public void FinalizeParse()
{
Helper.CheckCondition <ParseException>(NodeExclusionList.Count == 0 || Version == 2 || Version == 3, "-nodeExclusionList is only supported for API level 1");
Helper.CheckCondition <ParseException>(Version == 1 || Version == 2 || Version == 3, "Only Version 1 or 2 or 3 is supported.");
Helper.CheckCondition <ParseException>(TaskCount > 0, "You must specify a positive task count using -TaskCount");
Helper.CheckCondition <ParseException>(Cluster != null, "You must specify a cluster name using -cluster name");
if (Dir == null)
{
Dir = Environment.CurrentDirectory;
string username = HpcLib.GetCurrentUsername();
Dir = username + Dir.Substring(2); // first 2 characters are the drive. e.g. D:
RelativeDir = true;
}
if (ExeName == null)
{
ExeName = SpecialFunctions.GetEntryOrCallingAssembly().GetName().Name;
}
if (UsePasswordHack)
{
Password = HpcLib.GetPassword(out Username);
}
if (Name == null)
{
Name = ExeName;
}
if (Cluster.Equals(":auto:", StringComparison.CurrentCultureIgnoreCase) || Cluster.Equals("$auto$", StringComparison.CurrentCultureIgnoreCase))
{
Cluster = HpcLib.GetMostAppropriateCluster(TaskCount).Cluster;
}
}
public override void Distribute(Distribute.IDistributable distributableObject)
{
ClusterSubmitter.SubmitAndWait(this, distributableObject, MaxSubmitAfterTasksFail);
if (CopyResults.Count > 0)
{
HpcLib.CopyFiles(CopyResults, ExternalRemoteDirectoryName, ".");
}
}
private bool Connect()
{
if (_scheduler != null)
{
return(true);
}
return(HpcLib.TryConnect(Cluster, out _scheduler));
}
/// <summary>
/// Attempts to connect the the scheduler specified by Cluster and retrieve the Job specified by JobID. Will throw a SchedulerException if either of those fails.
/// </summary>
/// <returns></returns>
public v2.ISchedulerJob GetV2Job()
{
v2.ISchedulerJob job;
v2.IScheduler scheduler;
HpcLib.TryConnect(Cluster, out scheduler).Enforce <SchedulerException>("Unable to connect to head node {0}", Cluster);
HpcLib.TryGetJob(scheduler, Username, JobID, out job).Enforce <SchedulerException>("Unable to load jobID {0} for user {1}", JobID, Username);
return(job);
}
private bool Connect()
{
if (_scheduler != null)
{
return(true);
}
bool connected = false;
using (ParallelOptionsScope.Suspend())
{
connected = HpcLib.TryConnect(Cluster, out _scheduler);
}
return(connected);
}
public static void CopyInputFiles(List <string> filesToCopy, string baseDestnDir)
{
HpcLib.CopyFiles(filesToCopy, "", baseDestnDir);
//string xcopyArgsFormatString = GetXCopyArgsFormatString(filesToCopy[0], baseDestnDir);
//foreach (string file in filesToCopy)
//{
// Process proc = new Process();
// proc.StartInfo.FileName = "xcopy";
// proc.StartInfo.Arguments = string.Format(xcopyArgsFormatString, file);
// proc.StartInfo.UseShellExecute = false;
// //proc.StartInfo.RedirectStandardOutput = true;
// //Console.WriteLine("xcopy " + proc.StartInfo.Arguments);
// proc.Start();
// proc.WaitForExit();
// if (proc.ExitCode > 0)
// throw new Exception("Unable to copy file using command xcopy " + proc.StartInfo.Arguments);
//}
}
private ISchedulerJob GetJob()
{
lock (_getJobLocker)
{
HpcLib.TryGetJob(_scheduler, ClusterArgs.Username, ClusterArgs.JobID, out _job);
return(_job);
}
//IFilterCollection filter = _scheduler.CreateFilterCollection();
//filter.Add(FilterOperator.Equal, PropId.Job_Name, ClusterArgs.RunName);
//IIntCollection potentialIDs = _scheduler.GetJobIdList(filter, null);
//var matchingIDs = potentialIDs.Where(id => id == ClusterArgs.JobID);
//if (matchingIDs.Count() == 1)
//{
// int id = matchingIDs.First();
// ISchedulerJob job = _scheduler.OpenJob(id);
// return job;
//}
//return null;
}
private static void CopyExes(ClusterSubmitterArgs clusterArgs)
{
//!! why was this removed??
//var lockObj = CCSLibSettings.KnownClusters[clusterArgs.Cluster]; // use an object that is specific to this cluster.
//lock (lockObj) // we should only have one thread trying to copy to the cluster at a time. Esp since copying is smart and will reuse identical exes if they're already there.
{
if (true || clusterArgs.ExeRelativeDirectoryName == null)
{
clusterArgs.ExeRelativeDirectoryName = HpcLib.CopyExesToCluster(clusterArgs.ExternalRemoteDirectoryName, clusterArgs.Name);
}
else
{
Console.WriteLine("Using exe directory specified by user: "******"\\" + clusterArgs.ExeRelativeDirectoryName;
Helper.CheckCondition(Directory.Exists(absoluteExeDir), "Directory {0} does not exist!", absoluteExeDir);
//string exeRelativeName = "exes\\" + clusterArgs.ExeRelativeDirectoryName;
//clusterArgs.ExeRelativeDirectoryName = "\"" + exeRelativeName + "\"";
}
}
}
private static Timer HeartbeatTimer(int jobID, string clusterName, JobWaitingParams jobWaitingParams, int heartBeatPeriod)
{
Timer timer = new Timer(state =>
{
try
{
jobWaitingParams.Job.Refresh();
//Console.WriteLine("Job is still connected. Status is " + jobWaitingParams.JobState);
}
catch
{
Console.WriteLine("Lost connection to job. Attempting reconnect.");
v2008R2.IScheduler scheduler = null;
for (int iTry = 0; iTry < 10 && scheduler == null; iTry++)
{
HpcLib.TryConnect(clusterName, out scheduler);
}
if (scheduler == null)
{
Console.WriteLine("Unable to reconnect to cluster. Going back to sleep.");
}
else
{
jobWaitingParams.Job = scheduler.OpenJob(jobID);
jobWaitingParams.Job.Refresh();
SetupJobEventHandler(jobWaitingParams);
Console.WriteLine("Reconnect succeeded.");
}
}
jobWaitingParams.JobState = jobWaitingParams.Job.State;
if (JobIsFinished(jobWaitingParams.Job.State))
{
jobWaitingParams.ManualResetEvent.Set();
}
}, null, heartBeatPeriod, heartBeatPeriod);
return(timer);
}