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

Example 6 with Twister2Exception

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

the class ZKMasterController method workerZnodeRemoved.

/**
 * when a worker znode is removed from the ephemeral znode of this job znode,
 * take necessary actions
 * Possibilities:
 * that worker may have completed and deleted its znode,
 * that worker may have failed
 * that worker may have been removed by scaling down
 * a failed and restarted worker may have deleted the znode from its previous run
 */
private void workerZnodeRemoved(PathChildrenCacheEvent event) {
    // if job master znode removed, it must have failed
    // job master is the last one to leave the job.
    // it does not send complete message as workers when it finishes.
    String workerPath = event.getData().getPath();
    int removedWorkerID = ZKUtils.getWorkerIDFromEphemPath(workerPath);
    // this is a scaled down worker, nothing to do
    if (scaledDownWorkers.contains(removedWorkerID)) {
        scaledDownWorkers.remove(Integer.valueOf(removedWorkerID));
        LOG.info("Removed scaled down worker: " + removedWorkerID);
        return;
    }
    // if the worker state is COMPLETED or FULLY_FAILED in workerMonitor, ignore this event.
    // those events are already handled by persistent state cache updates
    JobMasterAPI.WorkerState ws = workerMonitor.getWorkerState(removedWorkerID);
    if (ws == JobMasterAPI.WorkerState.COMPLETED || ws == JobMasterAPI.WorkerState.FULLY_FAILED) {
        return;
    }
    // get worker info and the state from persistent storage
    WorkerWithState workerWithState;
    try {
        workerWithState = ZKPersStateManager.getWorkerWithState(client, rootPath, jobID, removedWorkerID);
        if (workerWithState == null) {
            LOG.severe("worker[" + removedWorkerID + "] removed, but its data can not be retrieved.");
            return;
        }
    } catch (Twister2Exception e) {
        LOG.log(Level.SEVERE, "worker[" + removedWorkerID + "] removed, but its data can not be retrieved.", e);
        return;
    }
    String workerBodyText = ZKEphemStateManager.decodeWorkerZnodeBody(event.getData().getData());
    // otherwise, the worker failed. We inform the failureListener.
    if (workerWithState.getState() == JobMasterAPI.WorkerState.COMPLETED || workerWithState.getState() == JobMasterAPI.WorkerState.FULLY_FAILED) {
        // removed event received for completed or fullyfailed worker, nothing to do
        return;
    } else if (ZKEphemStateManager.DELETE_TAG.equals(workerBodyText)) {
        // restarting worker deleted the previous ephemeral znode
        // ignore this event, because the worker is already re-joining
        LOG.info("Restarting worker deleted znode from previous run: " + workerPath);
        return;
    } else {
        // worker failed
        LOG.info(String.format("Worker[%s] FAILED. Worker last status: %s", removedWorkerID, workerWithState.getState()));
        workerMonitor.failed(removedWorkerID);
        // workers may mark their status as FAILED before failing
        if (workerWithState.getState() != JobMasterAPI.WorkerState.FAILED) {
            try {
                ZKPersStateManager.updateWorkerStatus(client, rootPath, jobID, workerWithState.getInfo(), workerWithState.getRestartCount(), JobMasterAPI.WorkerState.FAILED);
            } catch (Twister2Exception e) {
                LOG.log(Level.SEVERE, e.getMessage(), e);
            }
        }
    }
}

Example 7 with Twister2Exception

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

the class ZKMasterController method workerRestarted.

@Override
public void workerRestarted(JobMasterAPI.WorkerInfo workerInfo) {
    // generate en event and inform all other workers
    JobMasterAPI.WorkerRestarted workerRestarted = JobMasterAPI.WorkerRestarted.newBuilder().setWorkerInfo(workerInfo).build();
    JobMasterAPI.JobEvent jobEvent = JobMasterAPI.JobEvent.newBuilder().setRestarted(workerRestarted).build();
    try {
        ZKEventsManager.publishEvent(client, rootPath, jobID, jobEvent);
    } catch (Twister2Exception e) {
        throw new Twister2RuntimeException(e);
    }
}

Example 8 with Twister2Exception

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

the class ZKBarrierHandler method getInitialWorkersAtBarrier.

private long getInitialWorkersAtBarrier(PathChildrenCache childrenCache, Set<Integer> workersAtBarrier) {
    long timeout = 0;
    List<ChildData> existingWorkerZnodes = childrenCache.getCurrentData();
    for (ChildData child : existingWorkerZnodes) {
        String fullPath = child.getPath();
        int workerID = ZKUtils.getWorkerIDFromPersPath(fullPath);
        workersAtBarrier.add(workerID);
        if (timeout == 0) {
            try {
                ZKBarrierManager.readWorkerTimeout(client, fullPath);
            } catch (Twister2Exception e) {
                throw new Twister2RuntimeException(e);
            }
        }
    }
    return timeout;
}

Example 9 with Twister2Exception

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

the class JobZNodeManager method updateJob.

public static boolean updateJob(CuratorFramework client, String rootPath, JobAPI.Job job, JobAPI.JobState state) throws Twister2Exception {
    String jobDir = ZKUtils.jobDir(rootPath, job.getJobId());
    JobWithState jobWithState = new JobWithState(job, state);
    try {
        client.setData().forPath(jobDir, jobWithState.toByteArray());
        return true;
    } catch (Exception e) {
        throw new Twister2Exception("Could not update Job in znode: " + jobDir, e);
    }
}

Example 10 with Twister2Exception

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

the class JobZNodeManager method getJobs.

/**
 * return all jobs
 */
public static List<JobWithState> getJobs(CuratorFramework client, String rootPath) throws Twister2Exception {
    try {
        List<String> jobPaths = client.getChildren().forPath(rootPath);
        LinkedList<JobWithState> jobs = new LinkedList();
        for (String jobID : jobPaths) {
            String childPath = rootPath + "/" + jobID;
            byte[] jobZNodeBody = client.getData().forPath(childPath);
            JobWithState jobWithState = JobWithState.decode(jobZNodeBody);
            jobs.add(jobWithState);
        }
        return jobs;
    } catch (Exception e) {
        throw new Twister2Exception("Could not get job znode data: " + rootPath, e);
    }
}