Examples with Job org.gridlab.gat.resources.Job used on opensource projects

Example 1 with Job

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

the class GATJob method processMetricEvent.

// MetricListener interface implementation
@Override
public void processMetricEvent(MetricEvent value) {
    Job job = (Job) value.getSource();
    JobState newJobState = (JobState) value.getValue();
    JobDescription jd = (JobDescription) job.getJobDescription();
    SoftwareDescription sd = jd.getSoftwareDescription();
    Integer jobId = (Integer) sd.getAttributes().get("jobId");
    logger.debug("Processing job ID = " + jobId);
    /*
         * Check if either the job has finished or there has been a submission error. We don't care about other state
         * transitions
         */
    if (newJobState == JobState.STOPPED) {
        if (Tracer.isActivated()) {
            Integer slot = (Integer) sd.getAttributes().get("slot");
            String host = getResourceNode().getHost();
            Tracer.freeSlot(host, slot);
        }
        /*
             * We must check whether the chosen adaptor is globus In that case, since globus doesn't provide the exit
             * status of a job, we must examine the standard error file
             */
        try {
            if (usingGlobus) {
                File errFile = sd.getStderr();
                // Error file should always be in the same host as the IT
                File localFile = GAT.createFile(context, errFile.toGATURI());
                if (localFile.length() > 0) {
                    GATjob = null;
                    RUNNING_JOBS.remove(this);
                    ErrorManager.warn("Error when creating file.");
                    listener.jobFailed(this, JobEndStatus.EXECUTION_FAILED);
                } else {
                    if (!debug) {
                        localFile.delete();
                    }
                    RUNNING_JOBS.remove(this);
                    listener.jobCompleted(this);
                }
            } else {
                if (job.getExitStatus() == 0) {
                    RUNNING_JOBS.remove(this);
                    listener.jobCompleted(this);
                } else {
                    GATjob = null;
                    RUNNING_JOBS.remove(this);
                    listener.jobFailed(this, JobEndStatus.EXECUTION_FAILED);
                }
            }
        } catch (Exception e) {
            ErrorManager.fatal(CALLBACK_PROCESSING_ERR + ": " + this, e);
        }
    } else if (newJobState == JobState.SUBMISSION_ERROR) {
        if (Tracer.isActivated()) {
            Integer slot = (Integer) sd.getAttributes().get("slot");
            String host = getResourceNode().getHost();
            Tracer.freeSlot(host, slot);
        }
        try {
            if (usingGlobus && job.getInfo().get("resManError").equals("NO_ERROR")) {
                RUNNING_JOBS.remove(this);
                listener.jobCompleted(this);
            } else {
                GATjob = null;
                RUNNING_JOBS.remove(this);
                listener.jobFailed(this, JobEndStatus.SUBMISSION_FAILED);
            }
        } catch (GATInvocationException e) {
            ErrorManager.fatal(CALLBACK_PROCESSING_ERR + ": " + this, e);
        }
    }
}

Example 2 with Job

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

the class GATJob method submit.

@Override
public void submit() throws Exception {
    // Prepare the job
    logger.info("Submit GATJob with ID " + jobId);
    JobDescription jobDescr = null;
    jobDescr = prepareJob();
    // Get a broker for the host
    ResourceBroker broker = null;
    String dest = (String) jobDescr.getResourceDescription().getResourceAttribute(RES_ATTR);
    if ((broker = brokers.get(dest)) == null) {
        broker = GAT.createResourceBroker(context, new URI(dest));
        brokers.put(dest, broker);
    }
    // Submit the job, registering for notifications of job state
    // transitions (associatedJM is the metric listener)
    Job job = null;
    try {
        job = broker.submitJob(jobDescr, this, JOB_STATUS);
        RUNNING_JOBS.add(this);
    } catch (Exception e) {
        if (Tracer.isActivated()) {
            Tracer.freeSlot(((GATWorkerNode) worker.getNode()).getHost(), (Integer) jobDescr.getSoftwareDescription().getAttributes().get("slot"));
        }
        throw e;
    }
    // Update mapping
    GATjob = job;
}

Example 3 with Job

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

the class GATScriptExecutor method executeScript.

public boolean executeScript(List<URI> scripts, List<String> params, String stdOutFileName) {
    try {
        pool.startThreads();
    } catch (Exception e) {
        logger.error(THREAD_POOL_START_ERR, e);
        return false;
    }
    synchronized (jobQueue) {
        jobCount = scripts.size();
    }
    for (int i = 0; i < scripts.size(); i++) {
        URI script = scripts.get(i);
        String cleanParam = params.get(i);
        if (script == null) {
            continue;
        }
        if (debug) {
            logger.debug("Clean call: " + script + " " + cleanParam);
        }
        try {
            if (!node.isUserNeeded() && script.getUserInfo() != null) {
                // Remove user from the URI
                script.setUserInfo(null);
            }
            String user = script.getUserInfo();
            if (user == null) {
                user = "";
            } else {
                user += "@";
            }
            SoftwareDescription sd = new SoftwareDescription();
            sd.addAttribute("uri", Protocol.ANY_URI.getSchema() + user + script.getHost());
            sd.setExecutable(script.getPath());
            sd.setArguments(cleanParam.split(" "));
            sd.addAttribute("job_number", i);
            sd.addAttribute(SoftwareDescription.SANDBOX_ROOT, File.separator + "tmp" + File.separator);
            sd.addAttribute(SoftwareDescription.SANDBOX_USEROOT, "true");
            sd.addAttribute(SoftwareDescription.SANDBOX_DELETE, "false");
            if (debug) {
                try {
                    org.gridlab.gat.io.File outFile = GAT.createFile(node.getContext(), Protocol.ANY_URI.getSchema() + File.separator + System.getProperty(COMPSsConstants.APP_LOG_DIR) + File.separator + stdOutFileName + ".out");
                    sd.setStdout(outFile);
                    org.gridlab.gat.io.File errFile = GAT.createFile(node.getContext(), Protocol.ANY_URI.getSchema() + File.separator + System.getProperty(COMPSsConstants.APP_LOG_DIR) + File.separator + stdOutFileName + ".err");
                    sd.setStderr(errFile);
                } catch (Exception e) {
                    logger.error(CLEAN_JOB_ERR, e);
                }
            }
            sdQueue.enqueue(sd);
        } catch (Exception e) {
            logger.error(CLEAN_JOB_ERR, e);
            return false;
        }
    }
    Long timeout = System.currentTimeMillis() + 60_000l;
    // Poll for completion of the clean jobs
    while (jobCount > 0 && System.currentTimeMillis() < timeout) {
        Job job = jobQueue.dequeue();
        if (job == null) {
            synchronized (jobQueue) {
                jobCount--;
            }
        } else if (job.getState() == JobState.STOPPED) {
            synchronized (jobQueue) {
                jobCount--;
            }
        } else if (job.getState() == JobState.SUBMISSION_ERROR) {
            logger.error(CLEAN_JOB_ERR + ": " + job);
            synchronized (jobQueue) {
                jobCount--;
            }
        } else {
            jobQueue.enqueue(job);
            try {
                Thread.sleep(50);
            } catch (Exception e) {
            }
        }
    }
    try {
        pool.stopThreads();
    } catch (Exception e) {
        logger.error(THREAD_POOL_STOP_ERR, e);
        return false;
    }
    // Move cleanX.out logs to default logger
    if (debug) {
        String stdOutFilePath = System.getProperty(COMPSsConstants.APP_LOG_DIR) + File.separator + stdOutFileName + ".out";
        try (FileReader cleanOut = new FileReader(stdOutFilePath);
            BufferedReader br = new BufferedReader(cleanOut)) {
            String line = br.readLine();
            while (line != null) {
                logger.debug(line);
                line = br.readLine();
            }
        } catch (Exception e) {
            logger.error("Error moving std out file", e);
        }
        // Delete file
        if (!new File(stdOutFilePath).delete()) {
            logger.error("Error deleting out file " + stdOutFilePath);
        }
    }
    // Move cleanX.err logs to default logger
    if (debug) {
        String stdErrFilePath = System.getProperty(COMPSsConstants.APP_LOG_DIR) + File.separator + stdOutFileName + ".err";
        try (FileReader cleanErr = new FileReader(stdErrFilePath);
            BufferedReader br = new BufferedReader(cleanErr)) {
            String line = br.readLine();
            while (line != null) {
                logger.error(line);
                line = br.readLine();
            }
        } catch (Exception e) {
            logger.error("Error moving std err file", e);
        }
        if (!new File(stdErrFilePath).delete()) {
            logger.error("Error deleting err file " + stdErrFilePath);
        }
    }
    return true;
}

Example 4 with Job

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

the class LsfResourceBrokerAdaptor method submitJob.

/*
     * (non-Javadoc)
     * 
     * @see org.gridlab.gat.resources.ResourceBroker#submitJob(org.gridlab.gat.resources.JobDescription)
     */
public Job submitJob(AbstractJobDescription abstractDescription, MetricListener listener, String metricDefinitionName) throws GATInvocationException {
    if (!(abstractDescription instanceof JobDescription)) {
        throw new GATInvocationException("can only handle JobDescriptions: " + abstractDescription.getClass());
    }
    JobDescription description = (JobDescription) abstractDescription;
    SoftwareDescription sd = description.getSoftwareDescription();
    if (sd == null) {
        throw new GATInvocationException("The job description does not contain a software description");
    }
    if (description.getProcessCount() < 1) {
        throw new GATInvocationException("Adaptor cannot handle: process count < 1: " + description.getProcessCount());
    }
    if (description.getResourceCount() != 1) {
        throw new GATInvocationException("Adaptor cannot handle: resource count > 1: " + description.getResourceCount());
    }
    String home = System.getProperty("user.home");
    if (home == null) {
        throw new GATInvocationException("lsf broker could not get user home dir");
    }
    Sandbox sandbox = new Sandbox(gatContext, description, "localhost", home, true, true, false, false);
    LsfJob lsfJob = new LsfJob(gatContext, description, sandbox);
    Job job = null;
    if (description instanceof WrapperJobDescription) {
        WrapperJobCpi tmp = new WrapperJobCpi(gatContext, lsfJob, listener, metricDefinitionName);
        listener = tmp;
        job = tmp;
    } else {
        job = lsfJob;
    }
    if (listener != null && metricDefinitionName != null) {
        Metric metric = lsfJob.getMetricDefinitionByName(metricDefinitionName).createMetric(null);
        lsfJob.addMetricListener(listener, metric);
    }
    lsfJob.setState(Job.JobState.PRE_STAGING);
    lsfJob.waitForTrigger(Job.JobState.PRE_STAGING);
    sandbox.prestage();
    String exe;
    if (sandbox.getResolvedExecutable() != null) {
        exe = sandbox.getResolvedExecutable().getPath();
    // try to set the executable bit, it might be lost
    /* CDIAZ: The command "exe" can be also in a remote host
             * 		  The command must have the right permissions in the remote host
            try {
                new CommandRunner("chmod", "+x", exe);
            } catch (Throwable t) {
                // ignore
            }
            */
    } else {
        exe = getExecutable(description);
    }
    String[] args = getArgumentsArray(description);
    // Directory where the lsf command will be executed
    java.io.File f = new java.io.File(sandbox.getSandboxPath());
    if (!f.exists()) {
        throw new GATInvocationException("Unable to find directory " + f.getAbsolutePath());
    }
    // Check and set the environment for a blaunch command
    Map<String, Object> env = sd.getEnvironment();
    this.prepareBLaunchEnv(env);
    // Encapsulate the original command into a blaunch command
    String host = brokerURI.getHost();
    String blExe = this.getBlaunchCommand();
    String[] blArgs = this.getBlaunchArgs(host, exe, args);
    ProcessBundle bundle = new ProcessBundle(description.getProcessCount(), blExe, blArgs, f, env);
    lsfJob.setSubmissionTime();
    lsfJob.setState(Job.JobState.SCHEDULED);
    try {
        lsfJob.setState(Job.JobState.RUNNING);
        lsfJob.waitForTrigger(Job.JobState.RUNNING);
        lsfJob.setStartTime();
        bundle.startBundle();
        lsfJob.setProcess(bundle);
    } catch (IOException e) {
        throw new CommandNotFoundException("LsfResourceBrokerAdaptor", e);
    }
    if (!sd.streamingStderrEnabled()) {
        try {
            if (sd.getStderr() != null) {
                OutputStream err = GAT.createFileOutputStream(gatContext, sd.getStderr());
                // to file
                StreamForwarder forwarder = new StreamForwarder(bundle.getStderr(), err, sd.getExecutable() + " [stderr]");
                lsfJob.setErrorStream(forwarder);
                if (logger.isDebugEnabled()) {
                    logger.debug("Created stderr forwarder to file " + sd.getStderr());
                }
            } else {
                // or throw it away
                new StreamForwarder(bundle.getStderr(), null, sd.getExecutable() + " [stderr]");
            }
        } catch (GATObjectCreationException e) {
            throw new GATInvocationException("Unable to create file output stream for stderr!", e);
        }
    }
    if (!sd.streamingStdoutEnabled()) {
        // read away the stdout
        try {
            if (sd.getStdout() != null) {
                // to file
                OutputStream out = GAT.createFileOutputStream(gatContext, sd.getStdout());
                StreamForwarder forwarder = new StreamForwarder(bundle.getStdout(), out, sd.getExecutable() + " [stdout]");
                lsfJob.setOutputStream(forwarder);
                if (logger.isDebugEnabled()) {
                    logger.debug("Created stdout forwarder to file " + sd.getStdout());
                }
            } else {
                // or throw it away
                new StreamForwarder(bundle.getStdout(), null, sd.getExecutable() + " [stdout]");
            }
        } catch (GATObjectCreationException e) {
            throw new GATInvocationException("Unable to create file output stream for stdout!", e);
        }
    }
    if (!sd.streamingStdinEnabled() && sd.getStdin() != null) {
        // forward the stdin from file
        try {
            InputStream in = GAT.createFileInputStream(gatContext, sd.getStdin());
            bundle.setStdin(sd.getExecutable(), in);
        } catch (GATObjectCreationException e) {
            throw new GATInvocationException("Unable to create file input stream for stdin!", e);
        }
    }
    lsfJob.monitorState();
    return job;
}

Example 5 with Job

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

the class SshLSFJob method getJobState.

protected void getJobState(String jobID) throws GATInvocationException {
    synchronized (this) {
        while (jobStateBusy) {
            try {
                wait();
            } catch (InterruptedException e) {
            // ignored
            }
        }
        jobStateBusy = true;
    }
    JobState resultState;
    try {
        if (state == JobState.POST_STAGING || state == JobState.STOPPED || state == JobState.SUBMISSION_ERROR) {
            return;
        }
        logger.debug("Getting task status in setState()");
        // getting the status via ssh ... squeue
        java.io.File squeueResultFile = null;
        try {
            // Create qstat job
            SoftwareDescription sd = new SoftwareDescription();
            // Use /bin/sh, so that $USER gets expanded.
            sd.setExecutable("/bin/sh");
            sd.setArguments("-c", "bjobs -noheader " + jobID + " | awk {' print $3 '}");
            sd.addAttribute(SoftwareDescription.SANDBOX_USEROOT, "true");
            squeueResultFile = java.io.File.createTempFile("GAT", "tmp");
            try {
                sd.setStdout(GAT.createFile(subContext, new URI("file:///" + squeueResultFile.getAbsolutePath().replace(File.separatorChar, '/'))));
            } catch (Throwable e1) {
                throw new GATInvocationException("Could not create GAT object for temporary " + squeueResultFile.getAbsolutePath(), e1);
            }
            JobDescription jd = new JobDescription(sd);
            Job job = jobHelper.submitJob(jd, this, "job.status");
            synchronized (job) {
                while (job.getState() != Job.JobState.STOPPED && job.getState() != Job.JobState.SUBMISSION_ERROR) {
                    try {
                        job.wait();
                    } catch (InterruptedException e) {
                    // ignore
                    }
                }
            }
            if (job.getState() != Job.JobState.STOPPED || job.getExitStatus() != 0) {
                throw new GATInvocationException("Could not submit squeue job " + sd.toString());
            }
            // submit success.
            BufferedReader in = new BufferedReader(new FileReader(squeueResultFile.getAbsolutePath()));
            String status = in.readLine();
            // or finished. Set to "" in this case. --Ceriel
            if (status == null) {
                status = "";
            }
            if (logger.isDebugEnabled()) {
                logger.debug("squeue line: " + status);
            }
            resultState = mapLSFStatetoGAT(status);
        } catch (IOException e) {
            logger.debug("retrieving job status sshslurmjob failed", e);
            throw new GATInvocationException("Unable to retrieve the Job Status", e);
        } finally {
            squeueResultFile.delete();
        }
    } finally {
        synchronized (this) {
            jobStateBusy = false;
            notifyAll();
        }
    }
    if (resultState != JobState.STOPPED) {
        setState(resultState);
    } else {
        setState(JobState.POST_STAGING);
    }
}