Examples with Twister2RuntimeException edu.iu.dsc.tws.api.exceptions.Twister2RuntimeException used on opensource projects

Example 16 with Twister2RuntimeException

use of edu.iu.dsc.tws.api.exceptions.Twister2RuntimeException in project twister2 by DSC-SPIDAL.

the class K8sWorkerUtils method initLogger.

/**
 * initialize the logger
 * entityName can be "jobMaster", "mpiMaster", etc.
 */
public static void initLogger(Config cnfg, String entityName, boolean pvExists) {
    // if logging to file is not requested, do nothing
    if (!LoggingContext.fileLoggingRequested()) {
        return;
    }
    if (!pvExists && "persistent".equalsIgnoreCase(LoggingContext.loggingStorageType(cnfg))) {
        LOG.warning("Persistent logging is requested but no PersistentVolume provided. " + "Not logging to file");
        return;
    }
    if (LoggingContext.redirectSysOutErr()) {
        LOG.warning("Redirecting System.out and System.err to the log file. " + "Check the log file for the upcoming log messages. ");
    }
    String logDirName = LoggingContext.loggingDir(cnfg);
    File logDir = new File(logDirName);
    // refresh parent directory the cache
    logDir.getParentFile().list();
    if (!logDir.exists()) {
        if (!logDir.mkdirs()) {
            throw new Twister2RuntimeException("Failed to create the log directory: " + logDir);
        }
    }
    String logFileName = entityName;
    LoggingHelper.setupLogging(cnfg, logDirName, logFileName);
    String logFileWithPath = logDirName + "/" + logFileName + ".log.0";
    LOG.info("Logging to file initialized: " + logFileWithPath);
}

Example 17 with Twister2RuntimeException

use of edu.iu.dsc.tws.api.exceptions.Twister2RuntimeException in project twister2 by DSC-SPIDAL.

the class Twister2Submitter method restartJob.

/**
 * Restart a Twister2 job
 */
public static Twister2JobState restartJob(String jobID, Config config) {
    // job package filename from failed submission
    String prevJobDir = FsContext.uploaderJobDirectory(config) + File.separator + jobID;
    String jobPackage = prevJobDir + File.separator + SchedulerContext.jobPackageFileName(config);
    Path jobPackageFile = Paths.get(jobPackage);
    if (Files.notExists(jobPackageFile)) {
        LOG.severe("Job Package File does not exist: " + jobPackage);
        return new Twister2JobState(false);
    }
    // unpack the previous job package to a temp directory
    Path tempDirPath;
    try {
        tempDirPath = Files.createTempDirectory(jobID);
    } catch (IOException e) {
        throw new Twister2RuntimeException("Can not create temp directory", e);
    }
    // todo: we can exclude user-job-file from being unpacked
    // usually that is the lastest file, so we can be more efficient
    TarGzipPacker.unpack(jobPackageFile, tempDirPath);
    // load Job object
    String unpackedJobDir = tempDirPath + File.separator + Context.JOB_ARCHIVE_DIRECTORY;
    String jobFile = unpackedJobDir + File.separator + SchedulerContext.createJobDescriptionFileName(jobID);
    JobAPI.Job job = JobUtils.readJobFile(jobFile);
    // load previous configurations
    Config prevConfig = ConfigLoader.loadConfig(Context.twister2Home(config), unpackedJobDir, Context.clusterType(config));
    // delete temp directory
    try {
        Files.delete(tempDirPath);
        LOG.info("Unpacked job directory deleted: " + tempDirPath);
    } catch (IOException e) {
        LOG.warning("Exception when deleting temp directory: " + tempDirPath);
    }
    URI packageURI = null;
    try {
        packageURI = new URI(prevJobDir);
    } catch (URISyntaxException e) {
        throw new Twister2RuntimeException("Can not ceate URI for directory: " + prevJobDir, e);
    }
    // add restore parameter
    // local packages path
    prevConfig = Config.newBuilder().putAll(prevConfig).put(CheckpointingContext.CHECKPOINTING_RESTORE_JOB, true).put(SchedulerContext.TEMPORARY_PACKAGES_PATH, prevJobDir).put(SchedulerContext.USER_JOB_FILE, job.getJobFormat().getJobFile()).put(SchedulerContext.JOB_PACKAGE_URI, packageURI).put(Context.TWISTER2_HOME.getKey(), Context.twister2Home(config)).put(Context.JOB_ID, jobID).put(Context.TWISTER2_CLUSTER_TYPE, Context.clusterType(config)).build();
    writeJobIdToFile(jobID);
    printJobInfo(job, prevConfig);
    // launch the launcher
    ResourceAllocator resourceAllocator = new ResourceAllocator(prevConfig, job);
    return resourceAllocator.resubmitJob();
}

Example 18 with Twister2RuntimeException

use of edu.iu.dsc.tws.api.exceptions.Twister2RuntimeException in project twister2 by DSC-SPIDAL.

the class ZKMasterController method workerFailed.

@Override
public void workerFailed(int workerID) {
    JobMasterAPI.WorkerFailed workerFailed = JobMasterAPI.WorkerFailed.newBuilder().setWorkerID(workerID).build();
    JobMasterAPI.JobEvent jobEvent = JobMasterAPI.JobEvent.newBuilder().setFailed(workerFailed).build();
    try {
        ZKEventsManager.publishEvent(client, rootPath, jobID, jobEvent);
    } catch (Twister2Exception e) {
        throw new Twister2RuntimeException(e);
    }
}

Example 19 with Twister2RuntimeException

use of edu.iu.dsc.tws.api.exceptions.Twister2RuntimeException in project twister2 by DSC-SPIDAL.

the class ZKMasterController method jmRestarted.

public void jmRestarted() {
    // generate en event and inform all other workers
    JobMasterAPI.JobMasterRestarted jmRestarted = JobMasterAPI.JobMasterRestarted.newBuilder().setNumberOfWorkers(numberOfWorkers).setJmAddress(jmAddress).build();
    JobMasterAPI.JobEvent jobEvent = JobMasterAPI.JobEvent.newBuilder().setJmRestarted(jmRestarted).build();
    try {
        ZKEventsManager.publishEvent(client, rootPath, jobID, jobEvent);
    } catch (Twister2Exception e) {
        throw new Twister2RuntimeException(e);
    }
}

Example 20 with Twister2RuntimeException

use of edu.iu.dsc.tws.api.exceptions.Twister2RuntimeException in project twister2 by DSC-SPIDAL.

the class AbstractPythonProcessor method initialValue.

@Override
protected Jep initialValue() {
    try {
        // get the jep instance for this thread
        Jep jep = JepInstance.getInstance();
        // todo temporary workaround for JepArray issue.
        // This won't be a significant performance bottleneck though
        String lambdaString = Base64.getEncoder().encodeToString(bytes);
        jep.set("func_bin", lambdaString);
        jep.eval(this.objectId + " = cp.loads(base64.b64decode(func_bin))");
        jep.eval("del func_bin");
        return jep;
    } catch (JepException e) {
        throw new Twister2RuntimeException("Error in building lambda function", e);
    }
}