Examples with ResourceAwareScheduler org.apache.storm.scheduler.resource.ResourceAwareScheduler used on opensource projects

Example 31 with ResourceAwareScheduler

use of org.apache.storm.scheduler.resource.ResourceAwareScheduler in project storm by apache.

the class TestLargeCluster method testLargeCluster.

/**
 * Create a large cluster, read topologies and configuration from resource directory and schedule.
 *
 * @throws Exception upon error.
 */
@Test
public void testLargeCluster() throws Exception {
    for (TEST_CLUSTER_NAME testClusterName : TEST_CLUSTER_NAME.values()) {
        LOG.info("********************************************");
        LOG.info("testLargeCluster: Start Processing cluster {}", testClusterName.getClusterName());
        String resourcePath = testClusterName.getResourcePath();
        Map<String, SupervisorDetails> supervisors = createSupervisors(testClusterName, 0);
        TopologyDetails[] topoDetailsArray = createTopoDetailsArray(resourcePath, false);
        Assert.assertTrue("No topologies found for cluster " + testClusterName.getClusterName(), topoDetailsArray.length > 0);
        Topologies topologies = new Topologies(topoDetailsArray);
        Config confWithDefaultStrategy = new Config();
        confWithDefaultStrategy.putAll(topoDetailsArray[0].getConf());
        confWithDefaultStrategy.put(Config.TOPOLOGY_SCHEDULER_STRATEGY, DefaultResourceAwareStrategy.class.getName());
        confWithDefaultStrategy.put(Config.STORM_NETWORK_TOPOGRAPHY_PLUGIN, TestUtilsForResourceAwareScheduler.GenSupervisorsDnsToSwitchMapping.class.getName());
        INimbus iNimbus = new INimbusTest();
        Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supervisors, new HashMap<>(), topologies, confWithDefaultStrategy);
        scheduler = new ResourceAwareScheduler();
        List<Class> classesToDebug = Arrays.asList(DefaultResourceAwareStrategy.class, GenericResourceAwareStrategy.class, ResourceAwareScheduler.class, Cluster.class);
        // switch to Level.DEBUG for verbose otherwise Level.INFO
        Level logLevel = Level.INFO;
        classesToDebug.forEach(x -> Configurator.setLevel(x.getName(), logLevel));
        long startTime = System.currentTimeMillis();
        scheduler.prepare(confWithDefaultStrategy, new StormMetricsRegistry());
        scheduler.schedule(topologies, cluster);
        long endTime = System.currentTimeMillis();
        LOG.info("Cluster={} Scheduling Time: {} topologies in {} seconds", testClusterName.getClusterName(), topoDetailsArray.length, (endTime - startTime) / 1000.0);
        for (TopologyDetails td : topoDetailsArray) {
            TestUtilsForResourceAwareScheduler.assertTopologiesFullyScheduled(cluster, td.getName());
        }
        // Remove topology and reschedule it
        for (int i = 0; i < topoDetailsArray.length; i++) {
            startTime = System.currentTimeMillis();
            TopologyDetails topoDetails = topoDetailsArray[i];
            cluster.unassign(topoDetails.getId());
            LOG.info("Cluster={},  ({}) Removed topology {}", testClusterName.getClusterName(), i, topoDetails.getName());
            IScheduler rescheduler = new ResourceAwareScheduler();
            rescheduler.prepare(confWithDefaultStrategy, new StormMetricsRegistry());
            rescheduler.schedule(topologies, cluster);
            TestUtilsForResourceAwareScheduler.assertTopologiesFullyScheduled(cluster, topoDetails.getName());
            endTime = System.currentTimeMillis();
            LOG.info("Cluster={}, ({}) Scheduling Time: Removed topology {} and rescheduled in {} seconds", testClusterName.getClusterName(), i, topoDetails.getName(), (endTime - startTime) / 1000.0);
        }
        classesToDebug.forEach(x -> Configurator.setLevel(x.getName(), Level.INFO));
        LOG.info("testLargeCluster: End Processing cluster {}", testClusterName.getClusterName());
        LOG.info("********************************************");
    }
}

Example 32 with ResourceAwareScheduler

use of org.apache.storm.scheduler.resource.ResourceAwareScheduler in project storm by apache.

the class TestDefaultEvictionStrategy method testEvictTopologyFromItself.

/**
 * If topologies from other users cannot be evicted to make space
 * check if there is a topology with lower priority that can be evicted from the current user
 */
@Test
public void testEvictTopologyFromItself() {
    INimbus iNimbus = new INimbusTest();
    Map<String, SupervisorDetails> supMap = genSupervisors(4, 4, 100, 1000);
    Map<String, Map<String, Number>> resourceUserPool = userResourcePool(userRes("jerry", 200, 2000), userRes("bobby", 100, 1000), userRes("derek", 100, 1000));
    Config config = createClusterConfig(100, 500, 500, resourceUserPool);
    Topologies topologies = new Topologies(genTopology("topo-1", config, 1, 0, 1, 0, currentTime - 2, 20, "jerry"), genTopology("topo-2", config, 1, 0, 1, 0, currentTime - 2, 20, "jerry"), genTopology("topo-5", config, 1, 0, 1, 0, currentTime - 2, 10, "bobby"), genTopology("topo-6", config, 1, 0, 1, 0, currentTime - 2, 29, "derek"));
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, config);
    scheduler = new ResourceAwareScheduler();
    scheduler.prepare(config, new StormMetricsRegistry());
    LOG.info("\n\n\t\tScheduling topos 1,2,5,6");
    scheduler.schedule(topologies, cluster);
    LOG.info("\n\n\t\tDone Scheduling...");
    assertTopologiesFullyScheduled(cluster, "topo-1", "topo-2", "topo-5", "topo-6");
    // user jerry submits another topology into a full cluster
    // topo3 should not be able to scheduled
    topologies = addTopologies(topologies, genTopology("topo-3", config, 1, 0, 1, 0, currentTime - 2, 29, "jerry"));
    cluster = new Cluster(cluster, topologies);
    LOG.info("\n\n\t\tScheduling topos 1,2,3,5,6");
    scheduler.schedule(topologies, cluster);
    LOG.info("\n\n\t\tDone Scheduling...");
    assertTopologiesFullyScheduled(cluster, "topo-1", "topo-2", "topo-5", "topo-6");
    assertTopologiesNotScheduled(cluster, "topo-3");
    // user jerry submits another topology but this one should be scheduled since it has higher priority than than the
    // rest of jerry's running topologies
    topologies = addTopologies(topologies, genTopology("topo-4", config, 1, 0, 1, 0, currentTime - 2, 10, "jerry"));
    cluster = new Cluster(cluster, topologies);
    LOG.info("\n\n\t\tScheduling topos 1-6");
    scheduler.schedule(topologies, cluster);
    LOG.info("\n\n\t\tDone Scheduling...");
    assertTopologiesFullyScheduled(cluster, "topo-1", "topo-4", "topo-5", "topo-6");
    assertTopologiesNotScheduled(cluster, "topo-2", "topo-3");
}

Example 33 with ResourceAwareScheduler

use of org.apache.storm.scheduler.resource.ResourceAwareScheduler in project storm by apache.

the class TestDefaultEvictionStrategy method testEvictMultipleTopologies.

@Test
public void testEvictMultipleTopologies() {
    INimbus iNimbus = new INimbusTest();
    Map<String, SupervisorDetails> supMap = genSupervisors(4, 4, 100, 1000);
    Map<String, Map<String, Number>> resourceUserPool = userResourcePool(userRes("jerry", 200, 2000), userRes("derek", 100, 1000));
    Config config = createClusterConfig(100, 500, 500, resourceUserPool);
    Topologies topologies = new Topologies(genTopology("topo-2", config, 1, 0, 1, 0, currentTime - 2, 10, "bobby"), genTopology("topo-3", config, 1, 0, 1, 0, currentTime - 2, 20, "bobby"), genTopology("topo-4", config, 1, 0, 1, 0, currentTime - 2, 29, "derek"), genTopology("topo-5", config, 1, 0, 1, 0, currentTime - 2, 29, "derek"));
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, config);
    scheduler = new ResourceAwareScheduler();
    scheduler.prepare(config, new StormMetricsRegistry());
    LOG.info("\n\n\t\tScheduling topos 2 to 5...");
    scheduler.schedule(topologies, cluster);
    LOG.info("\n\n\t\tDone scheduling...");
    assertTopologiesFullyScheduled(cluster, "topo-2", "topo-3", "topo-4", "topo-5");
    // user jerry submits another topology
    topologies = addTopologies(topologies, genTopology("topo-1", config, 2, 0, 1, 0, currentTime - 2, 10, "jerry"));
    cluster = new Cluster(cluster, topologies);
    LOG.info("\n\n\t\tScheduling topos 1 to 5");
    scheduler.schedule(topologies, cluster);
    LOG.info("\n\n\t\tDone scheduling...");
    // bobby has no guarantee so topo-2 and topo-3 evicted
    assertTopologiesFullyScheduled(cluster, "topo-1", "topo-4", "topo-5");
    assertTopologiesNotScheduled(cluster, "topo-2", "topo-3");
}

Example 34 with ResourceAwareScheduler

use of org.apache.storm.scheduler.resource.ResourceAwareScheduler in project storm by apache.

the class TestGenericResourceAwareSchedulingPriorityStrategy method testDefaultSchedulingPriorityStrategyEvicting.

/*
     * DefaultSchedulingPriorityStrategy does not take generic resources into account when calculating score
     * So even if a user is requesting a lot of generic resources other than CPU and memory, scheduler will still score it very low and kick out other topologies
     *
     *  Ethan asks for medium cpu and memory while Rui asks for little cpu and memory but heavy generic resource
     *  However, Rui's generic request can not be met and default scoring system is not taking generic resources into account,
     *  so the score of Rui's new topo will be much lower than all Ethan's topos'.
     *  Then all Ethan's topo will be evicted in trying to make rooms for Rui.
     */
@Test
public void testDefaultSchedulingPriorityStrategyEvicting() {
    Map<String, Double> requestedgenericResourcesMap = new HashMap<>();
    requestedgenericResourcesMap.put("generic.resource.1", 40.0);
    Config ruiConf = createGrasClusterConfig(10, 10, 10, null, requestedgenericResourcesMap);
    Config ethanConf = createGrasClusterConfig(60, 200, 300, null, Collections.emptyMap());
    Topologies topologies = new Topologies(genTopology("ethan-topo-1", ethanConf, 1, 0, 1, 0, currentTime - 2, 10, "ethan"), genTopology("ethan-topo-2", ethanConf, 1, 0, 1, 0, currentTime - 2, 20, "ethan"), genTopology("ethan-topo-3", ethanConf, 1, 0, 1, 0, currentTime - 2, 28, "ethan"), genTopology("ethan-topo-4", ethanConf, 1, 0, 1, 0, currentTime - 2, 29, "ethan"));
    Topologies withNewTopo = addTopologies(topologies, genTopology("rui-topo-1", ruiConf, 1, 0, 5, 0, currentTime - 2, 10, "rui"));
    Config config = mkClusterConfig(DefaultSchedulingPriorityStrategy.class.getName());
    Cluster cluster = mkTestCluster(topologies, config);
    scheduler = new ResourceAwareScheduler();
    scheduler.prepare(config, new StormMetricsRegistry());
    scheduler.schedule(topologies, cluster);
    assertTopologiesFullyScheduled(cluster, "ethan-topo-1", "ethan-topo-2", "ethan-topo-3", "ethan-topo-4");
    cluster = new Cluster(cluster, withNewTopo);
    scheduler.schedule(withNewTopo, cluster);
    Map<String, Set<String>> evictedTopos = ((ResourceAwareScheduler) scheduler).getEvictedTopologiesMap();
    assertTopologiesFullyScheduled(cluster, "ethan-topo-1", "ethan-topo-2", "ethan-topo-3", "ethan-topo-4");
    assertTopologiesBeenEvicted(cluster, collectMapValues(evictedTopos), "ethan-topo-1", "ethan-topo-2", "ethan-topo-3", "ethan-topo-4");
    assertTopologiesNotScheduled(cluster, "rui-topo-1");
}

Example 35 with ResourceAwareScheduler

use of org.apache.storm.scheduler.resource.ResourceAwareScheduler 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);
}