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

Example 31 with ExecutorDetails

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

the class TestLargeCluster method createTopoDetailsArray.

/**
 * Create an array of TopologyDetails by reading serialized files for topology and configuration in the
 * resource path. Skip topologies with no executors/components.
 *
 * @param failOnParseError throw exception if there are unmatched files, otherwise ignore unmatched and read errors.
 * @return An array of TopologyDetails representing resource files.
 * @throws Exception upon error in reading topology serialized files.
 */
public static TopologyDetails[] createTopoDetailsArray(String resourcePath, boolean failOnParseError) throws Exception {
    List<TopologyDetails> topoDetailsList = new ArrayList<>();
    List<String> errors = new ArrayList<>();
    List<String> resources = getResourceFiles(resourcePath);
    Map<String, String> codeResourceMap = new TreeMap<>();
    Map<String, String> confResourceMap = new HashMap<>();
    for (String resource : resources) {
        int idxOfSlash = resource.lastIndexOf("/");
        int idxOfDash = resource.lastIndexOf("-");
        String nm = idxOfDash > idxOfSlash ? resource.substring(idxOfSlash + 1, idxOfDash) : resource.substring(idxOfSlash + 1, resource.length() - COMPRESSED_SERIALIZED_TOPOLOGY_FILENAME_ENDING.length());
        if (resource.endsWith(COMPRESSED_SERIALIZED_TOPOLOGY_FILENAME_ENDING)) {
            codeResourceMap.put(nm, resource);
        } else if (resource.endsWith(COMPRESSED_SERIALIZED_CONFIG_FILENAME_ENDING)) {
            confResourceMap.put(nm, resource);
        } else {
            LOG.info("Ignoring unsupported resource file " + resource);
        }
    }
    String[] examinedConfParams = { Config.TOPOLOGY_NAME, Config.TOPOLOGY_SCHEDULER_STRATEGY, Config.TOPOLOGY_PRIORITY, Config.TOPOLOGY_WORKERS, Config.TOPOLOGY_WORKER_MAX_HEAP_SIZE_MB, Config.TOPOLOGY_SUBMITTER_USER, Config.TOPOLOGY_ACKER_CPU_PCORE_PERCENT, Config.TOPOLOGY_ACKER_RESOURCES_OFFHEAP_MEMORY_MB, Config.TOPOLOGY_ACKER_RESOURCES_ONHEAP_MEMORY_MB };
    for (String topoId : codeResourceMap.keySet()) {
        String codeResource = codeResourceMap.get(topoId);
        if (!confResourceMap.containsKey(topoId)) {
            String err = String.format("Ignoring topology file %s because of missing config file for %s", codeResource, topoId);
            errors.add(err);
            LOG.error(err);
            continue;
        }
        String confResource = confResourceMap.get(topoId);
        LOG.info("Found matching topology and config files: {}, {}", codeResource, confResource);
        StormTopology stormTopology;
        try {
            stormTopology = Utils.deserialize(getResourceAsBytes(codeResource), StormTopology.class);
        } catch (Exception ex) {
            String err = String.format("Cannot read topology from resource %s", codeResource);
            errors.add(err);
            LOG.error(err, ex);
            continue;
        }
        Map<String, Object> conf;
        try {
            conf = Utils.fromCompressedJsonConf(getResourceAsBytes(confResource));
        } catch (RuntimeException | IOException ex) {
            String err = String.format("Cannot read configuration from resource %s", confResource);
            errors.add(err);
            LOG.error(err, ex);
            continue;
        }
        // fix 0.10 conf class names
        String[] configParamsToFix = { Config.TOPOLOGY_SCHEDULER_STRATEGY, Config.STORM_NETWORK_TOPOGRAPHY_PLUGIN, DaemonConfig.RESOURCE_AWARE_SCHEDULER_PRIORITY_STRATEGY };
        for (String configParam : configParamsToFix) {
            if (!conf.containsKey(configParam)) {
                continue;
            }
            String className = (String) conf.get(configParam);
            if (className.startsWith("backtype")) {
                className = className.replace("backtype", "org.apache");
                conf.put(configParam, className);
            }
        }
        // fix conf params used by ConstraintSolverStrategy
        if (!conf.containsKey(DaemonConfig.RESOURCE_AWARE_SCHEDULER_MAX_STATE_SEARCH)) {
            conf.put(DaemonConfig.RESOURCE_AWARE_SCHEDULER_MAX_STATE_SEARCH, 10_000);
        }
        if (!conf.containsKey(Config.TOPOLOGY_RAS_CONSTRAINT_MAX_STATE_SEARCH)) {
            conf.put(Config.TOPOLOGY_RAS_CONSTRAINT_MAX_STATE_SEARCH, 10_000);
        }
        if (!conf.containsKey(Config.TOPOLOGY_RAS_ACKER_EXECUTORS_PER_WORKER)) {
            conf.put(Config.TOPOLOGY_RAS_ACKER_EXECUTORS_PER_WORKER, 1);
        }
        String topoName = (String) conf.getOrDefault(Config.TOPOLOGY_NAME, topoId);
        // conf
        StringBuilder sb = new StringBuilder("Config for " + topoId + ": ");
        for (String param : examinedConfParams) {
            Object val = conf.getOrDefault(param, "<null>");
            sb.append(param).append("=").append(val).append(", ");
        }
        LOG.info(sb.toString());
        // topo
        Map<ExecutorDetails, String> execToComp = TestUtilsForResourceAwareScheduler.genExecsAndComps(stormTopology);
        LOG.info("Topology \"{}\" spouts={}, bolts={}, execToComp size is {}", topoName, stormTopology.get_spouts_size(), stormTopology.get_bolts_size(), execToComp.size());
        if (execToComp.isEmpty()) {
            LOG.error("Topology \"{}\" Ignoring BAD topology with zero executors", topoName);
            continue;
        }
        int numWorkers = Integer.parseInt("" + conf.getOrDefault(Config.TOPOLOGY_WORKERS, "0"));
        TopologyDetails topo = new TopologyDetails(topoId, conf, stormTopology, numWorkers, execToComp, Time.currentTimeSecs(), "user");
        // sanity check - normally this should not fail
        topo.getUserTopolgyComponents();
        topoDetailsList.add(topo);
    }
    if (!errors.isEmpty() && failOnParseError) {
        throw new Exception("Unable to parse all serialized objects\n\t" + String.join("\n\t", errors));
    }
    return topoDetailsList.toArray(new TopologyDetails[0]);
}

Example 32 with ExecutorDetails

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

the class TestDefaultResourceAwareStrategy method testDefaultResourceAwareStrategyWithoutSettingAckerExecutors.

/**
 * test if the scheduling logic for the DefaultResourceAwareStrategy is correct
 * when topology.acker.executors.per.worker is set to different values.
 *
 * If {@link Config#TOPOLOGY_ACKER_EXECUTORS} is not set,
 * it will be calculated by Nimbus as (num of estimated worker * topology.acker.executors.per.worker).
 * In this test, {@link Config#TOPOLOGY_ACKER_EXECUTORS} is set to 2 (num of estimated workers based on topo resources usage)
 *
 * For different value for {@link Config#TOPOLOGY_RAS_ACKER_EXECUTORS_PER_WORKER}:
 * -1: Note we don't really set it to be -1.
 *     It is just a special case in this test that topology.acker.executors.per.worker is unset, nimbus will set to 1 by default.
 * 0:  Since {@link Config#TOPOLOGY_ACKER_EXECUTORS} is not set either, acking is disabled.
 * 1:  2 ackers in total. Distribute 1 acker per worker. With ackers being added, this topology will now need 3 workers.
 *     Then first two worker will get 1 acker and last worker get 0.
 * 2:  4 ackers in total. First two workers will get 2 acker per worker respectively.
 */
@ParameterizedTest
@ValueSource(ints = { -1, 0, 1, 2 })
public void testDefaultResourceAwareStrategyWithoutSettingAckerExecutors(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");
    // Parameterized test on different numOfAckersPerWorker
    if (numOfAckersPerWorker == -1) {
    // Both Config.TOPOLOGY_ACKER_EXECUTORS and Config.TOPOLOGY_RAS_ACKER_EXECUTORS_PER_WORKER are not set
    // Default will be 2 (estimate num of workers) and 1 respectively
    } 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);
    // Ordered execs: [[6, 6], [2, 2], [4, 4], [5, 5], [1, 1], [3, 3], [0, 0], [8, 8], [7, 7]]
    // Ackers: [[8, 8], [7, 7]] (+ [[9, 9], [10, 10]] when numOfAckersPerWorker=2)
    HashSet<HashSet<ExecutorDetails>> expectedScheduling = new HashSet<>();
    if (numOfAckersPerWorker == -1 || numOfAckersPerWorker == 1) {
        // Setting topology.acker.executors = null and topology.acker.executors.per.worker = null
        // are equivalent to topology.acker.executors = null and topology.acker.executors.per.worker = 1
        expectedScheduling.add(new HashSet<>(Arrays.asList(// bolt-3
        new ExecutorDetails(6, 6), // bolt-1
        new ExecutorDetails(2, 2), // bolt-2
        new ExecutorDetails(4, 4), // acker
        new ExecutorDetails(8, 8))));
        expectedScheduling.add(new HashSet<>(Arrays.asList(// bolt-3
        new ExecutorDetails(5, 5), // bolt-1
        new ExecutorDetails(1, 1), // bolt-2
        new ExecutorDetails(3, 3), // acker
        new ExecutorDetails(7, 7))));
        expectedScheduling.add(new HashSet<>(Arrays.asList(// spout
        new ExecutorDetails(0, 0))));
    } else if (numOfAckersPerWorker == 0) {
        expectedScheduling.add(new HashSet<>(Arrays.asList(// bolt-3
        new ExecutorDetails(6, 6), // bolt-1
        new ExecutorDetails(2, 2), // bolt-2
        new ExecutorDetails(4, 4), // bolt-3
        new ExecutorDetails(5, 5))));
        expectedScheduling.add(new HashSet<>(Arrays.asList(// spout
        new ExecutorDetails(0, 0), // bolt-2
        new ExecutorDetails(3, 3), // bolt-1
        new ExecutorDetails(1, 1))));
    } else if (numOfAckersPerWorker == 2) {
        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-2
        new ExecutorDetails(4, 4), // bolt-3
        new ExecutorDetails(5, 5), // acker
        new ExecutorDetails(9, 9), // acker
        new ExecutorDetails(10, 10))));
        expectedScheduling.add(new HashSet<>(Arrays.asList(// bolt-1
        new ExecutorDetails(1, 1), // bolt-2
        new ExecutorDetails(3, 3), // spout
        new ExecutorDetails(0, 0))));
    }
    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 33 with ExecutorDetails

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

the class TestDefaultResourceAwareStrategy method testDefaultResourceAwareStrategyInFavorOfShuffle.

/**
 * test if the scheduling logic for the DefaultResourceAwareStrategy (when made by network proximity needs.) is correct
 */
@Test
public void testDefaultResourceAwareStrategyInFavorOfShuffle() 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");
    StormTopology stormToplogy = builder.createTopology();
    INimbus iNimbus = new INimbusTest();
    Map<String, SupervisorDetails> supMap = genSupervisors(4, 4, 200, 2000);
    Config conf = createClusterConfig(50, 250, 250, null);
    conf.put(Config.TOPOLOGY_PRIORITY, 0);
    conf.put(Config.TOPOLOGY_NAME, "testTopology");
    conf.put(Config.TOPOLOGY_WORKER_MAX_HEAP_SIZE_MB, Double.MAX_VALUE);
    conf.put(Config.TOPOLOGY_SUBMITTER_USER, "user");
    conf.put(Config.TOPOLOGY_RAS_ORDER_EXECUTORS_BY_PROXIMITY_NEEDS, true);
    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);
    ResourceAwareScheduler rs = new ResourceAwareScheduler();
    rs.prepare(conf, new StormMetricsRegistry());
    rs.schedule(topologies, cluster);
    // Sorted execs: [[0, 0], [2, 2], [6, 6], [4, 4], [1, 1], [5, 5], [3, 3], [7, 7]]
    // Ackers: [[7, 7]]]
    HashSet<HashSet<ExecutorDetails>> expectedScheduling = new HashSet<>();
    expectedScheduling.add(new HashSet<>(Arrays.asList(// spout
    new ExecutorDetails(0, 0), // bolt-2
    new ExecutorDetails(6, 6), // bolt-1
    new ExecutorDetails(2, 2), // acker
    new ExecutorDetails(7, 7))));
    expectedScheduling.add(new HashSet<>(Arrays.asList(// bolt-3
    new ExecutorDetails(3, 3), // bolt-2
    new ExecutorDetails(5, 5), // bolt-3
    new ExecutorDetails(4, 4), // 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 34 with ExecutorDetails

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

the class TestUtilsForBlacklistScheduler method genExecsAndComps.

public static Map<ExecutorDetails, String> genExecsAndComps(StormTopology topology, int spoutParallelism, int boltParallelism) {
    Map<ExecutorDetails, String> retMap = new HashMap<>();
    int startTask = 0;
    int endTask = 1;
    for (Map.Entry<String, SpoutSpec> entry : topology.get_spouts().entrySet()) {
        for (int i = 0; i < spoutParallelism; i++) {
            retMap.put(new ExecutorDetails(startTask, endTask), entry.getKey());
            startTask++;
            endTask++;
        }
    }
    for (Map.Entry<String, Bolt> entry : topology.get_bolts().entrySet()) {
        for (int i = 0; i < boltParallelism; i++) {
            retMap.put(new ExecutorDetails(startTask, endTask), entry.getKey());
            startTask++;
            endTask++;
        }
    }
    return retMap;
}

Example 35 with ExecutorDetails

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

the class TestResourceAwareScheduler method testTopologyWorkerMaxHeapSize.

@Test
public void testTopologyWorkerMaxHeapSize() {
    // Test1: If RAS spreads executors across multiple workers based on the set limit for a worker used by the topology
    INimbus iNimbus = new INimbusTest();
    Map<String, SupervisorDetails> supMap = genSupervisors(2, 2, 400, 2000);
    TopologyBuilder builder1 = new TopologyBuilder();
    builder1.setSpout("wordSpout1", new TestWordSpout(), 4);
    StormTopology stormTopology1 = builder1.createTopology();
    Config config1 = new Config();
    config1.putAll(defaultTopologyConf);
    config1.put(Config.TOPOLOGY_WORKER_MAX_HEAP_SIZE_MB, 128.0);
    Map<ExecutorDetails, String> executorMap1 = genExecsAndComps(stormTopology1);
    TopologyDetails topology1 = new TopologyDetails("topology1", config1, stormTopology1, 1, executorMap1, 0, "user");
    ResourceAwareScheduler rs = new ResourceAwareScheduler();
    Topologies topologies = new Topologies(topology1);
    Cluster cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, config1);
    rs.prepare(config1, new StormMetricsRegistry());
    try {
        rs.schedule(topologies, cluster);
        assertFalse(cluster.needsSchedulingRas(topology1));
        assertTrue(cluster.getStatusMap().get(topology1.getId()).startsWith("Running - Fully Scheduled by DefaultResourceAwareStrategy"));
        assertEquals(4, cluster.getAssignedNumWorkers(topology1));
    } finally {
        rs.cleanup();
    }
    // Test2: test when no more workers are available due to topology worker max heap size limit but there is memory is still available
    // wordSpout2 is going to contain 5 executors that needs scheduling. Each of those executors has a memory requirement of 128.0 MB
    // The cluster contains 4 free WorkerSlots. For this topolology each worker is limited to a max heap size of 128.0
    // Thus, one executor not going to be able to get scheduled thus failing the scheduling of this topology and no executors of this
    // topology will be scheduled
    TopologyBuilder builder2 = new TopologyBuilder();
    builder2.setSpout("wordSpout2", new TestWordSpout(), 5);
    StormTopology stormTopology2 = builder2.createTopology();
    Config config2 = new Config();
    config2.putAll(defaultTopologyConf);
    config2.put(Config.TOPOLOGY_WORKER_MAX_HEAP_SIZE_MB, 128.0);
    Map<ExecutorDetails, String> executorMap2 = genExecsAndComps(stormTopology2);
    TopologyDetails topology2 = new TopologyDetails("topology2", config2, stormTopology2, 1, executorMap2, 0, "user");
    topologies = new Topologies(topology2);
    cluster = new Cluster(iNimbus, new ResourceMetrics(new StormMetricsRegistry()), supMap, new HashMap<>(), topologies, config2);
    rs.prepare(config2, new StormMetricsRegistry());
    try {
        rs.schedule(topologies, cluster);
        assertTrue(cluster.needsSchedulingRas(topology2));
        String status = cluster.getStatusMap().get(topology2.getId());
        assert status.startsWith("Not enough resources to schedule") : status;
        // assert status.endsWith("5 executors not scheduled") : status;
        assertEquals(5, cluster.getUnassignedExecutors(topology2).size());
    } finally {
        rs.cleanup();
    }
}