public static void SubmitAndWait(ClusterSubmitterArgs clusterArgs, IDistributable distributableObj, int maxSubmitAfterTasksFail = 0)
{
using (ParallelOptionsScope.Suspend())
{
int numberOfTries = 0;
retry:
Submit(clusterArgs, distributableObj);
JobWaitingParams jobWaitingParams = WaitForJobInternal(clusterArgs);
if (jobWaitingParams.JobState == v2008R2.Properties.JobState.Canceled)
{
throw new Exception("Job canceled.");
}
else if (jobWaitingParams.JobState == v2008R2.Properties.JobState.Failed)
{
if (numberOfTries < maxSubmitAfterTasksFail)
{
++numberOfTries;
Console.WriteLine("Job failed, trying again...");
goto retry;
}
throw new Exception("Job failed.");
}
//HpcLib.HpcLib.CopyFiles(new List<String> { "" }, _remoteTaskOutDir, TASK_OUT_DIR);
}
}
public bool Refresh()
{
bool changed = false;
ISchedulerCounters counters;
using (ParallelOptionsScope.Suspend())
{
if (Connect() && null != (counters = GetCounters()))
{
changed =
BusyCores != counters.BusyCores ||
IdleCores != counters.IdleCores ||
QueuedTasks != counters.QueuedTasks;
BusyCores = counters.BusyCores;
IdleCores = counters.IdleCores;
QueuedTasks = counters.QueuedTasks;
}
else
{
changed = BusyCores != -1;
BusyCores = -1;
IdleCores = -1;
QueuedTasks = -1;
}
}
return(changed);
}
private static void SubmitInternal(ClusterSubmitterArgs clusterArgs, IDistributable distributableObj)
{
lock (_submitterLockObj) // for now, just let one thread submit at a time.
{
if (clusterArgs.Archive != null)
{
MBT.Escience.FileUtils.ArchiveExes(clusterArgs.Archive);
}
//ArgumentCollection argsToUse = (ArgumentCollection)applicationArgs.Clone();
CopyExes(clusterArgs);
clusterArgs.StdErrDirName = CreateUniqueDirectory(clusterArgs.ExternalRemoteDirectoryName, "Stderr", clusterArgs.Name);
clusterArgs.StdOutDirName = CreateUniqueDirectory(clusterArgs.ExternalRemoteDirectoryName, "Stdout", clusterArgs.Name);
if (clusterArgs.CopyInputFiles.Count > 0)
{
CopyInputFiles(clusterArgs.CopyInputFiles, clusterArgs.ExternalRemoteDirectoryName);
}
using (ParallelOptionsScope.Suspend())
{
switch (clusterArgs.Version)
{
case 1:
SubmitViaAPI1(clusterArgs, distributableObj);
break;
case 2:
Console.Error.WriteLine("Api2 and 3 are the same. Submitting via Api3.");
SubmitViaAPI3(clusterArgs, distributableObj);
break;
case 3:
SubmitViaAPI3(clusterArgs, distributableObj);
break;
default:
throw new NotSupportedException(string.Format("Cluster version {0} is not supported.", clusterArgs.Version));
}
}
Console.WriteLine("Processed job to cluster {0} with path {1}", clusterArgs.Cluster, clusterArgs.ExternalRemoteDirectoryName);
Console.WriteLine("Writing log file");
HpcLibSettings.TryWriteToLog(clusterArgs);
Console.WriteLine("Writing log entry to cluster directory");
HpcLibSettings.WriteLogEntryToClusterDirectory(clusterArgs);
Console.WriteLine("Done");
}
return;
}
public void ValidateSuspend()
{
using (ParallelOptionsScope scope = ParallelOptionsScope.Create(Constants.ParallelProcess))
{
Assert.IsNotNull(scope);
Assert.AreEqual(Constants.ParallelProcess, ParallelOptionsScope.Current.MaxDegreeOfParallelism);
Assert.IsTrue(ParallelOptionsScope.Exists);
IDisposable disposable = ParallelOptionsScope.Suspend();
Assert.IsFalse(ParallelOptionsScope.Exists);
disposable.Dispose();
}
ApplicationLog.WriteLine(string.Concat(
"ParallelOptionsScope BVT: Validation of Suspend method completed successfully."));
}
private bool Connect()
{
if (_scheduler != null)
{
return(true);
}
bool connected = false;
using (ParallelOptionsScope.Suspend())
{
connected = HpcLib.TryConnect(Cluster, out _scheduler);
}
return(connected);
}