEGABYTE), memoryPool, new TestingGcMonitor(), localQueryRunner.getExecutor(), localQueryRunner.getScheduler(), new DataSize(256, MEGABYTE), spillSpaceTracker, NOOP_SNAPSHOT_UTILS).addTaskContext(new TaskStateMachine(new TaskId("query", 0, 0), localQueryRunner.getExecutor()), setSession, false, false, OptionalInt.empty(), Optional.empty(), TESTING_SERDE_FACTORY);
    CpuTimer cpuTimer = new CpuTimer();
    Map<String, Long> executionStats = execute(taskContext);
    CpuDuration executionTime = cpuTimer.elapsedTime();
    TaskStats taskStats = taskContext.getTaskStats();
    long inputRows = taskStats.getRawInputPositions();
    long inputBytes = taskStats.getRawInputDataSize().toBytes();
    long outputRows = taskStats.getOutputPositions();
    long outputBytes = taskStats.getOutputDataSize().toBytes();
    double inputMegaBytes = new DataSize(inputBytes, BYTE).getValue(MEGABYTE);
    return ImmutableMap.<String, Long>builder().putAll(executionStats).put("elapsed_millis", executionTime.getWall().toMillis()).put("input_rows_per_second", (long) (inputRows / executionTime.getWall().getValue(SECONDS))).put("output_rows_per_second", (long) (outputRows / executionTime.getWall().getValue(SECONDS))).put("input_megabytes", (long) inputMegaBytes).put("input_megabytes_per_second", (long) (inputMegaBytes / executionTime.getWall().getValue(SECONDS))).put("wall_nanos", executionTime.getWall().roundTo(NANOSECONDS)).put("cpu_nanos", executionTime.getCpu().roundTo(NANOSECONDS)).put("user_nanos", executionTime.getUser().roundTo(NANOSECONDS)).put("input_rows", inputRows).put("input_bytes", inputBytes).put("output_rows", outputRows).put("output_bytes", outputBytes).build();
}
Also used : SpillSpaceTracker(io.prestosql.spiller.SpillSpaceTracker) TaskContext(io.prestosql.operator.TaskContext) TaskId(io.prestosql.execution.TaskId) QueryId(io.prestosql.spi.QueryId) QueryContext(io.prestosql.memory.QueryContext) TaskStats(io.prestosql.operator.TaskStats) TaskStateMachine(io.prestosql.execution.TaskStateMachine) DataSize(io.airlift.units.DataSize) TestingGcMonitor(io.airlift.stats.TestingGcMonitor) CpuTimer(io.airlift.stats.CpuTimer) CpuDuration(io.airlift.stats.CpuTimer.CpuDuration) MemoryPoolId(io.prestosql.spi.memory.MemoryPoolId) Session(io.prestosql.Session) MemoryPool(io.prestosql.memory.MemoryPool)

Example 10 with TestingGcMonitor

use of io.airlift.stats.TestingGcMonitor in project hetu-core by openlookeng.

the class TestMemoryPools method setUp.

private void setUp(Supplier<List<Driver>> driversSupplier) {
    checkState(localQueryRunner == null, "Already set up");
    Session session = testSessionBuilder().setCatalog("tpch").setSchema("tiny").setSystemProperty("task_default_concurrency", "1").build();
    localQueryRunner = queryRunnerWithInitialTransaction(session);
    // add tpch
    localQueryRunner.createCatalog("tpch", new TpchConnectorFactory(1), ImmutableMap.of());
    userPool = new MemoryPool(new MemoryPoolId("test"), TEN_MEGABYTES);
    fakeQueryId = new QueryId("fake");
    SpillSpaceTracker spillSpaceTracker = new SpillSpaceTracker(new DataSize(1, GIGABYTE));
    QueryContext queryContext = new QueryContext(new QueryId("query"), TEN_MEGABYTES, new DataSize(20, MEGABYTE), userPool, new TestingGcMonitor(), localQueryRunner.getExecutor(), localQueryRunner.getScheduler(), TEN_MEGABYTES, spillSpaceTracker, NOOP_SNAPSHOT_UTILS);
    taskContext = createTaskContext(queryContext, localQueryRunner.getExecutor(), session);
    drivers = driversSupplier.get();
}
Also used : TpchConnectorFactory(io.prestosql.plugin.tpch.TpchConnectorFactory) SpillSpaceTracker(io.prestosql.spiller.SpillSpaceTracker) QueryId(io.prestosql.spi.QueryId) DataSize(io.airlift.units.DataSize) TestingGcMonitor(io.airlift.stats.TestingGcMonitor) MemoryPoolId(io.prestosql.spi.memory.MemoryPoolId) Session(io.prestosql.Session)

Aggregations

TestingGcMonitor (io.airlift.stats.TestingGcMonitor)16 DataSize (io.airlift.units.DataSize)9 QueryId (io.prestosql.spi.QueryId)9 MemoryPoolId (io.prestosql.spi.memory.MemoryPoolId)7 SpillSpaceTracker (io.prestosql.spiller.SpillSpaceTracker)7 QueryId (io.trino.spi.QueryId)7 QueryContext (io.prestosql.memory.QueryContext)6 SpillSpaceTracker (io.trino.spiller.SpillSpaceTracker)6 MemoryPool (io.prestosql.memory.MemoryPool)5 TaskId (io.prestosql.execution.TaskId)4 TaskStateMachine (io.prestosql.execution.TaskStateMachine)4 MemoryPool (io.trino.memory.MemoryPool)4 QueryContext (io.trino.memory.QueryContext)4 CounterStat (io.airlift.stats.CounterStat)3 Session (io.prestosql.Session)3 TaskContext (io.prestosql.operator.TaskContext)3 Session (io.trino.Session)3 StageId (io.trino.execution.StageId)3 TaskId (io.trino.execution.TaskId)3 TaskStateMachine (io.trino.execution.TaskStateMachine)3