Search in sources :

Example 21 with Twister2Job

use of edu.iu.dsc.tws.api.Twister2Job in project twister2 by DSC-SPIDAL.

the class BatchTsetExample method submitJob.

public static void submitJob(Config config, int containers, JobConfig jobConfig, String clazz) {
    Twister2Job twister2Job;
    twister2Job = Twister2Job.newBuilder().setJobName(clazz.substring(clazz.lastIndexOf(".") + 1)).setWorkerClass(clazz).addComputeResource(1, 512, containers).setConfig(jobConfig).build();
    // now submit the job
    Twister2JobState state = Twister2Submitter.submitJob(twister2Job, config);
}
Also used : Twister2JobState(edu.iu.dsc.tws.api.scheduler.Twister2JobState) Twister2Job(edu.iu.dsc.tws.api.Twister2Job)

Example 22 with Twister2Job

use of edu.iu.dsc.tws.api.Twister2Job in project beam by apache.

the class Twister2Runner method run.

@Override
public PipelineResult run(Pipeline pipeline) {
    // create a worker and pass in the pipeline and then do the translation
    Twister2PipelineExecutionEnvironment env = new Twister2PipelineExecutionEnvironment(options);
    LOG.info("Translating pipeline to Twister2 program.");
    pipeline.replaceAll(getDefaultOverrides());
    // TODO(BEAM-10670): Use SDF read as default when we address performance issue.
    if (!ExperimentalOptions.hasExperiment(pipeline.getOptions(), "beam_fn_api")) {
        SplittableParDo.convertReadBasedSplittableDoFnsToPrimitiveReadsIfNecessary(pipeline);
    }
    env.translate(pipeline);
    setupSystem(options);
    Map configMap = new HashMap();
    JobConfig jobConfig = new JobConfig();
    if (isLocalMode(options)) {
        options.setParallelism(1);
        configMap.put(SIDEINPUTS, extractNames(env.getSideInputs()));
        configMap.put(LEAVES, extractNames(env.getLeaves()));
        configMap.put(GRAPH, env.getTSetGraph());
        configMap.put("twister2.network.buffer.size", 32000);
        configMap.put("twister2.network.sendBuffer.count", options.getParallelism());
        LOG.warning("Twister2 Local Mode currently only supports single worker");
    } else {
        jobConfig.put(SIDEINPUTS, extractNames(env.getSideInputs()));
        jobConfig.put(LEAVES, extractNames(env.getLeaves()));
        jobConfig.put(GRAPH, env.getTSetGraph());
    }
    Config config = ResourceAllocator.loadConfig(configMap);
    int workers = options.getParallelism();
    Twister2Job twister2Job = Twister2Job.newBuilder().setJobName(options.getJobName()).setWorkerClass(BeamBatchWorker.class).addComputeResource(options.getWorkerCPUs(), options.getRamMegaBytes(), workers).setConfig(jobConfig).build();
    Twister2JobState jobState;
    if (isLocalMode(options)) {
        jobState = LocalSubmitter.submitJob(twister2Job, config);
    } else {
        jobState = Twister2Submitter.submitJob(twister2Job, config);
    }
    Twister2PipelineResult result = new Twister2PipelineResult(jobState);
    return result;
}
Also used : HashMap(java.util.HashMap) LinkedHashMap(java.util.LinkedHashMap) Twister2JobState(edu.iu.dsc.tws.api.scheduler.Twister2JobState) Config(edu.iu.dsc.tws.api.config.Config) JobConfig(edu.iu.dsc.tws.api.JobConfig) HashMap(java.util.HashMap) LinkedHashMap(java.util.LinkedHashMap) Map(java.util.Map) JobConfig(edu.iu.dsc.tws.api.JobConfig) Twister2Job(edu.iu.dsc.tws.api.Twister2Job) PTransformOverride(org.apache.beam.sdk.runners.PTransformOverride)

Example 23 with Twister2Job

use of edu.iu.dsc.tws.api.Twister2Job in project beam by apache.

the class Twister2Runner method runTest.

public PipelineResult runTest(Pipeline pipeline) {
    // create a worker and pass in the pipeline and then do the translation
    Twister2PipelineExecutionEnvironment env = new Twister2PipelineExecutionEnvironment(options);
    LOG.info("Translating pipeline to Twister2 program.");
    pipeline.replaceAll(getDefaultOverrides());
    // TODO(BEAM-10670): Use SDF read as default when we address performance issue.
    if (!ExperimentalOptions.hasExperiment(pipeline.getOptions(), "beam_fn_api")) {
        SplittableParDo.convertReadBasedSplittableDoFnsToPrimitiveReadsIfNecessary(pipeline);
    }
    env.translate(pipeline);
    setupSystemTest(options);
    Map configMap = new HashMap();
    configMap.put(SIDEINPUTS, extractNames(env.getSideInputs()));
    configMap.put(LEAVES, extractNames(env.getLeaves()));
    configMap.put(GRAPH, env.getTSetGraph());
    configMap.put("twister2.network.buffer.size", 32000);
    configMap.put("twister2.network.sendBuffer.count", options.getParallelism());
    Config config = ResourceAllocator.loadConfig(configMap);
    JobConfig jobConfig = new JobConfig();
    int workers = options.getParallelism();
    Twister2Job twister2Job = Twister2Job.newBuilder().setJobName(options.getJobName()).setWorkerClass(BeamBatchWorker.class).addComputeResource(options.getWorkerCPUs(), options.getRamMegaBytes(), workers).setConfig(jobConfig).build();
    Twister2JobState jobState = LocalSubmitter.submitJob(twister2Job, config);
    Twister2PipelineResult result = new Twister2PipelineResult(jobState);
    // TODO: Need to fix the check for "RUNNING" once fix for this is done on Twister2 end.
    if (result.state == PipelineResult.State.FAILED) {
        throw new RuntimeException("Pipeline execution failed", jobState.getCause());
    }
    return result;
}
Also used : Twister2RuntimeException(edu.iu.dsc.tws.api.exceptions.Twister2RuntimeException) HashMap(java.util.HashMap) LinkedHashMap(java.util.LinkedHashMap) Twister2JobState(edu.iu.dsc.tws.api.scheduler.Twister2JobState) Config(edu.iu.dsc.tws.api.config.Config) JobConfig(edu.iu.dsc.tws.api.JobConfig) HashMap(java.util.HashMap) LinkedHashMap(java.util.LinkedHashMap) Map(java.util.Map) JobConfig(edu.iu.dsc.tws.api.JobConfig) Twister2Job(edu.iu.dsc.tws.api.Twister2Job)

Example 24 with Twister2Job

use of edu.iu.dsc.tws.api.Twister2Job in project twister2 by DSC-SPIDAL.

the class JobMasterExample method main.

/**
 * this main method is for locally testing only
 * A JobMaster instance is started locally on the default port:
 * edu.iu.dsc.tws.master.JobMasterContext.JOB_MASTER_PORT_DEFAULT = 11011
 * <p>
 * numberOfWorkers to join is expected as a parameter
 * <p>
 * When all workers joined and all have sent completed messages,
 * this server also completes and exits
 * <p>
 * En example usage of JobMaster can be seen in:
 * edu.iu.dsc.tws.rsched.schedulers.k8s.master.JobMasterStarter
 */
public static void main(String[] args) {
    if (args.length != 1) {
        LOG.info("usage: java JobMasterExample numberOfWorkers");
        return;
    }
    int numberOfWorkers = Integer.parseInt(args[0]);
    String host = "0.0.0.0";
    // we assume that the twister2Home is the current directory
    // String configDir = "../twister2/config/src/yaml/";
    String configDir = "";
    String twister2Home = Paths.get(configDir).toAbsolutePath().toString();
    Config config = ConfigLoader.loadConfig(twister2Home, "conf", "kubernetes");
    config = JobMasterClientExample.updateConfig(config, config, host);
    LOG.info("Loaded: " + config.size() + " configuration parameters.");
    // Twister2Job twister2Job = Twister2Job.loadTwister2Job(config, null);
    Twister2Job twister2Job = Twister2Job.newBuilder().setJobName("hello-world-job").setWorkerClass(HelloWorld.class).addComputeResource(.2, 128, numberOfWorkers).build();
    twister2Job.setUserName(System.getProperty("user.name"));
    JobAPI.Job job = twister2Job.serialize();
    LOG.info("JobID: " + job.getJobId());
    JobMasterAPI.JobMasterState initialState = JobMasterAPI.JobMasterState.JM_STARTED;
    JobMasterStarter.job = job;
    if (ZKContext.isZooKeeperServerUsed(config)) {
        if ("start".equalsIgnoreCase(args[0])) {
            JobMasterStarter.initializeZooKeeper(config, job.getJobId(), host, initialState);
        } else if ("restart".equalsIgnoreCase(args[0])) {
            initialState = JobMasterAPI.JobMasterState.JM_RESTARTED;
            JobMasterStarter.initializeZooKeeper(config, job.getJobId(), host, initialState);
            job = JobMasterStarter.job;
        } else {
            LOG.info("usage: java JobMasterExample start/restart");
            return;
        }
    }
    // write jobID to file
    String dir = System.getProperty("user.home") + "/.twister2";
    if (!FileUtils.isDirectoryExists(dir)) {
        FileUtils.createDirectory(dir);
    }
    String filename = dir + "/last-job-id.txt";
    FileUtils.writeToFile(filename, (job.getJobId() + "").getBytes(), true);
    LOG.info("Written jobID to file: " + job.getJobId());
    String ip = null;
    try {
        ip = Inet4Address.getLocalHost().getHostAddress();
    } catch (UnknownHostException e) {
        LOG.log(Level.SEVERE, e.getMessage(), e);
        return;
    }
    JobMasterAPI.NodeInfo jobMasterNode = NodeInfoUtils.createNodeInfo(ip, null, null);
    KubernetesController controller = KubernetesController.init("default");
    K8sScaler k8sScaler = new K8sScaler(config, job, controller);
    IJobTerminator jobTerminator = new NullTerminator();
    JobMaster jobMaster = new JobMaster(config, host, jobTerminator, job, jobMasterNode, k8sScaler, initialState);
    try {
        // jobMaster.startJobMasterThreaded();
        jobMaster.startJobMasterBlocking();
    } catch (Twister2Exception e) {
        LOG.log(Level.SEVERE, "Exception when starting Job master: ", e);
        throw new RuntimeException(e);
    }
    LOG.info("Threaded Job Master started:" + "\nnumberOfWorkers: " + job.getNumberOfWorkers() + "\njobID: " + job.getJobId());
}
Also used : JobMaster(edu.iu.dsc.tws.master.server.JobMaster) Twister2Exception(edu.iu.dsc.tws.api.exceptions.Twister2Exception) UnknownHostException(java.net.UnknownHostException) Config(edu.iu.dsc.tws.api.config.Config) KubernetesController(edu.iu.dsc.tws.rsched.schedulers.k8s.KubernetesController) JobAPI(edu.iu.dsc.tws.proto.system.job.JobAPI) Twister2Job(edu.iu.dsc.tws.api.Twister2Job) K8sScaler(edu.iu.dsc.tws.rsched.schedulers.k8s.driver.K8sScaler) JobMasterAPI(edu.iu.dsc.tws.proto.jobmaster.JobMasterAPI) IJobTerminator(edu.iu.dsc.tws.master.IJobTerminator) NullTerminator(edu.iu.dsc.tws.rsched.schedulers.NullTerminator)

Example 25 with Twister2Job

use of edu.iu.dsc.tws.api.Twister2Job in project twister2 by DSC-SPIDAL.

the class BasicKubernetesJob method submitJob.

/**
 * submit the job
 */
public static void submitJob(Config config) {
    // build JobConfig
    HashMap<String, Object> configurations = new HashMap<>();
    configurations.put(SchedulerContext.THREADS_PER_WORKER, 8);
    JobConfig jobConfig = new JobConfig();
    jobConfig.putAll(configurations);
    // It gets: job-name, worker-class and ComputeResource list from that file
    Twister2Job twister2Job = Twister2Job.loadTwister2Job(config, jobConfig);
    // now submit the job
    Twister2Submitter.submitJob(twister2Job, config);
}
Also used : HashMap(java.util.HashMap) JobConfig(edu.iu.dsc.tws.api.JobConfig) Twister2Job(edu.iu.dsc.tws.api.Twister2Job)

Aggregations

Twister2Job (edu.iu.dsc.tws.api.Twister2Job)39 JobConfig (edu.iu.dsc.tws.api.JobConfig)27 Config (edu.iu.dsc.tws.api.config.Config)23 HashMap (java.util.HashMap)11 CommandLine (org.apache.commons.cli.CommandLine)8 CommandLineParser (org.apache.commons.cli.CommandLineParser)7 DefaultParser (org.apache.commons.cli.DefaultParser)7 Options (org.apache.commons.cli.Options)7 Twister2JobState (edu.iu.dsc.tws.api.scheduler.Twister2JobState)4 DataFlowJobConfig (edu.iu.dsc.tws.task.cdfw.DataFlowJobConfig)4 JobAPI (edu.iu.dsc.tws.proto.system.job.JobAPI)3 CDFWWorker (edu.iu.dsc.tws.task.impl.cdfw.CDFWWorker)3 Twister2RuntimeException (edu.iu.dsc.tws.api.exceptions.Twister2RuntimeException)2 KubernetesController (edu.iu.dsc.tws.rsched.schedulers.k8s.KubernetesController)2 LinkedHashMap (java.util.LinkedHashMap)2 Map (java.util.Map)2 DataObject (edu.iu.dsc.tws.api.dataset.DataObject)1 DriverJobState (edu.iu.dsc.tws.api.driver.DriverJobState)1 Twister2Exception (edu.iu.dsc.tws.api.exceptions.Twister2Exception)1 HelloWorld (edu.iu.dsc.tws.examples.basic.HelloWorld)1