Examples with NodeSchedulerConfig io.trino.execution.scheduler.NodeSchedulerConfig used on opensource projects

Example 1 with NodeSchedulerConfig

use of io.trino.execution.scheduler.NodeSchedulerConfig in project trino by trinodb.

the class TaskTestUtils method createTestingPlanner.

public static LocalExecutionPlanner createTestingPlanner() {
    PageSourceManager pageSourceManager = new PageSourceManager();
    pageSourceManager.addConnectorPageSourceProvider(CONNECTOR_ID, new TestingPageSourceProvider());
    // we don't start the finalizer so nothing will be collected, which is ok for a test
    FinalizerService finalizerService = new FinalizerService();
    BlockTypeOperators blockTypeOperators = new BlockTypeOperators(PLANNER_CONTEXT.getTypeOperators());
    NodeScheduler nodeScheduler = new NodeScheduler(new UniformNodeSelectorFactory(new InMemoryNodeManager(), new NodeSchedulerConfig().setIncludeCoordinator(true), new NodeTaskMap(finalizerService)));
    NodePartitioningManager nodePartitioningManager = new NodePartitioningManager(nodeScheduler, blockTypeOperators);
    PageFunctionCompiler pageFunctionCompiler = new PageFunctionCompiler(PLANNER_CONTEXT.getFunctionManager(), 0);
    return new LocalExecutionPlanner(PLANNER_CONTEXT, createTestingTypeAnalyzer(PLANNER_CONTEXT), Optional.empty(), pageSourceManager, new IndexManager(), nodePartitioningManager, new PageSinkManager(), new MockDirectExchangeClientSupplier(), new ExpressionCompiler(PLANNER_CONTEXT.getFunctionManager(), pageFunctionCompiler), pageFunctionCompiler, new JoinFilterFunctionCompiler(PLANNER_CONTEXT.getFunctionManager()), new IndexJoinLookupStats(), new TaskManagerConfig(), new GenericSpillerFactory((types, spillContext, memoryContext) -> {
        throw new UnsupportedOperationException();
    }), (types, spillContext, memoryContext) -> {
        throw new UnsupportedOperationException();
    }, (types, partitionFunction, spillContext, memoryContext) -> {
        throw new UnsupportedOperationException();
    }, new PagesIndex.TestingFactory(false), new JoinCompiler(PLANNER_CONTEXT.getTypeOperators()), new TrinoOperatorFactories(), new OrderingCompiler(PLANNER_CONTEXT.getTypeOperators()), new DynamicFilterConfig(), blockTypeOperators, new TableExecuteContextManager(), new ExchangeManagerRegistry(new ExchangeHandleResolver()));
}

Example 2 with NodeSchedulerConfig

use of io.trino.execution.scheduler.NodeSchedulerConfig in project trino by trinodb.

the class TestNodeSchedulerConfig method testExplicitPropertyMappings.

@Test
public void testExplicitPropertyMappings() {
    Map<String, String> properties = ImmutableMap.<String, String>builder().put("node-scheduler.policy", "topology").put("node-scheduler.min-candidates", "11").put("node-scheduler.include-coordinator", "false").put("node-scheduler.max-pending-splits-per-task", "11").put("node-scheduler.max-splits-per-node", "101").put("node-scheduler.max-unacknowledged-splits-per-task", "501").put("node-scheduler.splits-balancing-policy", "node").put("node-scheduler.optimized-local-scheduling", "false").put("node-scheduler.max-absolute-full-nodes-per-query", "17").put("node-scheduler.max-fraction-full-nodes-per-query", "0.3").put("node-scheduler.allocator-type", "fixed_count").buildOrThrow();
    NodeSchedulerConfig expected = new NodeSchedulerConfig().setNodeSchedulerPolicy("topology").setIncludeCoordinator(false).setMaxSplitsPerNode(101).setMaxPendingSplitsPerTask(11).setMaxUnacknowledgedSplitsPerTask(501).setMinCandidates(11).setSplitsBalancingPolicy(NODE).setOptimizedLocalScheduling(false).setMaxAbsoluteFullNodesPerQuery(17).setMaxFractionFullNodesPerQuery(0.3).setNodeAllocatorType("fixed_count");
    assertFullMapping(properties, expected);
}

Example 3 with NodeSchedulerConfig

use of io.trino.execution.scheduler.NodeSchedulerConfig in project trino by trinodb.

the class TestLocalExchange method setUp.

@BeforeMethod
public void setUp() {
    NodeScheduler nodeScheduler = new NodeScheduler(new UniformNodeSelectorFactory(new InMemoryNodeManager(), new NodeSchedulerConfig().setIncludeCoordinator(true), new NodeTaskMap(new FinalizerService())));
    nodePartitioningManager = new NodePartitioningManager(nodeScheduler, new BlockTypeOperators(new TypeOperators()));
}

Example 4 with NodeSchedulerConfig

use of io.trino.execution.scheduler.NodeSchedulerConfig in project trino by trinodb.

the class TestNodeScheduler method testTopologyAwareScheduling.

@Test(timeOut = 60 * 1000)
public void testTopologyAwareScheduling() {
    NodeTaskMap nodeTaskMap = new NodeTaskMap(finalizerService);
    InMemoryNodeManager nodeManager = new InMemoryNodeManager();
    ImmutableList.Builder<InternalNode> nodeBuilder = ImmutableList.builder();
    nodeBuilder.add(new InternalNode("node1", URI.create("http://host1.rack1:11"), NodeVersion.UNKNOWN, false));
    nodeBuilder.add(new InternalNode("node2", URI.create("http://host2.rack1:12"), NodeVersion.UNKNOWN, false));
    nodeBuilder.add(new InternalNode("node3", URI.create("http://host3.rack2:13"), NodeVersion.UNKNOWN, false));
    ImmutableList<InternalNode> nodes = nodeBuilder.build();
    nodeManager.addNode(CONNECTOR_ID, nodes);
    // contents of taskMap indicate the node-task map for the current stage
    Map<InternalNode, RemoteTask> taskMap = new HashMap<>();
    NodeSchedulerConfig nodeSchedulerConfig = new NodeSchedulerConfig().setMaxSplitsPerNode(25).setIncludeCoordinator(false).setMaxPendingSplitsPerTask(20);
    TestNetworkTopology topology = new TestNetworkTopology();
    NodeSelectorFactory nodeSelectorFactory = new TopologyAwareNodeSelectorFactory(topology, nodeManager, nodeSchedulerConfig, nodeTaskMap, getNetworkTopologyConfig());
    NodeScheduler nodeScheduler = new NodeScheduler(nodeSelectorFactory);
    NodeSelector nodeSelector = nodeScheduler.createNodeSelector(session, Optional.of(CONNECTOR_ID));
    // Fill up the nodes with non-local data
    ImmutableSet.Builder<Split> nonRackLocalBuilder = ImmutableSet.builder();
    for (int i = 0; i < (25 + 11) * 3; i++) {
        nonRackLocalBuilder.add(new Split(CONNECTOR_ID, new TestSplitRemote(HostAddress.fromParts("data.other_rack", 1)), Lifespan.taskWide()));
    }
    Set<Split> nonRackLocalSplits = nonRackLocalBuilder.build();
    Multimap<InternalNode, Split> assignments = nodeSelector.computeAssignments(nonRackLocalSplits, ImmutableList.copyOf(taskMap.values())).getAssignments();
    MockRemoteTaskFactory remoteTaskFactory = new MockRemoteTaskFactory(remoteTaskExecutor, remoteTaskScheduledExecutor);
    int task = 0;
    for (InternalNode node : assignments.keySet()) {
        TaskId taskId = new TaskId(new StageId("test", 1), task, 0);
        task++;
        MockRemoteTaskFactory.MockRemoteTask remoteTask = remoteTaskFactory.createTableScanTask(taskId, node, ImmutableList.copyOf(assignments.get(node)), nodeTaskMap.createPartitionedSplitCountTracker(node, taskId));
        remoteTask.startSplits(25);
        nodeTaskMap.addTask(node, remoteTask);
        taskMap.put(node, remoteTask);
    }
    // Continue assigning to fill up part of the queue
    nonRackLocalSplits = Sets.difference(nonRackLocalSplits, new HashSet<>(assignments.values()));
    assignments = nodeSelector.computeAssignments(nonRackLocalSplits, ImmutableList.copyOf(taskMap.values())).getAssignments();
    for (InternalNode node : assignments.keySet()) {
        RemoteTask remoteTask = taskMap.get(node);
        remoteTask.addSplits(ImmutableMultimap.<PlanNodeId, Split>builder().putAll(new PlanNodeId("sourceId"), assignments.get(node)).build());
    }
    nonRackLocalSplits = Sets.difference(nonRackLocalSplits, new HashSet<>(assignments.values()));
    // Check that 3 of the splits were rejected, since they're non-local
    assertEquals(nonRackLocalSplits.size(), 3);
    // Assign rack-local splits
    ImmutableSet.Builder<Split> rackLocalSplits = ImmutableSet.builder();
    HostAddress dataHost1 = HostAddress.fromParts("data.rack1", 1);
    HostAddress dataHost2 = HostAddress.fromParts("data.rack2", 1);
    for (int i = 0; i < 6 * 2; i++) {
        rackLocalSplits.add(new Split(CONNECTOR_ID, new TestSplitRemote(dataHost1), Lifespan.taskWide()));
    }
    for (int i = 0; i < 6; i++) {
        rackLocalSplits.add(new Split(CONNECTOR_ID, new TestSplitRemote(dataHost2), Lifespan.taskWide()));
    }
    assignments = nodeSelector.computeAssignments(rackLocalSplits.build(), ImmutableList.copyOf(taskMap.values())).getAssignments();
    for (InternalNode node : assignments.keySet()) {
        RemoteTask remoteTask = taskMap.get(node);
        remoteTask.addSplits(ImmutableMultimap.<PlanNodeId, Split>builder().putAll(new PlanNodeId("sourceId"), assignments.get(node)).build());
    }
    Set<Split> unassigned = Sets.difference(rackLocalSplits.build(), new HashSet<>(assignments.values()));
    // Compute the assignments a second time to account for the fact that some splits may not have been assigned due to asynchronous
    // loading of the NetworkLocationCache
    assignments = nodeSelector.computeAssignments(unassigned, ImmutableList.copyOf(taskMap.values())).getAssignments();
    for (InternalNode node : assignments.keySet()) {
        RemoteTask remoteTask = taskMap.get(node);
        remoteTask.addSplits(ImmutableMultimap.<PlanNodeId, Split>builder().putAll(new PlanNodeId("sourceId"), assignments.get(node)).build());
    }
    unassigned = Sets.difference(unassigned, new HashSet<>(assignments.values()));
    assertEquals(unassigned.size(), 3);
    int rack1 = 0;
    int rack2 = 0;
    for (Split split : unassigned) {
        String rack = topology.locate(split.getAddresses().get(0)).getSegments().get(0);
        switch(rack) {
            case "rack1":
                rack1++;
                break;
            case "rack2":
                rack2++;
                break;
            default:
                throw new AssertionError("Unexpected rack: " + rack);
        }
    }
    assertEquals(rack1, 2);
    assertEquals(rack2, 1);
    // Assign local splits
    ImmutableSet.Builder<Split> localSplits = ImmutableSet.builder();
    localSplits.add(new Split(CONNECTOR_ID, new TestSplitRemote(HostAddress.fromParts("host1.rack1", 1)), Lifespan.taskWide()));
    localSplits.add(new Split(CONNECTOR_ID, new TestSplitRemote(HostAddress.fromParts("host2.rack1", 1)), Lifespan.taskWide()));
    localSplits.add(new Split(CONNECTOR_ID, new TestSplitRemote(HostAddress.fromParts("host3.rack2", 1)), Lifespan.taskWide()));
    assignments = nodeSelector.computeAssignments(localSplits.build(), ImmutableList.copyOf(taskMap.values())).getAssignments();
    assertEquals(assignments.size(), 3);
    assertEquals(assignments.keySet().size(), 3);
}

Example 5 with NodeSchedulerConfig

use of io.trino.execution.scheduler.NodeSchedulerConfig in project trino by trinodb.

the class TestNodeScheduler method setUp.

@BeforeMethod
public void setUp() {
    session = TestingSession.testSessionBuilder().build();
    finalizerService = new FinalizerService();
    nodeTaskMap = new NodeTaskMap(finalizerService);
    nodeManager = new InMemoryNodeManager();
    nodeSchedulerConfig = new NodeSchedulerConfig().setMaxSplitsPerNode(20).setIncludeCoordinator(false).setMaxPendingSplitsPerTask(10);
    nodeScheduler = new NodeScheduler(new UniformNodeSelectorFactory(nodeManager, nodeSchedulerConfig, nodeTaskMap));
    // contents of taskMap indicate the node-task map for the current stage
    taskMap = new HashMap<>();
    nodeSelector = nodeScheduler.createNodeSelector(session, Optional.of(CONNECTOR_ID));
    remoteTaskExecutor = newCachedThreadPool(daemonThreadsNamed("remoteTaskExecutor-%s"));
    remoteTaskScheduledExecutor = newScheduledThreadPool(2, daemonThreadsNamed("remoteTaskScheduledExecutor-%s"));
    finalizerService.start();
}