Examples with Topologies org.apache.storm.scheduler.Topologies used on opensource projects

Example 66 with Topologies

use of org.apache.storm.scheduler.Topologies in project storm by apache.

the class TestConstraintSolverStrategy method testZeroExecutorScheduling.

@Test
public void testZeroExecutorScheduling() {
    ConstraintSolverStrategy cs = new ConstraintSolverStrategy();
    cs.prepare(new HashMap<>());
    Map<String, Object> topoConf = Utils.readDefaultConfig();
    topoConf.put(Config.TOPOLOGY_RAS_CONSTRAINT_MAX_STATE_SEARCH, 1_000);
    topoConf.put(Config.TOPOLOGY_RAS_ONE_EXECUTOR_PER_WORKER, false);
    topoConf.put(Config.TOPOLOGY_RAS_ONE_COMPONENT_PER_WORKER, false);
    TopologyDetails topo = makeTopology(topoConf, 1);
    Cluster cluster = makeCluster(new Topologies(topo));
    cs.schedule(cluster, topo);
    LOG.info("********************* Scheduling Zero Unassigned Executors *********************");
    // reschedule a fully schedule topology
    cs.schedule(cluster, topo);
    LOG.info("********************* End of Scheduling Zero Unassigned Executors *********************");
}

Example 67 with Topologies

use of org.apache.storm.scheduler.Topologies in project storm by apache.

the class TestConstraintSolverStrategy method basicUnitTestWithKillAndRecover.

public void basicUnitTestWithKillAndRecover(ConstraintSolverStrategy cs, int boltParallel, int coLocationCnt) {
    Map<String, Object> config = makeTestTopoConf(coLocationCnt);
    cs.prepare(config);
    TopologyDetails topo = makeTopology(config, boltParallel);
    Topologies topologies = new Topologies(topo);
    Cluster cluster = makeCluster(topologies);
    LOG.info("Scheduling...");
    SchedulingResult result = cs.schedule(cluster, topo);
    LOG.info("Done scheduling {}...", result);
    Assert.assertTrue("Assert scheduling topology success " + result, result.isSuccess());
    Assert.assertEquals("Assert no unassigned executors, found unassigned: " + cluster.getUnassignedExecutors(topo), 0, cluster.getUnassignedExecutors(topo).size());
    Assert.assertTrue("Valid Scheduling?", ConstraintSolverStrategy.validateSolution(cluster, topo));
    LOG.info("Slots Used {}", cluster.getAssignmentById(topo.getId()).getSlots());
    LOG.info("Assignment {}", cluster.getAssignmentById(topo.getId()).getSlotToExecutors());
    // simulate worker loss
    SchedulerAssignment assignment = cluster.getAssignmentById(topo.getId());
    Set<WorkerSlot> slotsToDelete = new HashSet<>();
    Set<WorkerSlot> slots = assignment.getSlots();
    int i = 0;
    for (WorkerSlot slot : slots) {
        if (i % 2 == 0) {
            slotsToDelete.add(slot);
        }
        i++;
    }
    LOG.info("KILL WORKER(s) {}", slotsToDelete);
    for (WorkerSlot slot : slotsToDelete) {
        cluster.freeSlot(slot);
    }
    cs = new ConstraintSolverStrategy();
    cs.prepare(config);
    LOG.info("Scheduling again...");
    result = cs.schedule(cluster, topo);
    LOG.info("Done scheduling {}...", result);
    Assert.assertTrue("Assert scheduling topology success " + result, result.isSuccess());
    Assert.assertEquals("topo all executors scheduled?", 0, cluster.getUnassignedExecutors(topo).size());
    Assert.assertTrue("Valid Scheduling?", ConstraintSolverStrategy.validateSolution(cluster, topo));
}

Example 68 with Topologies

use of org.apache.storm.scheduler.Topologies in project storm by apache.

the class TestDefaultResourceAwareStrategy method testMultipleRacks.

/**
 * Test whether strategy will choose correct rack
 */
@Test
public void testMultipleRacks() {
    final Map<String, SupervisorDetails> supMap = new HashMap<>();
    final Map<String, SupervisorDetails> supMapRack0 = genSupervisors(10, 4, 0, 400, 8000);
    // generate another rack of supervisors with less resources
    final Map<String, SupervisorDetails> supMapRack1 = genSupervisors(10, 4, 10, 200, 4000);
    // generate some supervisors that are depleted of one resource
    final Map<String, SupervisorDetails> supMapRack2 = genSupervisors(10, 4, 20, 0, 8000);
    // generate some that has alot of memory but little of cpu
    final Map<String, SupervisorDetails> supMapRack3 = genSupervisors(10, 4, 30, 10, 8000 * 2 + 4000);
    // generate some that has alot of cpu but little of memory
    final Map<String, SupervisorDetails> supMapRack4 = genSupervisors(10, 4, 40, 400 + 200 + 10, 1000);
    // Generate some that have neither resource, to verify that the strategy will prioritize this last
    // Also put a generic resource with 0 value in the resources list, to verify that it doesn't affect the sorting
    final Map<String, SupervisorDetails> supMapRack5 = genSupervisors(10, 4, 50, 0.0, 0.0, Collections.singletonMap("gpu.count", 0.0));
    supMap.putAll(supMapRack0);
    supMap.putAll(supMapRack1);
    supMap.putAll(supMapRack2);
    supMap.putAll(supMapRack3);
    supMap.putAll(supMapRack4);
    supMap.putAll(supMapRack5);
    Config config = createClusterConfig(100, 500, 500, null);
    config.put(Config.TOPOLOGY_WORKER_MAX_HEAP_SIZE_MB, Double.MAX_VALUE);
    INimbus iNimbus = new INimbusTest();
    // create test DNSToSwitchMapping plugin
    DNSToSwitchMapping TestNetworkTopographyPlugin = new TestDNSToSwitchMapping(supMapRack0, supMapRack1, supMapRack2, supMapRack3, supMapRack4, supMapRack5);
    // generate topologies
    TopologyDetails topo1 = genTopology("topo-1", config, 8, 0, 2, 0, CURRENT_TIME - 2, 10, "user");
    TopologyDetails topo2 = genTopology("topo-2", config, 8, 0, 2, 0, CURRENT_TIME - 2, 10, "user");
    Topologies topologies = new Topologies(topo1, topo2);
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, config);
    List<String> supHostnames = new LinkedList<>();
    for (SupervisorDetails sup : supMap.values()) {
        supHostnames.add(sup.getHost());
    }
    Map<String, List<String>> rackToNodes = new HashMap<>();
    Map<String, String> resolvedSuperVisors = TestNetworkTopographyPlugin.resolve(supHostnames);
    for (Map.Entry<String, String> entry : resolvedSuperVisors.entrySet()) {
        String hostName = entry.getKey();
        String rack = entry.getValue();
        rackToNodes.computeIfAbsent(rack, rid -> new ArrayList<>()).add(hostName);
    }
    cluster.setNetworkTopography(rackToNodes);
    DefaultResourceAwareStrategyOld rs = new DefaultResourceAwareStrategyOld();
    rs.prepareForScheduling(cluster, topo1);
    INodeSorter nodeSorter = new NodeSorterHostProximity(cluster, topo1, BaseResourceAwareStrategy.NodeSortType.DEFAULT_RAS);
    nodeSorter.prepare(null);
    Iterable<ObjectResourcesItem> sortedRacks = nodeSorter.getSortedRacks();
    Iterator<ObjectResourcesItem> it = sortedRacks.iterator();
    // Ranked first since rack-0 has the most balanced set of resources
    Assert.assertEquals("rack-0 should be ordered first", "rack-0", it.next().id);
    // Ranked second since rack-1 has a balanced set of resources but less than rack-0
    Assert.assertEquals("rack-1 should be ordered second", "rack-1", it.next().id);
    // Ranked third since rack-4 has a lot of cpu but not a lot of memory
    Assert.assertEquals("rack-4 should be ordered third", "rack-4", it.next().id);
    // Ranked fourth since rack-3 has alot of memory but not cpu
    Assert.assertEquals("rack-3 should be ordered fourth", "rack-3", it.next().id);
    // Ranked fifth since rack-2 has not cpu resources
    Assert.assertEquals("rack-2 should be ordered fifth", "rack-2", it.next().id);
    // Ranked last since rack-5 has neither CPU nor memory available
    assertEquals("Rack-5 should be ordered sixth", "rack-5", it.next().id);
    SchedulingResult schedulingResult = rs.schedule(cluster, topo1);
    assert (schedulingResult.isSuccess());
    SchedulerAssignment assignment = cluster.getAssignmentById(topo1.getId());
    for (WorkerSlot ws : assignment.getSlotToExecutors().keySet()) {
        // make sure all workers on scheduled in rack-0
        Assert.assertEquals("assert worker scheduled on rack-0", "rack-0", resolvedSuperVisors.get(rs.idToNode(ws.getNodeId()).getHostname()));
    }
    Assert.assertEquals("All executors in topo-1 scheduled", 0, cluster.getUnassignedExecutors(topo1).size());
    // Test if topology is already partially scheduled on one rack
    Iterator<ExecutorDetails> executorIterator = topo2.getExecutors().iterator();
    List<String> nodeHostnames = rackToNodes.get("rack-1");
    for (int i = 0; i < topo2.getExecutors().size() / 2; i++) {
        String nodeHostname = nodeHostnames.get(i % nodeHostnames.size());
        RasNode node = rs.hostnameToNodes(nodeHostname).get(0);
        WorkerSlot targetSlot = node.getFreeSlots().iterator().next();
        ExecutorDetails targetExec = executorIterator.next();
        // to keep track of free slots
        node.assign(targetSlot, topo2, Arrays.asList(targetExec));
    }
    rs = new DefaultResourceAwareStrategyOld();
    // schedule topo2
    schedulingResult = rs.schedule(cluster, topo2);
    assert (schedulingResult.isSuccess());
    assignment = cluster.getAssignmentById(topo2.getId());
    for (WorkerSlot ws : assignment.getSlotToExecutors().keySet()) {
        // make sure all workers on scheduled in rack-1
        Assert.assertEquals("assert worker scheduled on rack-1", "rack-1", resolvedSuperVisors.get(rs.idToNode(ws.getNodeId()).getHostname()));
    }
    Assert.assertEquals("All executors in topo-2 scheduled", 0, cluster.getUnassignedExecutors(topo2).size());
}

Example 69 with Topologies

use of org.apache.storm.scheduler.Topologies in project storm by apache.

the class TestDefaultResourceAwareStrategy method testDefaultResourceAwareStrategyWithSettingAckerExecutors.

/**
 * test if the scheduling logic for the DefaultResourceAwareStrategy is correct
 * when topology.acker.executors is set.
 *
 * If yes, topology.acker.executors.per.worker setting will be ignored and calculated as
 * Math.ceil(topology.acker.executors / estimate num of workers) by Nimbus
 */
@ParameterizedTest
@ValueSource(ints = { -1, 0, 2, 300 })
public void testDefaultResourceAwareStrategyWithSettingAckerExecutors(int numOfAckersPerWorker) throws InvalidTopologyException {
    int spoutParallelism = 1;
    int boltParallelism = 2;
    TopologyBuilder builder = new TopologyBuilder();
    builder.setSpout("spout", new TestSpout(), spoutParallelism);
    builder.setBolt("bolt-1", new TestBolt(), boltParallelism).shuffleGrouping("spout");
    builder.setBolt("bolt-2", new TestBolt(), boltParallelism).shuffleGrouping("bolt-1");
    builder.setBolt("bolt-3", new TestBolt(), boltParallelism).shuffleGrouping("bolt-2");
    String topoName = "testTopology";
    StormTopology stormToplogy = builder.createTopology();
    INimbus iNimbus = new INimbusTest();
    Map<String, SupervisorDetails> supMap = genSupervisors(4, 4, 200, 2000);
    Config conf = createClusterConfig(50, 450, 0, null);
    conf.put(Config.TOPOLOGY_PRIORITY, 0);
    conf.put(Config.TOPOLOGY_NAME, topoName);
    conf.put(Config.TOPOLOGY_WORKER_MAX_HEAP_SIZE_MB, 2000);
    conf.put(Config.TOPOLOGY_SUBMITTER_USER, "user");
    conf.put(Config.TOPOLOGY_ACKER_EXECUTORS, 4);
    conf.put(Config.TOPOLOGY_RAS_ACKER_EXECUTORS_PER_WORKER, numOfAckersPerWorker);
    if (numOfAckersPerWorker == -1) {
    // Leave topology.acker.executors.per.worker unset
    } else {
        conf.put(Config.TOPOLOGY_RAS_ACKER_EXECUTORS_PER_WORKER, numOfAckersPerWorker);
    }
    int estimatedNumWorker = ServerUtils.getEstimatedWorkerCountForRasTopo(conf, stormToplogy);
    Nimbus.setUpAckerExecutorConfigs(topoName, conf, conf, estimatedNumWorker);
    conf.put(Config.TOPOLOGY_ACKER_RESOURCES_ONHEAP_MEMORY_MB, 250);
    conf.put(Config.TOPOLOGY_ACKER_CPU_PCORE_PERCENT, 50);
    TopologyDetails topo = new TopologyDetails("testTopology-id", conf, stormToplogy, 0, genExecsAndComps(StormCommon.systemTopology(conf, stormToplogy)), CURRENT_TIME, "user");
    Topologies topologies = new Topologies(topo);
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, conf);
    scheduler = new ResourceAwareScheduler();
    scheduler.prepare(conf, new StormMetricsRegistry());
    scheduler.schedule(topologies, cluster);
    // Sorted execs: [[6, 6], [2, 2], [4, 4], [5, 5], [1, 1], [3, 3], [0, 0], [8, 8], [7, 7], [10, 10], [9, 9]]
    // Ackers: [[8, 8], [7, 7], [10, 10], [9, 9]]
    HashSet<HashSet<ExecutorDetails>> expectedScheduling = new HashSet<>();
    expectedScheduling.add(new HashSet<>(Arrays.asList(// bolt-3
    new ExecutorDetails(6, 6), // bolt-1
    new ExecutorDetails(2, 2), // acker
    new ExecutorDetails(7, 7), // acker
    new ExecutorDetails(8, 8))));
    expectedScheduling.add(new HashSet<>(Arrays.asList(// bolt-3
    new ExecutorDetails(5, 5), // bolt-2
    new ExecutorDetails(4, 4), // acker
    new ExecutorDetails(9, 9), // acker
    new ExecutorDetails(10, 10))));
    expectedScheduling.add(new HashSet<>(Arrays.asList(// spout
    new ExecutorDetails(0, 0), // bolt-2
    new ExecutorDetails(3, 3), // bolt-1
    new ExecutorDetails(1, 1))));
    HashSet<HashSet<ExecutorDetails>> foundScheduling = new HashSet<>();
    SchedulerAssignment assignment = cluster.getAssignmentById("testTopology-id");
    for (Collection<ExecutorDetails> execs : assignment.getSlotToExecutors().values()) {
        foundScheduling.add(new HashSet<>(execs));
    }
    Assert.assertEquals(expectedScheduling, foundScheduling);
}

Example 70 with Topologies

use of org.apache.storm.scheduler.Topologies in project storm by apache.

the class TestDefaultResourceAwareStrategy method testMultipleRacksWithFavoritism.

/**
 * Test whether strategy will choose correct rack
 */
@Test
public void testMultipleRacksWithFavoritism() {
    final Map<String, SupervisorDetails> supMap = new HashMap<>();
    final Map<String, SupervisorDetails> supMapRack0 = genSupervisors(10, 4, 0, 400, 8000);
    // generate another rack of supervisors with less resources
    final Map<String, SupervisorDetails> supMapRack1 = genSupervisors(10, 4, 10, 200, 4000);
    // generate some supervisors that are depleted of one resource
    final Map<String, SupervisorDetails> supMapRack2 = genSupervisors(10, 4, 20, 0, 8000);
    // generate some that has alot of memory but little of cpu
    final Map<String, SupervisorDetails> supMapRack3 = genSupervisors(10, 4, 30, 10, 8000 * 2 + 4000);
    // generate some that has alot of cpu but little of memory
    final Map<String, SupervisorDetails> supMapRack4 = genSupervisors(10, 4, 40, 400 + 200 + 10, 1000);
    supMap.putAll(supMapRack0);
    supMap.putAll(supMapRack1);
    supMap.putAll(supMapRack2);
    supMap.putAll(supMapRack3);
    supMap.putAll(supMapRack4);
    Config config = createClusterConfig(100, 500, 500, null);
    config.put(Config.TOPOLOGY_WORKER_MAX_HEAP_SIZE_MB, Double.MAX_VALUE);
    INimbus iNimbus = new INimbusTest();
    // create test DNSToSwitchMapping plugin
    DNSToSwitchMapping TestNetworkTopographyPlugin = new TestDNSToSwitchMapping(supMapRack0, supMapRack1, supMapRack2, supMapRack3, supMapRack4);
    Config t1Conf = new Config();
    t1Conf.putAll(config);
    final List<String> t1FavoredHostNames = Arrays.asList("host-41", "host-42", "host-43");
    t1Conf.put(Config.TOPOLOGY_SCHEDULER_FAVORED_NODES, t1FavoredHostNames);
    final List<String> t1UnfavoredHostIds = Arrays.asList("host-1", "host-2", "host-3");
    t1Conf.put(Config.TOPOLOGY_SCHEDULER_UNFAVORED_NODES, t1UnfavoredHostIds);
    // generate topologies
    TopologyDetails topo1 = genTopology("topo-1", t1Conf, 8, 0, 2, 0, CURRENT_TIME - 2, 10, "user");
    Config t2Conf = new Config();
    t2Conf.putAll(config);
    t2Conf.put(Config.TOPOLOGY_SCHEDULER_FAVORED_NODES, Arrays.asList("host-31", "host-32", "host-33"));
    t2Conf.put(Config.TOPOLOGY_SCHEDULER_UNFAVORED_NODES, Arrays.asList("host-11", "host-12", "host-13"));
    TopologyDetails topo2 = genTopology("topo-2", t2Conf, 8, 0, 2, 0, CURRENT_TIME - 2, 10, "user");
    Topologies topologies = new Topologies(topo1, topo2);
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, config);
    List<String> supHostnames = new LinkedList<>();
    for (SupervisorDetails sup : supMap.values()) {
        supHostnames.add(sup.getHost());
    }
    Map<String, List<String>> rackToNodes = new HashMap<>();
    Map<String, String> resolvedSuperVisors = TestNetworkTopographyPlugin.resolve(supHostnames);
    for (Map.Entry<String, String> entry : resolvedSuperVisors.entrySet()) {
        String hostName = entry.getKey();
        String rack = entry.getValue();
        List<String> nodesForRack = rackToNodes.get(rack);
        if (nodesForRack == null) {
            nodesForRack = new ArrayList<>();
            rackToNodes.put(rack, nodesForRack);
        }
        nodesForRack.add(hostName);
    }
    cluster.setNetworkTopography(rackToNodes);
    DefaultResourceAwareStrategyOld rs = new DefaultResourceAwareStrategyOld();
    rs.prepareForScheduling(cluster, topo1);
    INodeSorter nodeSorter = new NodeSorterHostProximity(cluster, topo1, BaseResourceAwareStrategy.NodeSortType.DEFAULT_RAS);
    nodeSorter.prepare(null);
    Iterable<ObjectResourcesItem> sortedRacks = nodeSorter.getSortedRacks();
    Iterator<ObjectResourcesItem> it = sortedRacks.iterator();
    // Ranked first since rack-0 has the most balanced set of resources
    Assert.assertEquals("rack-0 should be ordered first", "rack-0", it.next().id);
    // Ranked second since rack-1 has a balanced set of resources but less than rack-0
    Assert.assertEquals("rack-1 should be ordered second", "rack-1", it.next().id);
    // Ranked third since rack-4 has a lot of cpu but not a lot of memory
    Assert.assertEquals("rack-4 should be ordered third", "rack-4", it.next().id);
    // Ranked fourth since rack-3 has alot of memory but not cpu
    Assert.assertEquals("rack-3 should be ordered fourth", "rack-3", it.next().id);
    // Ranked last since rack-2 has not cpu resources
    Assert.assertEquals("rack-2 should be ordered fifth", "rack-2", it.next().id);
    SchedulingResult schedulingResult = rs.schedule(cluster, topo1);
    assert (schedulingResult.isSuccess());
    SchedulerAssignment assignment = cluster.getAssignmentById(topo1.getId());
    for (WorkerSlot ws : assignment.getSlotToExecutors().keySet()) {
        String hostName = rs.idToNode(ws.getNodeId()).getHostname();
        String rackId = resolvedSuperVisors.get(hostName);
        Assert.assertTrue(ws + " is neither on a favored node " + t1FavoredHostNames + " nor the highest priority rack (rack-0)", t1FavoredHostNames.contains(hostName) || "rack-0".equals(rackId));
        Assert.assertFalse(ws + " is a part of an unfavored node " + t1UnfavoredHostIds, t1UnfavoredHostIds.contains(hostName));
    }
    Assert.assertEquals("All executors in topo-1 scheduled", 0, cluster.getUnassignedExecutors(topo1).size());
    // Test if topology is already partially scheduled on one rack
    Iterator<ExecutorDetails> executorIterator = topo2.getExecutors().iterator();
    List<String> nodeHostnames = rackToNodes.get("rack-1");
    for (int i = 0; i < topo2.getExecutors().size() / 2; i++) {
        String nodeHostname = nodeHostnames.get(i % nodeHostnames.size());
        RasNode node = rs.hostnameToNodes(nodeHostname).get(0);
        WorkerSlot targetSlot = node.getFreeSlots().iterator().next();
        ExecutorDetails targetExec = executorIterator.next();
        // to keep track of free slots
        node.assign(targetSlot, topo2, Arrays.asList(targetExec));
    }
    rs = new DefaultResourceAwareStrategyOld();
    // schedule topo2
    schedulingResult = rs.schedule(cluster, topo2);
    assert (schedulingResult.isSuccess());
    assignment = cluster.getAssignmentById(topo2.getId());
    for (WorkerSlot ws : assignment.getSlotToExecutors().keySet()) {
        // make sure all workers on scheduled in rack-1
        // The favored nodes would have put it on a different rack, but because that rack does not have free space to run the
        // topology it falls back to this rack
        Assert.assertEquals("assert worker scheduled on rack-1", "rack-1", resolvedSuperVisors.get(rs.idToNode(ws.getNodeId()).getHostname()));
    }
    Assert.assertEquals("All executors in topo-2 scheduled", 0, cluster.getUnassignedExecutors(topo2).size());
}