Examples with Twister2Exception edu.iu.dsc.tws.api.exceptions.Twister2Exception used on opensource projects

Example 26 with Twister2Exception

use of edu.iu.dsc.tws.api.exceptions.Twister2Exception in project twister2 by DSC-SPIDAL.

the class JobZNodeManager method checkJstZNodeWaitIfNeeded.

/**
 * Job master creates job submission time znode under job znode,
 * as the last action to create job related znodes at ZK server
 * workers wait for the job master to create this znode.
 * They proceed after seeing that this jst znode is created
 * <p>
 * this jst znode may exist from previous runs in the case of restarting from a checkpoint
 * because of this, its value has to be compared
 */
public static boolean checkJstZNodeWaitIfNeeded(CuratorFramework client, String rootPath, String jobID, long jsTime) throws Twister2Exception {
    String jstPath = ZKUtils.jobSubmisionTimePath(rootPath, jobID);
    // 100 seconds
    long timeLimit = 100000;
    long sleepInterval = 300;
    long duration = 0;
    long startTime = System.currentTimeMillis();
    // log interval in milliseconds
    long logInterval = 3000;
    long nextLogTime = logInterval;
    int checkCount = 1;
    while (duration < timeLimit) {
        try {
            if (client.checkExists().forPath(jstPath) != null) {
                byte[] jstBytes = client.getData().forPath(jstPath);
                long jstAtZK = Longs.fromByteArray(jstBytes);
                if (jstAtZK == jsTime) {
                    LOG.info("matched job submission times. Proceeding. checkCount: " + checkCount);
                    return true;
                }
            }
        } catch (Exception e) {
            throw new Twister2Exception("Can not get job submission znode data.", e);
        }
        try {
            Thread.sleep(sleepInterval);
        } catch (InterruptedException e) {
            LOG.warning("Sleeping thread interrupted.");
        }
        duration = System.currentTimeMillis() - startTime;
        checkCount++;
        if (duration > nextLogTime) {
            LOG.info("Still waiting for job submission time znode to be created: " + jstPath);
            nextLogTime += logInterval;
        }
    }
    throw new Twister2Exception("Job Submission Time znode is not created by job master " + "on the time limit: " + timeLimit + " ms");
}

Example 27 with Twister2Exception

use of edu.iu.dsc.tws.api.exceptions.Twister2Exception in project twister2 by DSC-SPIDAL.

the class ZKBarrierManager method createWorkerZNode.

/**
 * create a worker znode at the barrier directory
 */
public static void createWorkerZNode(CuratorFramework client, String workerPath, long timeout) throws Twister2Exception {
    try {
        client.create().creatingParentsIfNeeded().withMode(CreateMode.PERSISTENT).forPath(workerPath, Longs.toByteArray(timeout));
        LOG.info("Worker Barrier Znode created: " + workerPath);
    } catch (Exception e) {
        throw new Twister2Exception("Worker Barrier Znode can not be created for the path: " + workerPath, e);
    }
}

Example 28 with Twister2Exception

use of edu.iu.dsc.tws.api.exceptions.Twister2Exception in project twister2 by DSC-SPIDAL.

the class ZKBarrierManager method deleteWorkerZNode.

/**
 * delete the worker znode at a barrier directory
 */
public static void deleteWorkerZNode(CuratorFramework client, String workerPath) throws Twister2Exception {
    try {
        client.delete().forPath(workerPath);
        LOG.info("Worker Barrier Znode deleted: " + workerPath);
    } catch (Exception e) {
        throw new Twister2Exception("Worker Barrier Znode can not be deleted for the path: " + workerPath, e);
    }
}

Example 29 with Twister2Exception

use of edu.iu.dsc.tws.api.exceptions.Twister2Exception in project twister2 by DSC-SPIDAL.

the class ZKEphemStateManager method removeEphemZNode.

/**
 * remove ephemeral worker znode from previous run if exist
 */
public static void removeEphemZNode(CuratorFramework client, String rootPath, String jobID, int workerID) throws Twister2Exception {
    String ephemDirPath = ZKUtils.ephemDir(rootPath, jobID);
    try {
        List<String> children = client.getChildren().forPath(ephemDirPath);
        for (String childZnodeName : children) {
            int wID = ZKUtils.getWorkerIDFromEphemPath(childZnodeName);
            if (wID == workerID) {
                String wPath = ephemDirPath + "/" + childZnodeName;
                client.setData().forPath(wPath, DELETE_TAG.getBytes(StandardCharsets.UTF_8));
                client.delete().forPath(wPath);
                LOG.info("EphemeralWorkerZnode deleted from previous run: " + wPath);
            }
        }
    } catch (Exception e) {
        throw new Twister2Exception("Can not remove ephemeral worker znode.", e);
    }
}

Example 30 with Twister2Exception

use of edu.iu.dsc.tws.api.exceptions.Twister2Exception in project twister2 by DSC-SPIDAL.

the class KubernetesLauncher method startJobMasterOnClient.

/**
 * start the JobMaster locally on submitting client
 * this is a blocking call
 * it finishes after the job has completed
 */
private boolean startJobMasterOnClient(JobAPI.Job job) {
    // get Dashboard IP address from dashboard service name
    String dashAddress = JobMasterContext.dashboardHost(config);
    // get the IP address for the service name by querying Kubernetes master
    if (dashAddress.endsWith("svc.cluster.local")) {
        String dashIP = getDashboardIP(dashAddress);
        String dashURL = "http://" + dashIP;
        if (dashIP == null) {
            LOG.warning("Could not get Dashboard server IP address from dashboard service name: " + dashAddress + " will not connect to Dashboard. *****");
            dashURL = null;
        }
        config = JobMasterContext.updateDashboardHost(config, dashURL);
        LOG.info("Dashboard server HTTP URL: " + dashURL);
    }
    String hostAdress = RequestObjectBuilder.getJobMasterIP();
    JobMasterAPI.NodeInfo nodeInfo = NodeInfoUtils.createNodeInfo(hostAdress, null, null);
    K8sScaler k8sScaler = new K8sScaler(config, job, controller);
    JobMasterAPI.JobMasterState initialState = JobMasterAPI.JobMasterState.JM_STARTED;
    JobTerminator jobTerminator = new JobTerminator(config, controller);
    JobMaster jobMaster = new JobMaster(config, hostAdress, jobTerminator, job, nodeInfo, k8sScaler, initialState);
    jobMaster.addShutdownHook(true);
    // jobMaster.startJobMasterThreaded();
    try {
        jobMaster.startJobMasterBlocking();
    } catch (Twister2Exception e) {
        LOG.log(Level.SEVERE, e.getMessage(), e);
        return false;
    }
    return true;
}