Examples with ComputeResourceDescription org.apache.airavata.model.appcatalog.computeresource.ComputeResourceDescription used on opensource projects

Example 16 with ComputeResourceDescription

use of org.apache.airavata.model.appcatalog.computeresource.ComputeResourceDescription in project airavata by apache.

the class BatchQueueValidator method validateUserConfiguration.

private List<ValidatorResult> validateUserConfiguration(ExperimentModel experiment, ProcessModel processModel) throws AppCatalogException {
    List<ValidatorResult> validatorResultList = new ArrayList<ValidatorResult>();
    try {
        UserConfigurationDataModel userConfigurationData = experiment.getUserConfigurationData();
        ComputationalResourceSchedulingModel computationalResourceScheduling = userConfigurationData.getComputationalResourceScheduling();
        if (userConfigurationData.isAiravataAutoSchedule()) {
            logger.info("User enabled Auto-Schedule. Hence we don't do validation..");
            ValidatorResult validatorResult = new ValidatorResult();
            validatorResult.setResult(true);
            validatorResultList.add(validatorResult);
        } else {
            ComputeResourceDescription computeResource;
            if (processModel == null) {
                computeResource = appCatalog.getComputeResource().getComputeResource(experiment.getUserConfigurationData().getComputationalResourceScheduling().getResourceHostId());
            } else {
                computeResource = appCatalog.getComputeResource().getComputeResource(processModel.getProcessResourceSchedule().getResourceHostId());
            }
            List<BatchQueue> batchQueues = computeResource.getBatchQueues();
            if (batchQueues != null && !batchQueues.isEmpty()) {
                if (computationalResourceScheduling != null) {
                    String experimentQueueName = computationalResourceScheduling.getQueueName().trim();
                    int experimentWallTimeLimit = computationalResourceScheduling.getWallTimeLimit();
                    int experimentNodeCount = computationalResourceScheduling.getNodeCount();
                    int experimentCPUCount = computationalResourceScheduling.getTotalCPUCount();
                    ValidatorResult queueNameResult = new ValidatorResult();
                    // Set the validation to false. Once all the queue's are looped, if nothing matches, then this gets passed.
                    queueNameResult.setResult(false);
                    queueNameResult.setErrorDetails("The specified queue " + experimentQueueName + " does not exist. If you believe this is an error, contact the administrator to verify App-Catalog Configurations");
                    for (BatchQueue queue : batchQueues) {
                        String resourceQueueName = queue.getQueueName();
                        int maxQueueRunTime = queue.getMaxRunTime();
                        int maxNodeCount = queue.getMaxNodes();
                        int maxcpuCount = queue.getMaxProcessors();
                        if (resourceQueueName != null && resourceQueueName.equals(experimentQueueName)) {
                            queueNameResult.setResult(true);
                            queueNameResult.setErrorDetails("");
                            // Validate if the specified wall time is within allowable limit
                            ValidatorResult wallTimeResult = new ValidatorResult();
                            if (experimentWallTimeLimit == 0) {
                                wallTimeResult.setResult(false);
                                wallTimeResult.setErrorDetails("Walltime cannot be zero for queue " + resourceQueueName);
                            } else {
                                if (maxQueueRunTime == 0) {
                                    wallTimeResult.setResult(true);
                                    wallTimeResult.setErrorDetails("Maximum wall time is not configured for the queue," + "Validation is being skipped");
                                    logger.info("Maximum wall time is not configured for the queue" + "Validation is being skipped");
                                } else {
                                    if (maxQueueRunTime < experimentWallTimeLimit) {
                                        wallTimeResult.setResult(false);
                                        wallTimeResult.setErrorDetails("Job Execution walltime " + experimentWallTimeLimit + "exceeds the allowable walltime" + maxQueueRunTime + "for queue " + resourceQueueName);
                                    } else {
                                        wallTimeResult.setResult(true);
                                        wallTimeResult.setErrorDetails("");
                                    }
                                }
                            }
                            // validate max node count
                            ValidatorResult nodeCountResult = new ValidatorResult();
                            if (maxNodeCount == 0) {
                                nodeCountResult.setResult(true);
                                nodeCountResult.setErrorDetails("Max node count is not configured for the queue," + "Validation is being skipped");
                                logger.info("Max node count is not configured for the queue" + "Validation is being skipped");
                            } else {
                                if (experimentNodeCount == 0) {
                                    nodeCountResult.setResult(false);
                                    nodeCountResult.setErrorDetails("Job Execution node count cannot be zero for queue " + resourceQueueName);
                                } else {
                                    if (maxNodeCount < experimentNodeCount) {
                                        nodeCountResult.setResult(false);
                                        nodeCountResult.setErrorDetails("Job Execution node count " + experimentNodeCount + "exceeds the allowable node count" + maxNodeCount + "for queue " + resourceQueueName);
                                    } else {
                                        nodeCountResult.setResult(true);
                                        nodeCountResult.setErrorDetails("");
                                    }
                                }
                            }
                            // validate cpu count
                            ValidatorResult cpuCountResult = new ValidatorResult();
                            if (maxcpuCount == 0) {
                                cpuCountResult.setResult(true);
                                cpuCountResult.setErrorDetails("Max cpu count is not configured for the queue," + "Validation is being skipped");
                                logger.info("Max cpu count is not configured for the queue" + "Validation is being skipped");
                            } else {
                                if (experimentCPUCount == 0) {
                                    cpuCountResult.setResult(false);
                                    cpuCountResult.setErrorDetails("Job Execution cpu count cannot be zero for queue " + resourceQueueName);
                                } else {
                                    if (maxcpuCount < experimentCPUCount) {
                                        cpuCountResult.setResult(false);
                                        cpuCountResult.setErrorDetails("Job Execution cpu count " + experimentCPUCount + "exceeds the allowable cpu count" + maxcpuCount + "for queue " + resourceQueueName);
                                    } else {
                                        cpuCountResult.setResult(true);
                                        cpuCountResult.setErrorDetails("");
                                    }
                                }
                            }
                            validatorResultList.add(wallTimeResult);
                            validatorResultList.add(nodeCountResult);
                            validatorResultList.add(cpuCountResult);
                        }
                    }
                    validatorResultList.add(queueNameResult);
                }
            } else {
                // for some compute resources, you dnt need to specify queue names
                ValidatorResult result = new ValidatorResult();
                logger.info("There are not queues defined under the compute resource. Airavata assumes this experiment " + "does not need a queue name...");
                result.setResult(true);
                validatorResultList.add(result);
            }
        }
    } catch (AppCatalogException e) {
        logger.error("Error while getting information from App catalog", e);
        throw new AppCatalogException("Error while getting information from App catalog", e);
    }
    return validatorResultList;
}

Example 17 with ComputeResourceDescription

use of org.apache.airavata.model.appcatalog.computeresource.ComputeResourceDescription in project airavata by apache.

the class OrchestratorServerHandler method getAppDeploymentForModule.

private ApplicationDeploymentDescription getAppDeploymentForModule(ProcessModel processModel, String selectedModuleId) throws AppCatalogException, ClassNotFoundException, ApplicationSettingsException, InstantiationException, IllegalAccessException {
    Map<String, String> moduleIdFilter = new HashMap<String, String>();
    moduleIdFilter.put(AppCatAbstractResource.ApplicationDeploymentConstants.APP_MODULE_ID, selectedModuleId);
    if (processModel.getProcessResourceSchedule() != null && processModel.getProcessResourceSchedule().getResourceHostId() != null) {
        moduleIdFilter.put(AppCatAbstractResource.ApplicationDeploymentConstants.COMPUTE_HOST_ID, processModel.getProcessResourceSchedule().getResourceHostId());
    }
    List<ApplicationDeploymentDescription> applicationDeployements = appCatalog.getApplicationDeployment().getApplicationDeployements(moduleIdFilter);
    Map<ComputeResourceDescription, ApplicationDeploymentDescription> deploymentMap = new HashMap<ComputeResourceDescription, ApplicationDeploymentDescription>();
    ComputeResource computeResource = appCatalog.getComputeResource();
    for (ApplicationDeploymentDescription deploymentDescription : applicationDeployements) {
        deploymentMap.put(computeResource.getComputeResource(deploymentDescription.getComputeHostId()), deploymentDescription);
    }
    List<ComputeResourceDescription> computeHostList = Arrays.asList(deploymentMap.keySet().toArray(new ComputeResourceDescription[] {}));
    Class<? extends HostScheduler> aClass = Class.forName(ServerSettings.getHostScheduler()).asSubclass(HostScheduler.class);
    HostScheduler hostScheduler = aClass.newInstance();
    ComputeResourceDescription ComputeResourceDescription = hostScheduler.schedule(computeHostList);
    return deploymentMap.get(ComputeResourceDescription);
}

Example 18 with ComputeResourceDescription

use of org.apache.airavata.model.appcatalog.computeresource.ComputeResourceDescription in project airavata by apache.

the class ClusterStatusMonitorJob method execute.

@Override
public void execute(JobExecutionContext jobExecutionContext) throws JobExecutionException {
    try {
        String superTenantGatewayId = ServerSettings.getSuperTenantGatewayId();
        RegistryService.Client registryClient = getRegistryClient();
        List<ComputeResourceProfile> computeResourceProfiles = new ArrayList<>();
        List<ComputeResourcePreference> computeResourcePreferences = null;
        try {
            computeResourcePreferences = registryClient.getAllGatewayComputeResourcePreferences(superTenantGatewayId);
        } catch (Exception ex) {
            logger.warn("Could not find super tenant compute resources preferences for cluster status monitoring...");
        }
        if (computeResourcePreferences != null && computeResourcePreferences.size() > 0) {
            computeResourcePreferences.stream().forEach(p -> {
                try {
                    String computeResourceId = p.getComputeResourceId();
                    String credentialStoreToken = p.getResourceSpecificCredentialStoreToken();
                    String loginUserName = p.getLoginUserName();
                    String hostName = null;
                    if (credentialStoreToken == null || credentialStoreToken.equals("")) {
                        credentialStoreToken = registryClient.getGatewayResourceProfile(superTenantGatewayId).getCredentialStoreToken();
                    }
                    int port = -1;
                    ArrayList queueNames = new ArrayList<>();
                    ComputeResourceDescription computeResourceDescription = registryClient.getComputeResource(computeResourceId);
                    hostName = computeResourceDescription.getHostName();
                    // FIXME This should come from compute resource description
                    port = 22;
                    computeResourceDescription.getBatchQueues().stream().forEach(q -> {
                        queueNames.add(q.getQueueName());
                    });
                    List<JobSubmissionInterface> jobSubmissionInterfaces = computeResourceDescription.getJobSubmissionInterfaces();
                    if (jobSubmissionInterfaces != null && jobSubmissionInterfaces.size() > 0) {
                        if (jobSubmissionInterfaces.get(0).getJobSubmissionProtocol().equals(JobSubmissionProtocol.SSH)) {
                            String resourceManagerType = registryClient.getSSHJobSubmission(jobSubmissionInterfaces.get(0).getJobSubmissionInterfaceId()).getResourceJobManager().getResourceJobManagerType().name();
                            ComputeResourceProfile computeResourceProfile = new ComputeResourceProfile(hostName, loginUserName, port, credentialStoreToken, queueNames, resourceManagerType);
                            computeResourceProfiles.add(computeResourceProfile);
                        }
                    }
                } catch (TException e) {
                    logger.error(e.getMessage());
                }
            });
        }
        ArrayList<QueueStatusModel> queueStatuses = new ArrayList<>();
        for (ComputeResourceProfile computeResourceProfile : computeResourceProfiles) {
            String userName = computeResourceProfile.getUserName();
            String hostName = computeResourceProfile.getHostName();
            int port = computeResourceProfile.getPort();
            try {
                JSch jsch = new JSch();
                CredentialStoreService.Client credentialClient = getCredentialStoreClient();
                SSHCredential sshCredential = credentialClient.getSSHCredential(computeResourceProfile.getCredentialStoreToken(), superTenantGatewayId);
                jsch.addIdentity(hostName, sshCredential.getPrivateKey().getBytes(), sshCredential.getPublicKey().getBytes(), sshCredential.getPassphrase().getBytes());
                Session session = jsch.getSession(userName, hostName, port);
                java.util.Properties config = new java.util.Properties();
                config.put("StrictHostKeyChecking", "no");
                session.setConfig(config);
                logger.debug("Connected to " + hostName);
                session.connect();
                for (String queue : computeResourceProfile.getQueueNames()) {
                    String command = "";
                    if (computeResourceProfile.getResourceManagerType().equals("SLURM"))
                        command = "sinfo -s -p " + queue + " -o \"%a %F\" | tail -1";
                    else if (computeResourceProfile.getResourceManagerType().equals("PBS"))
                        command = "qstat -Q " + queue + "| tail -1";
                    if (command.equals("")) {
                        logger.warn("No matching resource manager type found for " + computeResourceProfile.getResourceManagerType());
                        continue;
                    }
                    Channel channel = session.openChannel("exec");
                    ((ChannelExec) channel).setCommand(command);
                    channel.setInputStream(null);
                    ((ChannelExec) channel).setErrStream(System.err);
                    InputStream in = channel.getInputStream();
                    channel.connect();
                    byte[] tmp = new byte[1024];
                    String result = "";
                    while (true) {
                        while (in.available() > 0) {
                            int i = in.read(tmp, 0, 1024);
                            if (i < 0)
                                break;
                            result += new String(tmp, 0, i);
                        }
                        if (channel.isClosed()) {
                            if (in.available() > 0)
                                continue;
                            logger.debug(hostName + " " + queue + " " + "exit-status: " + channel.getExitStatus());
                            break;
                        }
                        try {
                            Thread.sleep(1000);
                        } catch (Exception ee) {
                        }
                    }
                    channel.disconnect();
                    if (result != null && result.length() > 0) {
                        QueueStatusModel queueStatus = null;
                        if (computeResourceProfile.getResourceManagerType().equals("SLURM")) {
                            String[] sparts = result.split(" ");
                            boolean isUp = sparts[0].equalsIgnoreCase("up");
                            String knts = sparts[1];
                            sparts = knts.split("/");
                            int running = Integer.parseInt(sparts[0].trim());
                            int queued = Integer.parseInt(sparts[1].trim());
                            queueStatus = new QueueStatusModel(hostName, queue, isUp, running, queued, System.currentTimeMillis());
                        } else if (computeResourceProfile.getResourceManagerType().equals("PBS")) {
                            result = result.replaceAll("\\s+", " ");
                            String[] sparts = result.split(" ");
                            boolean isUp = sparts[3].equalsIgnoreCase("yes");
                            int running = Integer.parseInt(sparts[6].trim());
                            int queued = Integer.parseInt(sparts[5].trim());
                            queueStatus = new QueueStatusModel(hostName, queue, isUp, running, queued, System.currentTimeMillis());
                        }
                        if (queueStatus != null)
                            queueStatuses.add(queueStatus);
                    }
                }
                session.disconnect();
            } catch (Exception ex) {
                logger.error("Failed to get cluster status from " + computeResourceProfile.getHostName());
                logger.error(ex.getMessage(), ex);
            }
        }
        if (queueStatuses != null && queueStatuses.size() > 0) {
            registryClient.registerQueueStatuses(queueStatuses);
        }
    } catch (Exception e) {
        throw new JobExecutionException(e);
    }
}

Example 19 with ComputeResourceDescription

use of org.apache.airavata.model.appcatalog.computeresource.ComputeResourceDescription in project airavata by apache.

the class AppDeploymentTest method testAppDeployment.

@Test
public void testAppDeployment() throws Exception {
    ApplicationDeployment appDep = appcatalog.getApplicationDeployment();
    ApplicationInterface appInt = appcatalog.getApplicationInterface();
    ComputeResource computeRs = appcatalog.getComputeResource();
    ComputeResourceDescription cm = new ComputeResourceDescription();
    cm.setHostName("localhost");
    cm.setResourceDescription("test compute host");
    String hostId = computeRs.addComputeResource(cm);
    ApplicationModule module = new ApplicationModule();
    module.setAppModuleName("WRF");
    module.setAppModuleVersion("1.0.0");
    String wrfModuleId = appInt.addApplicationModule(module, ServerSettings.getDefaultUserGateway());
    ApplicationDeploymentDescription description = new ApplicationDeploymentDescription();
    description.setAppModuleId(wrfModuleId);
    description.setComputeHostId(hostId);
    description.setExecutablePath("/home/a/b/c");
    description.setAppDeploymentDescription("test app deployment");
    CommandObject cmd1 = new CommandObject();
    cmd1.setCommand("cmd1");
    cmd1.setCommandOrder(1);
    CommandObject cmd2 = new CommandObject();
    cmd2.setCommand("cmd1");
    cmd2.setCommandOrder(1);
    description.addToModuleLoadCmds(cmd1);
    description.addToModuleLoadCmds(cmd2);
    List<SetEnvPaths> libPrepandPaths = new ArrayList<SetEnvPaths>();
    libPrepandPaths.add(createSetEnvPath("name1", "val1", 1));
    libPrepandPaths.add(createSetEnvPath("name2", "val2", 2));
    description.setLibPrependPaths(libPrepandPaths);
    List<SetEnvPaths> libApendPaths = new ArrayList<SetEnvPaths>();
    libApendPaths.add(createSetEnvPath("name3", "val3", 1));
    libApendPaths.add(createSetEnvPath("name4", "val4", 2));
    description.setLibAppendPaths(libApendPaths);
    List<SetEnvPaths> appEvns = new ArrayList<SetEnvPaths>();
    appEvns.add(createSetEnvPath("name5", "val5", 1));
    appEvns.add(createSetEnvPath("name6", "val6", 2));
    description.setSetEnvironment(appEvns);
    String appDepId = appDep.addApplicationDeployment(description, ServerSettings.getDefaultUserGateway());
    ApplicationDeploymentDescription app = null;
    if (appDep.isAppDeploymentExists(appDepId)) {
        app = appDep.getApplicationDeployement(appDepId);
        System.out.println("*********** application deployment id ********* : " + app.getAppDeploymentId());
        System.out.println("*********** application deployment desc ********* : " + app.getAppDeploymentDescription());
    }
    description.setAppDeploymentDescription("test app deployment2");
    appDep.updateApplicationDeployment(appDepId, description);
    if (appDep.isAppDeploymentExists(appDepId)) {
        app = appDep.getApplicationDeployement(appDepId);
        System.out.println("*********** application deployment desc ********* : " + app.getAppDeploymentDescription());
    }
    Map<String, String> moduleIdFilter = new HashMap<String, String>();
    moduleIdFilter.put(AppCatAbstractResource.ApplicationDeploymentConstants.APP_MODULE_ID, wrfModuleId);
    List<ApplicationDeploymentDescription> applicationDeployements = appDep.getApplicationDeployements(moduleIdFilter);
    System.out.println("******** Size of App deployments for module *********** : " + applicationDeployements.size());
    Map<String, String> hostFilter = new HashMap<String, String>();
    hostFilter.put(AppCatAbstractResource.ApplicationDeploymentConstants.COMPUTE_HOST_ID, hostId);
    List<ApplicationDeploymentDescription> applicationDeployementsForHost = appDep.getApplicationDeployements(hostFilter);
    System.out.println("******** Size of App deployments for host *********** : " + applicationDeployementsForHost.size());
    List<String> allApplicationDeployementIds = appDep.getAllApplicationDeployementIds();
    System.out.println("******** Size of all App deployments ids *********** : " + allApplicationDeployementIds.size());
    List<ApplicationDeploymentDescription> allApplicationDeployements = appDep.getAllApplicationDeployements(ServerSettings.getDefaultUserGateway());
    System.out.println("******** Size of all App deployments *********** : " + allApplicationDeployements.size());
    assertTrue("App interface saved successfully", app != null);
}

Example 20 with ComputeResourceDescription

use of org.apache.airavata.model.appcatalog.computeresource.ComputeResourceDescription in project airavata by apache.

the class GatewayProfileTest method gatewayProfileTest.

@Test
public void gatewayProfileTest() throws Exception {
    GwyResourceProfile gatewayProfile = appcatalog.getGatewayProfile();
    GatewayResourceProfile gf = new GatewayResourceProfile();
    ComputeResource computeRs = appcatalog.getComputeResource();
    ComputeResourceDescription cm1 = new ComputeResourceDescription();
    cm1.setHostName("localhost");
    cm1.setResourceDescription("test compute host");
    String hostId1 = computeRs.addComputeResource(cm1);
    ComputeResourceDescription cm2 = new ComputeResourceDescription();
    cm2.setHostName("localhost");
    cm2.setResourceDescription("test compute host");
    String hostId2 = computeRs.addComputeResource(cm2);
    ComputeResourcePreference preference1 = new ComputeResourcePreference();
    preference1.setComputeResourceId(hostId1);
    preference1.setOverridebyAiravata(true);
    preference1.setPreferredJobSubmissionProtocol(JobSubmissionProtocol.SSH);
    preference1.setPreferredDataMovementProtocol(DataMovementProtocol.SCP);
    preference1.setPreferredBatchQueue("queue1");
    preference1.setScratchLocation("/tmp");
    preference1.setAllocationProjectNumber("project1");
    ComputeResourcePreference preference2 = new ComputeResourcePreference();
    preference2.setComputeResourceId(hostId2);
    preference2.setOverridebyAiravata(true);
    preference2.setPreferredJobSubmissionProtocol(JobSubmissionProtocol.LOCAL);
    preference2.setPreferredDataMovementProtocol(DataMovementProtocol.GridFTP);
    preference2.setPreferredBatchQueue("queue2");
    preference2.setScratchLocation("/tmp");
    preference2.setAllocationProjectNumber("project2");
    List<ComputeResourcePreference> list = new ArrayList<ComputeResourcePreference>();
    list.add(preference1);
    list.add(preference2);
    gf.setComputeResourcePreferences(list);
    gf.setGatewayID("testGateway");
    String gwId = gatewayProfile.addGatewayResourceProfile(gf);
    GatewayResourceProfile retrievedProfile = null;
    if (gatewayProfile.isGatewayResourceProfileExists(gwId)) {
        retrievedProfile = gatewayProfile.getGatewayProfile(gwId);
        System.out.println("************ gateway id ************** :" + retrievedProfile.getGatewayID());
    }
    List<ComputeResourcePreference> preferences = gatewayProfile.getAllComputeResourcePreferences(gwId);
    System.out.println("compute preferences size : " + preferences.size());
    if (preferences != null && !preferences.isEmpty()) {
        for (ComputeResourcePreference cm : preferences) {
            System.out.println("******** host id ********* : " + cm.getComputeResourceId());
            System.out.println(cm.getPreferredBatchQueue());
            System.out.println(cm.getPreferredDataMovementProtocol());
            System.out.println(cm.getPreferredJobSubmissionProtocol());
        }
    }
    assertTrue("App interface saved successfully", retrievedProfile != null);
}