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

Example 76 with Topologies

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

the class TestNodeSorterHostProximity method testWithImpairedClusterNetworkTopography.

/**
 * Assign and then clear out a rack to host list mapping in cluster.networkTopography.
 * Expected behavior is that:
 *  <li>the rack without hosts does not show up in {@link NodeSorterHostProximity#getSortedRacks()}</li>
 *  <li>all the supervisor nodes still get returned in {@link NodeSorterHostProximity#sortAllNodes()} ()}</li>
 *  <li>supervisors on cleared rack show up under {@link DNSToSwitchMapping#DEFAULT_RACK}</li>
 *
 *  <p>
 *      Force an usual condition, where one of the racks is still passed to LazyNodeSortingIterator with
 *      an empty list and then ensure that code is resilient.
 *  </p>
 */
@Test
void testWithImpairedClusterNetworkTopography() {
    INimbus iNimbus = new INimbusTest();
    double compPcore = 100;
    double compOnHeap = 775;
    double compOffHeap = 25;
    int topo1NumSpouts = 1;
    int topo1NumBolts = 5;
    int topo1SpoutParallelism = 100;
    int topo1BoltParallelism = 200;
    final int numSupersPerRack = 10;
    final int numPortsPerSuper = 66;
    long compPerRack = (topo1NumSpouts * topo1SpoutParallelism + topo1NumBolts * topo1BoltParallelism + 10);
    long compPerSuper = compPerRack / numSupersPerRack;
    double cpuPerSuper = compPcore * compPerSuper;
    double memPerSuper = (compOnHeap + compOffHeap) * compPerSuper;
    double topo1MaxHeapSize = memPerSuper;
    final String topoName1 = "topology1";
    int numRacks = 3;
    Map<String, SupervisorDetails> supMap = genSupervisorsWithRacks(numRacks, numSupersPerRack, numPortsPerSuper, 0, 0, cpuPerSuper, memPerSuper, new HashMap<>());
    TestDNSToSwitchMapping testDNSToSwitchMapping = new TestDNSToSwitchMapping(supMap.values());
    Config config = new Config();
    config.putAll(createGrasClusterConfig(compPcore, compOnHeap, compOffHeap, null, null));
    config.put(Config.TOPOLOGY_SCHEDULER_STRATEGY, GenericResourceAwareStrategy.class.getName());
    IScheduler scheduler = new ResourceAwareScheduler();
    scheduler.prepare(config, new StormMetricsRegistry());
    TopologyDetails td1 = genTopology(topoName1, config, topo1NumSpouts, topo1NumBolts, topo1SpoutParallelism, topo1BoltParallelism, 0, 0, "user", topo1MaxHeapSize);
    Topologies topologies = new Topologies(td1);
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, config);
    cluster.setNetworkTopography(testDNSToSwitchMapping.getRackToHosts());
    Map<String, List<String>> networkTopography = cluster.getNetworkTopography();
    assertEquals("Expecting " + numRacks + " racks found " + networkTopography.size(), numRacks, networkTopography.size());
    assertTrue("Expecting racks count to be >= 3, found " + networkTopography.size(), networkTopography.size() >= 3);
    // Impair cluster.networkTopography and set one rack to have zero hosts, getSortedRacks should exclude this rack.
    // Keep, the supervisorDetails unchanged - confirm that these nodes are not lost even with incomplete networkTopography
    String rackIdToZero = networkTopography.keySet().stream().findFirst().get();
    impairClusterRack(cluster, rackIdToZero, true, false);
    NodeSorterHostProximity nodeSorterHostProximity = new NodeSorterHostProximity(cluster, td1);
    nodeSorterHostProximity.getSortedRacks().forEach(x -> assertNotEquals(x.id, rackIdToZero));
    // confirm that the above action has not lost the hosts and that they appear under the DEFAULT rack
    {
        Set<String> seenRacks = new HashSet<>();
        nodeSorterHostProximity.getSortedRacks().forEach(x -> seenRacks.add(x.id));
        assertEquals("Expecting rack cnt to be still " + numRacks, numRacks, seenRacks.size());
        assertTrue("Expecting to see default-rack=" + DNSToSwitchMapping.DEFAULT_RACK + " in sortedRacks", seenRacks.contains(DNSToSwitchMapping.DEFAULT_RACK));
    }
    // now check if node/supervisor is missing when sorting all nodes
    Set<String> expectedNodes = supMap.keySet();
    Set<String> seenNodes = new HashSet<>();
    nodeSorterHostProximity.prepare(null);
    nodeSorterHostProximity.sortAllNodes().forEach(n -> seenNodes.add(n));
    assertEquals("Expecting see all supervisors ", expectedNodes, seenNodes);
    // Now fully impair the cluster - confirm no default rack
    {
        cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, config);
        cluster.setNetworkTopography(new TestDNSToSwitchMapping(supMap.values()).getRackToHosts());
        impairClusterRack(cluster, rackIdToZero, true, true);
        Set<String> seenRacks = new HashSet<>();
        NodeSorterHostProximity nodeSorterHostProximity2 = new NodeSorterHostProximity(cluster, td1);
        nodeSorterHostProximity2.getSortedRacks().forEach(x -> seenRacks.add(x.id));
        Map<String, Set<String>> rackIdToHosts = nodeSorterHostProximity2.getRackIdToHosts();
        String dumpOfRacks = rackIdToHosts.entrySet().stream().map(x -> String.format("rack %s -> hosts [%s]", x.getKey(), String.join(",", x.getValue()))).collect(Collectors.joining("\n\t"));
        assertEquals("Expecting rack cnt to be " + (numRacks - 1) + " but found " + seenRacks.size() + "\n\t" + dumpOfRacks, numRacks - 1, seenRacks.size());
        assertFalse("Found default-rack=" + DNSToSwitchMapping.DEFAULT_RACK + " in \n\t" + dumpOfRacks, seenRacks.contains(DNSToSwitchMapping.DEFAULT_RACK));
    }
}

Example 77 with Topologies

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

the class TestNodeSorterHostProximity method testMultipleRacks.

/**
 * Test whether strategy will choose correct rack.
 */
@Test
public void testMultipleRacks() {
    final Map<String, SupervisorDetails> supMap = new HashMap<>();
    final int numRacks = 1;
    final int numSupersPerRack = 10;
    final int numPortsPerSuper = 4;
    final int numZonesPerHost = 1;
    final double numaResourceMultiplier = 1.0;
    int rackStartNum = 0;
    int supStartNum = 0;
    final Map<String, SupervisorDetails> supMapRack0 = genSupervisorsWithRacksAndNuma(numRacks, numSupersPerRack, numZonesPerHost, numPortsPerSuper, rackStartNum++, supStartNum, 400, 8000, Collections.emptyMap(), numaResourceMultiplier);
    // generate another rack of supervisors with less resources
    supStartNum += numSupersPerRack;
    final Map<String, SupervisorDetails> supMapRack1 = genSupervisorsWithRacksAndNuma(numRacks, numSupersPerRack, numZonesPerHost, numPortsPerSuper, rackStartNum++, supStartNum, 200, 4000, Collections.emptyMap(), numaResourceMultiplier);
    // generate some supervisors that are depleted of one resource
    supStartNum += numSupersPerRack;
    final Map<String, SupervisorDetails> supMapRack2 = genSupervisorsWithRacksAndNuma(numRacks, numSupersPerRack, numZonesPerHost, numPortsPerSuper, rackStartNum++, supStartNum, 0, 8000, Collections.emptyMap(), numaResourceMultiplier);
    // generate some that has a lot of memory but little of cpu
    supStartNum += numSupersPerRack;
    final Map<String, SupervisorDetails> supMapRack3 = genSupervisorsWithRacksAndNuma(numRacks, numSupersPerRack, numZonesPerHost, numPortsPerSuper, rackStartNum++, supStartNum, 10, 8000 * 2 + 4000, Collections.emptyMap(), numaResourceMultiplier);
    // generate some that has a lot of cpu but little of memory
    supStartNum += numSupersPerRack;
    final Map<String, SupervisorDetails> supMapRack4 = genSupervisorsWithRacksAndNuma(numRacks, numSupersPerRack, numZonesPerHost, numPortsPerSuper, rackStartNum++, supStartNum, 400 + 200 + 10, 1000, Collections.emptyMap(), numaResourceMultiplier);
    // 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
    supStartNum += numSupersPerRack;
    final Map<String, SupervisorDetails> supMapRack5 = genSupervisorsWithRacksAndNuma(numRacks, numSupersPerRack, numZonesPerHost, numPortsPerSuper, rackStartNum++, supStartNum, 0.0, 0.0, Collections.singletonMap("gpu.count", 0.0), numaResourceMultiplier);
    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
    TestDNSToSwitchMapping testDNSToSwitchMapping = 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>> rackToHosts = testDNSToSwitchMapping.getRackToHosts();
    cluster.setNetworkTopography(rackToHosts);
    NodeSorterHostProximity nodeSorter = new NodeSorterHostProximity(cluster, topo1, BaseResourceAwareStrategy.NodeSortType.DEFAULT_RAS);
    nodeSorter.prepare(null);
    List<ObjectResourcesItem> sortedRacks = StreamSupport.stream(nodeSorter.getSortedRacks().spliterator(), false).collect(Collectors.toList());
    String rackSummaries = sortedRacks.stream().map(x -> String.format("Rack %s -> scheduled-cnt %d, min-avail %f, avg-avail %f, cpu %f, mem %f", x.id, nodeSorter.getScheduledExecCntByRackId().getOrDefault(x.id, new AtomicInteger(-1)).get(), x.minResourcePercent, x.avgResourcePercent, x.availableResources.getTotalCpu(), x.availableResources.getTotalMemoryMb())).collect(Collectors.joining("\n\t"));
    Assert.assertEquals(rackSummaries + "\n# of racks sorted", 6, sortedRacks.size());
    Iterator<ObjectResourcesItem> it = sortedRacks.iterator();
    Assert.assertEquals(rackSummaries + "\nrack-000 should be ordered first since it has the most balanced set of resources", "rack-000", it.next().id);
    Assert.assertEquals(rackSummaries + "\nrack-001 should be ordered second since it has a balanced set of resources but less than rack-000", "rack-001", it.next().id);
    Assert.assertEquals(rackSummaries + "\nrack-004 should be ordered third since it has a lot of cpu but not a lot of memory", "rack-004", it.next().id);
    Assert.assertEquals(rackSummaries + "\nrack-003 should be ordered fourth since it has a lot of memory but not cpu", "rack-003", it.next().id);
    Assert.assertEquals(rackSummaries + "\nrack-002 should be ordered fifth since it has not cpu resources", "rack-002", it.next().id);
    Assert.assertEquals(rackSummaries + "\nRack-005 should be ordered sixth since it has neither CPU nor memory available", "rack-005", it.next().id);
}

Example 78 with Topologies

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

the class TestResourceAwareScheduler method minCpuPreventsThirdTopo.

/**
 * Min CPU for worker set to 40%.  1 supervisor with 100% CPU.
 * 2 topologies with 2 10% components should schedule.  A third topology should then fail scheduling due to lack of CPU.
 */
@Test
public void minCpuPreventsThirdTopo() {
    INimbus iNimbus = new INimbusTest();
    Map<String, SupervisorDetails> supMap = genSupervisors(1, 4, 100, 60000);
    Config config = createClusterConfig(10, 500, 500, null);
    config.put(DaemonConfig.STORM_WORKER_MIN_CPU_PCORE_PERCENT, 40.0);
    TopologyDetails topo1 = genTopology("topo-1", config, 2, 0, 1, 1, currentTime - 2, 20, "jerry");
    TopologyDetails topo2 = genTopology("topo-2", config, 2, 0, 1, 1, currentTime - 2, 20, "jerry");
    TopologyDetails topo3 = genTopology("topo-3", config, 2, 0, 1, 1, currentTime - 2, 20, "jerry");
    Topologies topologies = new Topologies(topo1, topo2, topo3);
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<String, SchedulerAssignmentImpl>(), topologies, config);
    scheduler = new ResourceAwareScheduler();
    scheduler.prepare(config, new StormMetricsRegistry());
    scheduler.schedule(topologies, cluster);
    assertFalse(cluster.needsSchedulingRas(topo1));
    assertFalse(cluster.needsSchedulingRas(topo2));
    assertTrue(cluster.needsSchedulingRas(topo3));
    assertTrue("topo-1 scheduled?", cluster.getAssignmentById(topo1.getId()) != null);
    assertTrue("topo-2 scheduled?", cluster.getAssignmentById(topo2.getId()) != null);
    assertFalse("topo-3 unscheduled?", cluster.getAssignmentById(topo3.getId()) != null);
    SchedulerAssignment assignment1 = cluster.getAssignmentById(topo1.getId());
    assertEquals(1, assignment1.getSlots().size());
    Map<WorkerSlot, WorkerResources> assignedSlots1 = assignment1.getScheduledResources();
    double assignedCpu = 0.0;
    for (Entry<WorkerSlot, WorkerResources> entry : assignedSlots1.entrySet()) {
        WorkerResources wr = entry.getValue();
        assignedCpu += wr.get_cpu();
    }
    assertEquals(40.0, assignedCpu, 0.001);
    SchedulerAssignment assignment2 = cluster.getAssignmentById(topo2.getId());
    assertEquals(1, assignment2.getSlots().size());
    Map<WorkerSlot, WorkerResources> assignedSlots2 = assignment2.getScheduledResources();
    assignedCpu = 0.0;
    for (Entry<WorkerSlot, WorkerResources> entry : assignedSlots2.entrySet()) {
        WorkerResources wr = entry.getValue();
        assignedCpu += wr.get_cpu();
    }
    assertEquals(40.0, assignedCpu, 0.001);
}

Example 79 with Topologies

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

the class TestResourceAwareScheduler method minCpuWorkerJustFits.

/**
 * Min CPU for worker set to 50%.  1 supervisor with 100% CPU.
 * A topology with 10 10% components should schedule.
 */
@Test
public void minCpuWorkerJustFits() {
    INimbus iNimbus = new INimbusTest();
    Map<String, SupervisorDetails> supMap = genSupervisors(1, 4, 100, 60000);
    Config config = createClusterConfig(10, 500, 500, null);
    config.put(DaemonConfig.STORM_WORKER_MIN_CPU_PCORE_PERCENT, 50.0);
    TopologyDetails topo1 = genTopology("topo-1", config, 10, 0, 1, 1, currentTime - 2, 20, "jerry");
    Topologies topologies = new Topologies(topo1);
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<String, SchedulerAssignmentImpl>(), topologies, config);
    scheduler = new ResourceAwareScheduler();
    scheduler.prepare(config, new StormMetricsRegistry());
    scheduler.schedule(topologies, cluster);
    assertFalse(cluster.needsSchedulingRas(topo1));
    assertTrue("Topo-1 scheduled?", cluster.getAssignmentById(topo1.getId()) != null);
}

Example 80 with Topologies

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

the class TestResourceAwareScheduler method testSchedulingAfterFailedScheduling.

/**
 * When the first topology failed to be scheduled make sure subsequent schedulings can still succeed
 */
@Test
public void testSchedulingAfterFailedScheduling() {
    INimbus iNimbus = new INimbusTest();
    Map<String, SupervisorDetails> supMap = genSupervisors(8, 4, 100, 1000);
    Config config = createClusterConfig(100, 500, 500, null);
    TopologyDetails topo1 = genTopology("topo-1", config, 8, 0, 2, 0, currentTime - 2, 10, "jerry");
    TopologyDetails topo2 = genTopology("topo-2", config, 2, 0, 2, 0, currentTime - 2, 20, "jerry");
    TopologyDetails topo3 = genTopology("topo-3", config, 1, 2, 1, 1, currentTime - 2, 20, "jerry");
    Topologies topologies = new Topologies(topo1, topo2, topo3);
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<String, SchedulerAssignmentImpl>(), topologies, config);
    scheduler = new ResourceAwareScheduler();
    scheduler.prepare(config, new StormMetricsRegistry());
    scheduler.schedule(topologies, cluster);
    assertFalse("Topo-1 unscheduled?", cluster.getAssignmentById(topo1.getId()) != null);
    assertTrue("Topo-2 scheduled?", cluster.getAssignmentById(topo2.getId()) != null);
    assertEquals("Topo-2 all executors scheduled?", 4, cluster.getAssignmentById(topo2.getId()).getExecutorToSlot().size());
    assertTrue("Topo-3 scheduled?", cluster.getAssignmentById(topo3.getId()) != null);
    assertEquals("Topo-3 all executors scheduled?", 3, cluster.getAssignmentById(topo3.getId()).getExecutorToSlot().size());
}