/// <summary>
/// Get instance from job id
/// </summary>
/// <param name="clustername">cluster name</param>
/// <param name="jobID">job id</param>
/// <returns></returns>
public static LogEntry GetInstanceFromJobID(string clustername, int jobID)
{
LogEntry entry = new LogEntry();
entry.ClusterArgs = new ClusterSubmitterArgs();
entry.ClusterArgs.Cluster = clustername;
entry.ClusterArgs.JobID = jobID;
JobListener.TryConnect(clustername, jobID, null, out entry._jobListener).Enforce("Unable to recover job {0} from cluster {1}.", jobID, clustername);
//entry.Connect();
//entry._job = entry.GetJob();
//entry.ClusterArgs.Name = entry._job.Name;
entry.Date = entry._jobListener.Job.SubmitTime;
ISchedulerTask exampleTask = entry._jobListener.Job.GetTaskList(null, null, true).Cast <ISchedulerTask>().First();
entry.ClusterArgs.StdErrDirName = exampleTask.StdErrFilePath;
entry.ClusterArgs.StdOutDirName = exampleTask.StdOutFilePath;
entry.ClusterArgs.Dir = exampleTask.WorkDirectory;
string clusterPath = entry.ClusterDir.ToLower();
string rootClusterPath = Path.Combine(HpcLibSettings.KnownClusters[entry.Cluster].StoragePath, "username");
string relativeDir = clusterPath.Replace(rootClusterPath.ToLower(), "");
if (!relativeDir.StartsWith("\\"))
{
relativeDir = "\\" + relativeDir;
}
entry.LocalDir = @"d:\projects" + relativeDir;
return(entry);
}
/// <summary>
/// Connect to HPC
/// </summary>
/// <param name="schedulerName">scheduler name</param>
/// <param name="jobID">job id</param>
/// <param name="usernameOrNull">user name</param>
/// <param name="jobListener">job listener</param>
/// <returns></returns>
public static bool TryConnect(string schedulerName, int jobID, string usernameOrNull, out JobListener jobListener)
{
IScheduler scheduler;
ISchedulerJob job;
if (HpcLib.TryConnect(schedulerName, out scheduler) && HpcLib.TryGetJob(scheduler, usernameOrNull, jobID, out job))
{
jobListener = new JobListener(scheduler, job);
return(true);
}
else
{
jobListener = null;
return(false);
}
}
private static JobState WaitForJobInternal(ClusterSubmitterArgs clusterArgs, int maxNumTimesToResubmitFailedTasks)
{
JobListener jobListener;
JobListener.TryConnect(clusterArgs.Cluster, clusterArgs.JobID, clusterArgs.Username, out jobListener).Enforce("Could not connect to scheduler {0} or find jobID {1} for user {2}.",
clusterArgs.Cluster, clusterArgs.JobID, clusterArgs.Username);
ManualResetEvent mre = new ManualResetEvent(false);
// setup a notification for when the job is done.
jobListener.OnJobStateChanged += (o, e) =>
{
if (jobListener.JobIsDone)
{
mre.Set();
}
};
// setup the notification to requeue failed tasks.
if (maxNumTimesToResubmitFailedTasks > 0)
{
jobListener.OnTaskStateChanged += (o, e) =>
{
if (jobListener.JobCounters.FailedTaskCount > 0 && maxNumTimesToResubmitFailedTasks-- > 0)
{
Console.WriteLine(Resource.Tasks_failed);
RequeueFailedAndCanceledTasks(jobListener.Scheduler, jobListener.Job);
}
};
}
if (!jobListener.JobIsDone)
{
mre.WaitOne();
}
return(jobListener.JobState);
}
/// <summary>
/// Start tracking job
/// </summary>
/// <returns></returns>
public bool StartTrackingJob()
{
if (HpcLibSettings.ActiveClusters.Contains(Cluster, StringComparer.CurrentCultureIgnoreCase) && JobListener.TryConnect(Cluster, ClusterArgs.JobID, ClusterArgs.Username, out _jobListener))
{
_jobListener.OnJobStateChanged += (o, e) => { RefreshJobStatus(); RaiseJobStateChangedEvent(); };
_jobListener.OnTaskStateChanged += (o, e) => { RefreshJobStatus(); RaiseTaskStateChangedEvent(); };
if (JobState != _jobListener.JobState)
{
RefreshJobStatus();
RaiseJobStateChangedEvent();
RaiseTaskStateChangedEvent();
}
else if (JobState == Microsoft.Hpc.Scheduler.Properties.JobState.Running)
{
RefreshJobStatus();
RaiseTaskStateChangedEvent();
}
return(true);
}
else
{
return(false);
}
}