Examples with StormMetricsRegistry org.apache.storm.metric.StormMetricsRegistry used on opensource projects

Example 21 with StormMetricsRegistry

use of org.apache.storm.metric.StormMetricsRegistry in project storm by apache.

the class TestResourceAwareScheduler method testMultipleUsers.

@Test
public void testMultipleUsers() {
    INimbus iNimbus = new INimbusTest();
    Map<String, SupervisorDetails> supMap = genSupervisors(20, 4, 1000, 1024 * 10);
    Map<String, Map<String, Number>> resourceUserPool = userResourcePool(userRes("jerry", 1_000, 8_192), userRes("bobby", 10_000, 32_768), userRes("derek", 5_000, 16_384));
    Config config = createClusterConfig(10, 128, 0, resourceUserPool);
    Topologies topologies = new Topologies(genTopology("topo-1", config, 5, 15, 1, 1, currentTime - 2, 20, "jerry"), genTopology("topo-2", config, 5, 15, 1, 1, currentTime - 8, 29, "jerry"), genTopology("topo-3", config, 5, 15, 1, 1, currentTime - 16, 29, "jerry"), genTopology("topo-4", config, 5, 15, 1, 1, currentTime - 16, 20, "jerry"), genTopology("topo-5", config, 5, 15, 1, 1, currentTime - 24, 29, "jerry"), genTopology("topo-6", config, 5, 15, 1, 1, currentTime - 2, 20, "bobby"), genTopology("topo-7", config, 5, 15, 1, 1, currentTime - 8, 29, "bobby"), genTopology("topo-8", config, 5, 15, 1, 1, currentTime - 16, 29, "bobby"), genTopology("topo-9", config, 5, 15, 1, 1, currentTime - 16, 20, "bobby"), genTopology("topo-10", config, 5, 15, 1, 1, currentTime - 24, 29, "bobby"), genTopology("topo-11", config, 5, 15, 1, 1, currentTime - 2, 20, "derek"), genTopology("topo-12", config, 5, 15, 1, 1, currentTime - 8, 29, "derek"), genTopology("topo-13", config, 5, 15, 1, 1, currentTime - 16, 29, "derek"), genTopology("topo-14", config, 5, 15, 1, 1, currentTime - 16, 20, "derek"), genTopology("topo-15", config, 5, 15, 1, 1, currentTime - 24, 29, "derek"));
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, config);
    scheduler = new ResourceAwareScheduler();
    scheduler.prepare(config, new StormMetricsRegistry());
    scheduler.schedule(topologies, cluster);
    for (TopologyDetails td : topologies) {
        assertTopologiesFullyScheduled(cluster, td.getName());
    }
}

Example 22 with StormMetricsRegistry

use of org.apache.storm.metric.StormMetricsRegistry in project storm by apache.

the class TestResourceAwareScheduler method testRASNodeSlotAssign.

@Test
public void testRASNodeSlotAssign() {
    Config config = new Config();
    config.putAll(defaultTopologyConf);
    INimbus iNimbus = new INimbusTest();
    Map<String, SupervisorDetails> supMap = genSupervisors(5, 4, 400, 2000);
    TopologyDetails topology1 = genTopology("topology1", config, 1, 0, 2, 0, 0, 0, "user");
    TopologyDetails topology2 = genTopology("topology2", config, 1, 0, 2, 0, 0, 0, "user");
    Topologies topologies = new Topologies(topology1, topology2);
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, config);
    Map<String, RasNode> nodes = RasNodes.getAllNodesFrom(cluster);
    assertEquals(5, nodes.size());
    RasNode node = nodes.get("r000s000");
    assertEquals("r000s000", node.getId());
    assertTrue(node.isAlive());
    assertEquals(0, node.getRunningTopologies().size());
    assertTrue(node.isTotallyFree());
    assertEquals(4, node.totalSlotsFree());
    assertEquals(0, node.totalSlotsUsed());
    assertEquals(4, node.totalSlots());
    List<ExecutorDetails> executors11 = new ArrayList<>();
    executors11.add(new ExecutorDetails(1, 1));
    node.assign(node.getFreeSlots().iterator().next(), topology1, executors11);
    assertEquals(1, node.getRunningTopologies().size());
    assertFalse(node.isTotallyFree());
    assertEquals(3, node.totalSlotsFree());
    assertEquals(1, node.totalSlotsUsed());
    assertEquals(4, node.totalSlots());
    List<ExecutorDetails> executors12 = new ArrayList<>();
    executors12.add(new ExecutorDetails(2, 2));
    node.assign(node.getFreeSlots().iterator().next(), topology1, executors12);
    assertEquals(1, node.getRunningTopologies().size());
    assertFalse(node.isTotallyFree());
    assertEquals(2, node.totalSlotsFree());
    assertEquals(2, node.totalSlotsUsed());
    assertEquals(4, node.totalSlots());
    List<ExecutorDetails> executors21 = new ArrayList<>();
    executors21.add(new ExecutorDetails(1, 1));
    node.assign(node.getFreeSlots().iterator().next(), topology2, executors21);
    assertEquals(2, node.getRunningTopologies().size());
    assertFalse(node.isTotallyFree());
    assertEquals(1, node.totalSlotsFree());
    assertEquals(3, node.totalSlotsUsed());
    assertEquals(4, node.totalSlots());
    List<ExecutorDetails> executors22 = new ArrayList<>();
    executors22.add(new ExecutorDetails(2, 2));
    node.assign(node.getFreeSlots().iterator().next(), topology2, executors22);
    assertEquals(2, node.getRunningTopologies().size());
    assertFalse(node.isTotallyFree());
    assertEquals(0, node.totalSlotsFree());
    assertEquals(4, node.totalSlotsUsed());
    assertEquals(4, node.totalSlots());
    node.freeAllSlots();
    assertEquals(0, node.getRunningTopologies().size());
    assertTrue(node.isTotallyFree());
    assertEquals(4, node.totalSlotsFree());
    assertEquals(0, node.totalSlotsUsed());
    assertEquals(4, node.totalSlots());
}

Example 23 with StormMetricsRegistry

use of org.apache.storm.metric.StormMetricsRegistry in project storm by apache.

the class TestUser method testResourcePoolUtilization.

@Test
public void testResourcePoolUtilization() {
    INimbus iNimbus = new INimbusTest();
    Map<String, SupervisorDetails> supMap = genSupervisors(4, 4, 100, 1000);
    Double cpuGuarantee = 400.0;
    Double memoryGuarantee = 1000.0;
    Map<String, Map<String, Number>> resourceUserPool = userResourcePool(userRes("user1", cpuGuarantee, memoryGuarantee));
    Config config = createClusterConfig(100, 200, 200, resourceUserPool);
    TopologyDetails topo1 = genTopology("topo-1", config, 1, 1, 2, 1, Time.currentTimeSecs() - 24, 9, "user1");
    Topologies topologies = new Topologies(topo1);
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, config);
    User user1 = new User("user1", toDouble(resourceUserPool.get("user1")));
    WorkerSlot slot = cluster.getAvailableSlots().get(0);
    cluster.assign(slot, topo1.getId(), topo1.getExecutors());
    Assert.assertEquals("check cpu resource guarantee", cpuGuarantee, user1.getCpuResourceGuaranteed(), 0.001);
    Assert.assertEquals("check memory resource guarantee", memoryGuarantee, user1.getMemoryResourceGuaranteed(), 0.001);
    Assert.assertEquals("check cpu resource pool utilization", ((100.0 * 3.0) / cpuGuarantee), user1.getCpuResourcePoolUtilization(cluster), 0.001);
    Assert.assertEquals("check memory resource pool utilization", ((200.0 + 200.0) * 3.0) / memoryGuarantee, user1.getMemoryResourcePoolUtilization(cluster), 0.001);
}

Example 24 with StormMetricsRegistry

use of org.apache.storm.metric.StormMetricsRegistry in project storm by apache.

the class TestNodeSorterHostProximity method testAntiAffinityWithMultipleTopologies.

/**
 * Schedule two topologies, once with special resources and another without.
 * There are enough special resources to hold one topology with special resource ("my.gpu").
 * If the sort order is incorrect, scheduling will not succeed.
 */
@Test
public void testAntiAffinityWithMultipleTopologies() {
    INimbus iNimbus = new INimbusTest();
    Map<String, SupervisorDetails> supMap = genSupervisorsWithRacks(1, 40, 66, 0, 0, 4700, 226200, new HashMap<>());
    HashMap<String, Double> extraResources = new HashMap<>();
    extraResources.put("my.gpu", 1.0);
    supMap.putAll(genSupervisorsWithRacks(1, 40, 66, 1, 0, 4700, 226200, extraResources));
    Config config = new Config();
    config.putAll(createGrasClusterConfig(88, 775, 25, null, null));
    IScheduler scheduler = new ResourceAwareScheduler();
    scheduler.prepare(config, new StormMetricsRegistry());
    TopologyDetails tdSimple = genTopology("topology-simple", config, 1, 5, 100, 300, 0, 0, "user", 8192);
    // Schedule the simple topology first
    Topologies topologies = new Topologies(tdSimple);
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, config);
    {
        NodeSorterHostProximity nodeSorter = new NodeSorterHostProximity(cluster, tdSimple);
        for (ExecutorDetails exec : tdSimple.getExecutors()) {
            nodeSorter.prepare(exec);
            List<ObjectResourcesItem> sortedRacks = StreamSupport.stream(nodeSorter.getSortedRacks().spliterator(), false).collect(Collectors.toList());
            String rackSummaries = StreamSupport.stream(sortedRacks.spliterator(), false).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"));
            NormalizedResourceRequest topoResourceRequest = tdSimple.getApproximateTotalResources();
            String topoRequest = String.format("Topo %s, approx-requested-resources %s", tdSimple.getId(), topoResourceRequest.toString());
            Assert.assertEquals(rackSummaries + "\n# of racks sorted", 2, sortedRacks.size());
            Assert.assertEquals(rackSummaries + "\nFirst rack sorted", "rack-000", sortedRacks.get(0).id);
            Assert.assertEquals(rackSummaries + "\nSecond rack sorted", "rack-001", sortedRacks.get(1).id);
        }
    }
    scheduler.schedule(topologies, cluster);
    TopologyBuilder builder = topologyBuilder(1, 5, 100, 300);
    builder.setBolt("gpu-bolt", new TestBolt(), 40).addResource("my.gpu", 1.0).shuffleGrouping("spout-0");
    TopologyDetails tdGpu = topoToTopologyDetails("topology-gpu", config, builder.createTopology(), 0, 0, "user", 8192);
    // Now schedule GPU but with the simple topology in place.
    topologies = new Topologies(tdSimple, tdGpu);
    cluster = new Cluster(cluster, topologies);
    {
        NodeSorterHostProximity nodeSorter = new NodeSorterHostProximity(cluster, tdGpu);
        for (ExecutorDetails exec : tdGpu.getExecutors()) {
            String comp = tdGpu.getComponentFromExecutor(exec);
            nodeSorter.prepare(exec);
            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"));
            NormalizedResourceRequest topoResourceRequest = tdSimple.getApproximateTotalResources();
            String topoRequest = String.format("Topo %s, approx-requested-resources %s", tdSimple.getId(), topoResourceRequest.toString());
            Assert.assertEquals(rackSummaries + "\n# of racks sorted", 2, sortedRacks.size());
            if (comp.equals("gpu-bolt")) {
                Assert.assertEquals(rackSummaries + "\nFirst rack sorted for " + comp, "rack-001", sortedRacks.get(0).id);
                Assert.assertEquals(rackSummaries + "\nSecond rack sorted for " + comp, "rack-000", sortedRacks.get(1).id);
            } else {
                Assert.assertEquals(rackSummaries + "\nFirst rack sorted for " + comp, "rack-000", sortedRacks.get(0).id);
                Assert.assertEquals(rackSummaries + "\nSecond rack sorted for " + comp, "rack-001", sortedRacks.get(1).id);
            }
        }
    }
    scheduler.schedule(topologies, cluster);
    Map<String, SchedulerAssignment> assignments = new TreeMap<>(cluster.getAssignments());
    assertEquals(2, assignments.size());
    Map<String, Map<String, AtomicLong>> topoPerRackCount = new HashMap<>();
    for (Map.Entry<String, SchedulerAssignment> entry : assignments.entrySet()) {
        SchedulerAssignment sa = entry.getValue();
        Map<String, AtomicLong> slotsPerRack = new TreeMap<>();
        for (WorkerSlot slot : sa.getSlots()) {
            String nodeId = slot.getNodeId();
            String rack = supervisorIdToRackName(nodeId);
            slotsPerRack.computeIfAbsent(rack, (r) -> new AtomicLong(0)).incrementAndGet();
        }
        LOG.info("{} => {}", entry.getKey(), slotsPerRack);
        topoPerRackCount.put(entry.getKey(), slotsPerRack);
    }
    Map<String, AtomicLong> simpleCount = topoPerRackCount.get("topology-simple-0");
    assertNotNull(simpleCount);
    // Because the simple topology was scheduled first we want to be sure that it didn't put anything on
    // the GPU nodes.
    // Only 1 rack is in use
    assertEquals(1, simpleCount.size());
    // r001 is the second rack with GPUs
    assertFalse(simpleCount.containsKey("r001"));
    // r000 is the first rack with no GPUs
    assertTrue(simpleCount.containsKey("r000"));
// We don't really care too much about the scheduling of topology-gpu-0, because it was scheduled.
}

Example 25 with StormMetricsRegistry

use of org.apache.storm.metric.StormMetricsRegistry in project storm by apache.

the class TestNodeSorterHostProximity method testFillUpRackAndSpilloverToNextRack.

/**
 * If the topology is too large for one rack, it should be partially scheduled onto the next rack (and next rack only).
 */
@Test
public void testFillUpRackAndSpilloverToNextRack() {
    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 numRacks = 3;
    final int numSupersPerRack = 10;
    final int numPortsPerSuper = 6;
    final int numZonesPerHost = 1;
    final double numaResourceMultiplier = 1.0;
    int rackStartNum = 0;
    int supStartNum = 0;
    // not enough for topo1
    long compPerRack = (topo1NumSpouts * topo1SpoutParallelism + topo1NumBolts * topo1BoltParallelism) * 4 / 5;
    long compPerSuper = compPerRack / numSupersPerRack;
    double cpuPerSuper = compPcore * compPerSuper;
    double memPerSuper = (compOnHeap + compOffHeap) * compPerSuper;
    double topo1MaxHeapSize = memPerSuper;
    final String topoName1 = "topology1";
    Map<String, SupervisorDetails> supMap = genSupervisorsWithRacksAndNuma(numRacks, numSupersPerRack, numZonesPerHost, numPortsPerSuper, rackStartNum, supStartNum, cpuPerSuper, memPerSuper, Collections.emptyMap(), numaResourceMultiplier);
    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);
    // Schedule the topo1 topology and ensure it fits on 2 racks
    Topologies topologies = new Topologies(td1);
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, config);
    cluster.setNetworkTopography(testDNSToSwitchMapping.getRackToHosts());
    scheduler.schedule(topologies, cluster);
    Set<String> assignedRacks = cluster.getAssignedRacks(td1.getId());
    assertEquals("Racks for topology=" + td1.getId() + " is " + assignedRacks, 2, assignedRacks.size());
}