Search in sources :

Example 1 with ResourceDescription

use of org.gridlab.gat.resources.ResourceDescription in project compss by bsc-wdc.

the class SshLsfResourceBrokerAdaptor method createBsubScript.

private java.io.File createBsubScript(JobDescription description, String returnValueFile, int nproc) throws GATInvocationException {
    // Adding bsub options
    String Queue = null;
    long Time = -1;
    Integer cpus = null;
    String jobname = null;
    java.io.File temp;
    LSFScriptWriter job = null;
    HashMap<String, Object> rd_HashMap = null;
    SoftwareDescription sd = description.getSoftwareDescription();
    ResourceDescription rd = description.getResourceDescription();
    // Corrected initialization of rd_HashMap: rd may be null ... --Ceriel
    if (rd != null) {
        rd_HashMap = (HashMap<String, Object>) rd.getDescription();
    }
    if (rd_HashMap == null) {
        rd_HashMap = new HashMap<String, Object>();
    }
    // try {
    temp = new java.io.File("lsf" + Math.random());
    try {
        job = new LSFScriptWriter(new BufferedWriter(new FileWriter(temp)));
        String userScript = (String) gatContext.getPreferences().get(SSHLSF_SCRIPT);
        if (userScript != null) {
            // a specified job script overrides everything, except for
            // pre-staging, post-staging,
            // and exit status.
            BufferedReader f = new BufferedReader(new FileReader(userScript));
            for (; ; ) {
                String s = f.readLine();
                if (s == null) {
                    break;
                }
                job.print(s + "\n");
            }
        } else {
            job.print("#!/bin/sh\n");
            job.print("# bsub script automatically generated by GAT SshLsf adaptor\n");
            // Resources: queue, walltime, memory size, et cetera.
            Queue = (String) rd_HashMap.get("machine.queue");
            if (Queue == null) {
                Queue = sd.getStringAttribute(SoftwareDescription.JOB_QUEUE, null);
            }
            if (Queue != null) {
                job.addOption("q", Queue);
            }
            Time = sd.getLongAttribute(SoftwareDescription.WALLTIME_MAX, -1L);
            cpus = (Integer) rd_HashMap.get(HardwareResourceDescription.CPU_COUNT);
            if (cpus == null) {
                cpus = sd.getIntAttribute("coreCount", 1);
            }
            job.addOption("n", cpus);
            // In a single node
            job.addOption("R", "\"span[ptile=" + cpus + "]\"");
            if (Time > 0) {
                // Reformat time.
                int minutes = (int) (Time % 60);
                job.addOption("W", minutes);
            } else {
                job.addOption("W", 60);
            }
            String nativeFlags = null;
            Object o = rd == null ? null : rd.getResourceAttribute(SSHLSF_NATIVE_FLAGS);
            if (o != null && o instanceof String) {
                nativeFlags = (String) o;
            } else {
                String s = sd == null ? null : sd.getStringAttribute(SSHLSF_NATIVE_FLAGS, null);
                if (s != null) {
                    nativeFlags = s;
                } else {
                    o = gatContext.getPreferences().get(SSHLSF_NATIVE_FLAGS);
                    if (o != null && o instanceof String) {
                        nativeFlags = (String) o;
                    }
                }
            }
            if (nativeFlags != null) {
                String[] splits = nativeFlags.split("##");
                for (String s : splits) {
                    job.addString(s);
                }
            }
            String path = sd.getStringAttribute(SoftwareDescription.SANDBOX_ROOT, "");
            if (!path.isEmpty() && !path.endsWith(File.separator)) {
                path = path + File.separator;
            }
            // Set working dir.
            // job.addOption("cwd", path);
            // Name for the job.
            jobname = (String) rd_HashMap.get("Jobname");
            if (jobname == null) {
                jobname = brokerURI.getUserInfo();
                if (jobname == null || "".equals(jobname)) {
                    jobname = "compss_remotejob_" + System.getProperty("user.name");
                }
            }
            if (jobname != null)
                job.addOption("J", jobname);
            if (sd.getStdout() != null) {
                job.addOption("oo", path + sd.getStdout().getName());
            }
            if (sd.getStderr() != null) {
                job.addOption("eo", path + sd.getStderr().getName());
            }
            addScriptExecution(job, sd, rd);
        }
        job.print("echo retvalue = $? > " + returnValueFile + "\n");
    } catch (Throwable e) {
        throw new GATInvocationException("Cannot create temporary bsub file " + temp.getAbsolutePath(), e);
    } finally {
        if (job != null)
            job.close();
    }
    return temp;
}
Also used : FileWriter(java.io.FileWriter) File(java.io.File) SoftwareDescription(org.gridlab.gat.resources.SoftwareDescription) BufferedWriter(java.io.BufferedWriter) GATInvocationException(org.gridlab.gat.GATInvocationException) HardwareResourceDescription(org.gridlab.gat.resources.HardwareResourceDescription) ResourceDescription(org.gridlab.gat.resources.ResourceDescription) BufferedReader(java.io.BufferedReader) FileReader(java.io.FileReader)

Example 2 with ResourceDescription

use of org.gridlab.gat.resources.ResourceDescription in project compss by bsc-wdc.

the class GATJob method prepareJob.

private JobDescription prepareJob() throws Exception {
    // Get the information related to the job
    logger.debug("Preparing GAT Job " + this.jobId);
    TaskDescription taskParams = this.taskParams;
    String targetPath = getResourceNode().getInstallDir();
    String targetHost = getResourceNode().getHost();
    String targetUser = getResourceNode().getUser();
    if (userNeeded && !targetUser.isEmpty()) {
        targetUser += "@";
    } else {
        targetUser = "";
    }
    SoftwareDescription sd = new SoftwareDescription();
    sd.setExecutable(targetPath + WORKER_SCRIPT_PATH + WORKER_SCRIPT_NAME);
    ArrayList<String> lArgs = new ArrayList<String>();
    // Common arguments: language working_dir lib_path num_obsolete [obs1... obsN] tracing [event_type task_id
    // slot_id]
    lArgs.add(LANG);
    lArgs.add(getResourceNode().getWorkingDir());
    lArgs.add(getResourceNode().getLibPath());
    LogicalData[] obsoleteFiles = getResource().pollObsoletes();
    if (obsoleteFiles != null) {
        lArgs.add("" + obsoleteFiles.length);
        for (LogicalData ld : obsoleteFiles) {
            String renaming = ld.getName();
            lArgs.add(renaming);
        }
    } else {
        lArgs.add("0");
    }
    // Check sandbox working dir
    boolean isSpecific = false;
    String sandboxDir = null;
    AbstractMethodImplementation absImpl = (AbstractMethodImplementation) this.impl;
    switch(absImpl.getMethodType()) {
        case BINARY:
            BinaryImplementation binaryImpl = (BinaryImplementation) absImpl;
            sandboxDir = binaryImpl.getWorkingDir();
            isSpecific = true;
            break;
        case MPI:
            MPIImplementation mpiImpl = (MPIImplementation) absImpl;
            sandboxDir = mpiImpl.getWorkingDir();
            isSpecific = true;
            break;
        case DECAF:
            DecafImplementation decafImpl = (DecafImplementation) absImpl;
            sandboxDir = decafImpl.getWorkingDir();
            isSpecific = true;
            break;
        case OMPSS:
            OmpSsImplementation ompssImpl = (OmpSsImplementation) absImpl;
            sandboxDir = ompssImpl.getWorkingDir();
            isSpecific = true;
            break;
        case OPENCL:
            OpenCLImplementation openclImpl = (OpenCLImplementation) absImpl;
            sandboxDir = openclImpl.getWorkingDir();
            isSpecific = true;
            break;
        case METHOD:
            sandboxDir = null;
            break;
    }
    if (sandboxDir == null || sandboxDir.isEmpty() || sandboxDir.equals(Constants.UNASSIGNED)) {
        sandboxDir = getResourceNode().getWorkingDir() + File.separator + "sandBox" + File.separator + "job_" + this.jobId;
        isSpecific = false;
    }
    // Processing parameters to get symlinks pairs to create (symlinks) and how to pass parameters in the GAT
    // Job(paramArgs)
    ArrayList<String> symlinks = new ArrayList<>();
    ArrayList<String> paramArgs = new ArrayList<>();
    processParameters(sandboxDir, symlinks, paramArgs);
    // Adding info to create symlinks between renamed files and original names
    lArgs.add(Boolean.toString(isSpecific));
    lArgs.add(sandboxDir);
    if (symlinks.size() > 0) {
        lArgs.add(String.valueOf(symlinks.size()));
        lArgs.addAll(symlinks);
    } else {
        lArgs.add("0");
    }
    lArgs.add(Boolean.toString(Tracer.isActivated()));
    lArgs.add(getHostName());
    if (debug) {
        logger.debug("hostName " + getHostName());
    }
    if (Tracer.isActivated()) {
        // event type
        lArgs.add(String.valueOf(Tracer.getTaskEventsType()));
        // task id
        lArgs.add(String.valueOf(this.taskParams.getId() + 1));
        int slot = Tracer.getNextSlot(targetHost);
        // slot id
        lArgs.add(String.valueOf(slot));
        sd.addAttribute("slot", slot);
    }
    // Language-dependent arguments: taskSandbox_dir app_dir classpath pythonpath debug storage_conf
    // method_impl_type method_impl_params
    // numSlaves [slave1,..,slaveN] numCus
    // has_target num_params par_type_1 par_1 ... par_type_n par_n
    lArgs.add(sandboxDir);
    lArgs.add(getResourceNode().getAppDir());
    lArgs.add(getClasspath());
    lArgs.add(getPythonpath());
    lArgs.add(String.valueOf(debug));
    lArgs.add(STORAGE_CONF);
    lArgs.add(String.valueOf(absImpl.getMethodType()));
    switch(absImpl.getMethodType()) {
        case METHOD:
            MethodImplementation methodImpl = (MethodImplementation) absImpl;
            lArgs.add(methodImpl.getDeclaringClass());
            String methodName = methodImpl.getAlternativeMethodName();
            if (methodName == null || methodName.isEmpty()) {
                methodName = taskParams.getName();
            }
            lArgs.add(methodName);
            break;
        case MPI:
            MPIImplementation mpiImpl = (MPIImplementation) absImpl;
            lArgs.add(mpiImpl.getMpiRunner());
            lArgs.add(mpiImpl.getBinary());
            break;
        case DECAF:
            DecafImplementation decafImpl = (DecafImplementation) absImpl;
            lArgs.add(targetPath + DecafImplementation.SCRIPT_PATH);
            String dfScript = decafImpl.getDfScript();
            if (!dfScript.startsWith(File.separator)) {
                String appPath = getResourceNode().getAppDir();
                dfScript = appPath + File.separator + dfScript;
            }
            lArgs.add(dfScript);
            String dfExecutor = decafImpl.getDfExecutor();
            if (dfExecutor == null || dfExecutor.isEmpty() || dfExecutor.equals(Constants.UNASSIGNED)) {
                dfExecutor = "executor.sh";
            }
            if (!dfExecutor.startsWith(File.separator) && !dfExecutor.startsWith("./")) {
                dfExecutor = "./" + dfExecutor;
            }
            lArgs.add(dfExecutor);
            String dfLib = decafImpl.getDfLib();
            if (dfLib == null || dfLib.isEmpty()) {
                dfLib = Constants.UNASSIGNED;
            }
            lArgs.add(dfLib);
            lArgs.add(decafImpl.getMpiRunner());
            break;
        case OMPSS:
            OmpSsImplementation ompssImpl = (OmpSsImplementation) absImpl;
            lArgs.add(ompssImpl.getBinary());
            break;
        case OPENCL:
            OpenCLImplementation openclImpl = (OpenCLImplementation) absImpl;
            lArgs.add(openclImpl.getKernel());
            break;
        case BINARY:
            BinaryImplementation binaryImpl = (BinaryImplementation) absImpl;
            lArgs.add(binaryImpl.getBinary());
            break;
    }
    // Slave nodes and cus description
    lArgs.add(String.valueOf(slaveWorkersNodeNames.size()));
    lArgs.addAll(slaveWorkersNodeNames);
    lArgs.add(String.valueOf(((MethodResourceDescription) this.impl.getRequirements()).getTotalCPUComputingUnits()));
    // Add parameter arguments already processed
    lArgs.addAll(paramArgs);
    // Conversion vector -> array
    String[] arguments = new String[lArgs.size()];
    arguments = lArgs.toArray(arguments);
    try {
        sd.setArguments(arguments);
    } catch (NullPointerException e) {
        StringBuilder sb = new StringBuilder("Null argument parameter of job " + this.jobId + " " + absImpl.getMethodDefinition() + "\n");
        int i = 0;
        for (Parameter param : taskParams.getParameters()) {
            sb.append("Parameter ").append(i).append("\n");
            DataType type = param.getType();
            sb.append("\t Type: ").append(param.getType()).append("\n");
            if (type == DataType.FILE_T || type == DataType.OBJECT_T) {
                DependencyParameter dPar = (DependencyParameter) param;
                DataAccessId dAccId = dPar.getDataAccessId();
                sb.append("\t Target: ").append(dPar.getDataTarget()).append("\n");
                if (type == DataType.OBJECT_T) {
                    if (dAccId instanceof RAccessId) {
                        sb.append("\t Direction: " + "R").append("\n");
                    } else {
                        // for the worker to know it must write the object to disk
                        sb.append("\t Direction: " + "W").append("\n");
                    }
                }
            } else if (type == DataType.STRING_T) {
                BasicTypeParameter btParS = (BasicTypeParameter) param;
                // Check spaces
                String value = btParS.getValue().toString();
                int numSubStrings = value.split(" ").length;
                sb.append("\t Num Substrings: " + Integer.toString(numSubStrings)).append("\n");
                sb.append("\t Value:" + value).append("\n");
            } else {
                // Basic types
                BasicTypeParameter btParB = (BasicTypeParameter) param;
                sb.append("\t Value: " + btParB.getValue().toString()).append("\n");
            }
            i++;
        }
        logger.error(sb.toString());
        listener.jobFailed(this, JobEndStatus.SUBMISSION_FAILED);
    }
    sd.addAttribute("jobId", jobId);
    // JEA Changed to allow execution in MN
    sd.addAttribute(SoftwareDescription.WALLTIME_MAX, absImpl.getRequirements().getWallClockLimit());
    if (absImpl.getRequirements().getHostQueues().size() > 0) {
        sd.addAttribute(SoftwareDescription.JOB_QUEUE, absImpl.getRequirements().getHostQueues().get(0));
    }
    sd.addAttribute("coreCount", absImpl.getRequirements().getTotalCPUComputingUnits());
    sd.addAttribute("gpuCount", absImpl.getRequirements().getTotalGPUComputingUnits());
    sd.addAttribute("fpgaCount", absImpl.getRequirements().getTotalFPGAComputingUnits());
    sd.addAttribute(SoftwareDescription.MEMORY_MAX, absImpl.getRequirements().getMemorySize());
    // sd.addAttribute(SoftwareDescription.SANDBOX_ROOT, "/tmp/");
    sd.addAttribute(SoftwareDescription.SANDBOX_ROOT, getResourceNode().getWorkingDir());
    sd.addAttribute(SoftwareDescription.SANDBOX_USEROOT, "true");
    sd.addAttribute(SoftwareDescription.SANDBOX_DELETE, "false");
    /*
         * sd.addAttribute(SoftwareDescription.SANDBOX_PRESTAGE_STDIN, "false");
         * sd.addAttribute(SoftwareDescription.SANDBOX_POSTSTAGE_STDOUT, "false");
         * sd.addAttribute(SoftwareDescription.SANDBOX_POSTSTAGE_STDERR, "false");
         */
    if (debug) {
        // Set standard output file for job
        File outFile = GAT.createFile(context, Protocol.ANY_URI.getSchema() + File.separator + JOBS_DIR + "job" + jobId + "_" + this.getHistory() + ".out");
        sd.setStdout(outFile);
    }
    if (debug || usingGlobus) {
        // Set standard error file for job
        File errFile = GAT.createFile(context, Protocol.ANY_URI.getSchema() + File.separator + JOBS_DIR + "job" + jobId + "_" + this.getHistory() + ".err");
        sd.setStderr(errFile);
    }
    Map<String, Object> attributes = new HashMap<String, Object>();
    attributes.put(RES_ATTR, Protocol.ANY_URI.getSchema() + targetUser + targetHost);
    attributes.put("Jobname", "compss_remote_job_" + jobId);
    ResourceDescription rd = new HardwareResourceDescription(attributes);
    if (debug) {
        logger.debug("Ready to submit job " + jobId + ":");
        logger.debug("  * Host: " + targetHost);
        logger.debug("  * Executable: " + sd.getExecutable());
        StringBuilder sb = new StringBuilder("  - Arguments:");
        for (String arg : sd.getArguments()) {
            sb.append(" ").append(arg);
        }
        logger.debug(sb.toString());
    }
    JobDescription jd = new JobDescription(sd, rd);
    // jd.setProcessCount(method.getRequirements().getProcessorCoreCount());
    return jd;
}
Also used : MethodImplementation(es.bsc.compss.types.implementations.MethodImplementation) AbstractMethodImplementation(es.bsc.compss.types.implementations.AbstractMethodImplementation) MPIImplementation(es.bsc.compss.types.implementations.MPIImplementation) RAccessId(es.bsc.compss.types.data.DataAccessId.RAccessId) HashMap(java.util.HashMap) ArrayList(java.util.ArrayList) DependencyParameter(es.bsc.compss.types.parameter.DependencyParameter) BasicTypeParameter(es.bsc.compss.types.parameter.BasicTypeParameter) OpenCLImplementation(es.bsc.compss.types.implementations.OpenCLImplementation) JobDescription(org.gridlab.gat.resources.JobDescription) TaskDescription(es.bsc.compss.types.TaskDescription) DecafImplementation(es.bsc.compss.types.implementations.DecafImplementation) DataType(es.bsc.compss.types.annotations.parameter.DataType) MethodResourceDescription(es.bsc.compss.types.resources.MethodResourceDescription) BinaryImplementation(es.bsc.compss.types.implementations.BinaryImplementation) AbstractMethodImplementation(es.bsc.compss.types.implementations.AbstractMethodImplementation) HardwareResourceDescription(org.gridlab.gat.resources.HardwareResourceDescription) SoftwareDescription(org.gridlab.gat.resources.SoftwareDescription) LogicalData(es.bsc.compss.types.data.LogicalData) MethodResourceDescription(es.bsc.compss.types.resources.MethodResourceDescription) HardwareResourceDescription(org.gridlab.gat.resources.HardwareResourceDescription) ResourceDescription(org.gridlab.gat.resources.ResourceDescription) Parameter(es.bsc.compss.types.parameter.Parameter) DependencyParameter(es.bsc.compss.types.parameter.DependencyParameter) BasicTypeParameter(es.bsc.compss.types.parameter.BasicTypeParameter) File(org.gridlab.gat.io.File) OmpSsImplementation(es.bsc.compss.types.implementations.OmpSsImplementation) DataAccessId(es.bsc.compss.types.data.DataAccessId)

Example 3 with ResourceDescription

use of org.gridlab.gat.resources.ResourceDescription in project compss by bsc-wdc.

the class AdvertJob method main.

public static void main(String[] args) throws Exception {
    try {
        GATContext c = new GATContext();
        Preferences prefs = new Preferences();
        prefs.put("File.adaptor.name", "local,commandlinessh");
        prefs.put("job.stop.on.exit", "false");
        c.addPreferences(prefs);
        SoftwareDescription sd = new SoftwareDescription();
        sd.setExecutable("/bin/sleep");
        sd.setArguments("100");
        // stdout & stderr
        File stdout = GAT.createFile(c, "std.out");
        File stderr = GAT.createFile(c, "std.err");
        sd.setStderr(stderr);
        sd.setStdout(stdout);
        ResourceDescription rd = new HardwareResourceDescription();
        JobDescription jd = new JobDescription(sd, rd);
        ResourceBroker broker = GAT.createResourceBroker(c, new URI("sshsge://fs0.das3.cs.vu.nl"));
        Job job = broker.submitJob(jd);
        AdvertService a = GAT.createAdvertService(c);
        MetaData m = new MetaData();
        m.put("name", "testJob");
        a.add(job, m, "/rob/testJob");
        a.exportDataBase(new URI("file:///mydb"));
        GAT.end();
        System.exit(0);
    } catch (Throwable e) {
        e.printStackTrace();
    }
}
Also used : GATContext(org.gridlab.gat.GATContext) ResourceBroker(org.gridlab.gat.resources.ResourceBroker) HardwareResourceDescription(org.gridlab.gat.resources.HardwareResourceDescription) AdvertService(org.gridlab.gat.advert.AdvertService) URI(org.gridlab.gat.URI) SoftwareDescription(org.gridlab.gat.resources.SoftwareDescription) JobDescription(org.gridlab.gat.resources.JobDescription) ResourceDescription(org.gridlab.gat.resources.ResourceDescription) HardwareResourceDescription(org.gridlab.gat.resources.HardwareResourceDescription) MetaData(org.gridlab.gat.advert.MetaData) Preferences(org.gridlab.gat.Preferences) Job(org.gridlab.gat.resources.Job) File(org.gridlab.gat.io.File)

Aggregations

HardwareResourceDescription (org.gridlab.gat.resources.HardwareResourceDescription)3 ResourceDescription (org.gridlab.gat.resources.ResourceDescription)3 SoftwareDescription (org.gridlab.gat.resources.SoftwareDescription)3 File (org.gridlab.gat.io.File)2 JobDescription (org.gridlab.gat.resources.JobDescription)2 TaskDescription (es.bsc.compss.types.TaskDescription)1 DataType (es.bsc.compss.types.annotations.parameter.DataType)1 DataAccessId (es.bsc.compss.types.data.DataAccessId)1 RAccessId (es.bsc.compss.types.data.DataAccessId.RAccessId)1 LogicalData (es.bsc.compss.types.data.LogicalData)1 AbstractMethodImplementation (es.bsc.compss.types.implementations.AbstractMethodImplementation)1 BinaryImplementation (es.bsc.compss.types.implementations.BinaryImplementation)1 DecafImplementation (es.bsc.compss.types.implementations.DecafImplementation)1 MPIImplementation (es.bsc.compss.types.implementations.MPIImplementation)1 MethodImplementation (es.bsc.compss.types.implementations.MethodImplementation)1 OmpSsImplementation (es.bsc.compss.types.implementations.OmpSsImplementation)1 OpenCLImplementation (es.bsc.compss.types.implementations.OpenCLImplementation)1 BasicTypeParameter (es.bsc.compss.types.parameter.BasicTypeParameter)1 DependencyParameter (es.bsc.compss.types.parameter.DependencyParameter)1 Parameter (es.bsc.compss.types.parameter.Parameter)1