public virtual object Run(TaskRunContext task)
{
var node = GetNode(task);
var pack = PackageByName(node, task.PackageName);
var taskId = task.TaskId;
var farmId = task.Resource.Controller.FarmId;
var esService = GetExecuteServiceClient(node);
var resorceHomeFolder = IncarnationParams.IncarnatePath(node.DataFolders.LocalFolder, task.TaskId, farmId, CopyPhase.None);
PrepareEnviroment(esService, pack, resorceHomeFolder, farmId);
CopyInputFiles(task, resorceHomeFolder);
string cmdLine = String.Format(task.CommandLine, pack.AppPath, taskId);
Log.Info("cmdline = " + cmdLine);
var result = esService.ExecuteTaskOnFarm(taskId, farmId, cmdLine);
Log.Info("Exec done. Job id = " + result);
esService.Close();
return(result + "\n" + node.NodeName);
}
public string CopyInputFiles(TaskRunContext task, out string fileNames)
{
var node = GetNode(task);
//string ftpFolder = IncarnationParams.IncarnatePath(node.DataFolders.ExchangeUrlFromSystem, taskId, CopyPhase.In);
//string jobFtpFolder = IncarnationParams.IncarnatePath(node.DataFolders.ExchangeUrlFromSystem, taskId, CopyPhase.None);
//string ftpInputFolder = IncarnationParams.IncarnatePath(node.DataFolders.ExchangeUrlFromResource, taskId, CopyPhase.In);
//string ftpOutputFolder = IncarnationParams.IncarnatePath(node.DataFolders.ExchangeUrlFromResource, taskId, CopyPhase.Out);
string clusterHomeFolder = IncarnationParams.IncarnatePath(node.DataFolders.LocalFolder, task.TaskId, CopyPhase.None);
//IOProxy.Ftp.MakePath(ftpInputFolder);
//IOProxy.Ftp.MakePath(ftpOutputFolder);
try
{
Log.Info(Thread.CurrentThread.ManagedThreadId + " entered.");
SshExec(node, "mkdir " + clusterHomeFolder);
Log.Info(Thread.CurrentThread.ManagedThreadId + " exited.");
}
catch (Exception e)
{
Log.Warn(e.ToString());
}
Log.Info("Copying input files for task " + task.TaskId.ToString());
fileNames = ""; //String.Join(" ", incarnation.FilesToCopy.Select(f => f.FileName));
foreach (var file in task.InputFiles)
{
string tmpFile = Path.GetTempFileName();
IOProxy.Storage.Download(file.StorageId, tmpFile);
string fileOnCluster = clusterHomeFolder.TrimEnd(new[] { '/', '\\' }) + "/" + file.FileName;
fileNames += " " + fileOnCluster;
Log.Info("Copying file " + fileOnCluster);
//ScpCopy(node, fileOnCluster, tmpFile);
UploadFile(node, fileOnCluster, tmpFile);
File.Delete(tmpFile);
}
return(clusterHomeFolder);
}
protected override bool GetFromResourceTaskStateInfo(TaskRunContext task, out string result)
{
var node = GetNode(task);
try
{
result = SshExec(node, GetTaskStateCommand(), (string)task.LocalId, null).ToLowerInvariant();
}
catch (Exception e)
{
Log.Error(String.Format("Exception while updating task's {0} state: {1}", task.TaskId, e));
result = "SshExec error while updating task's state";
}
string clusterFolder = IncarnationParams.IncarnatePath(node.DataFolders.LocalFolder, task.TaskId, CopyPhase.Out);
string result2 = SshExec(node, SshUnixCommands.Ls, clusterFolder);
return(result.Contains(task.LocalId.ToString()) && !result2.Contains(ClavireFinishFileName));
}
public TaskStateInfo GetTaskStateInfo(TaskRunContext task)
{
lock (_gridLock)
{
RefreshCertificate();
ulong taskId = task.TaskId;
string localId = (string)task.LocalId;
string state = SshExec(PilotCommands.JobStatus, localId).ToLower();
if (state.Contains("is new"))
{
return(new TaskStateInfo(TaskState.Started, state));
}
//return Tuple.Create(TaskState.Scheduled, state);
if (state.Contains("is running") || state.Contains("is starting"))
{
return(new TaskStateInfo(TaskState.Started, state));
}
var node = GetNode(task);
string ftpOutFolderFromSystem = IncarnationParams.IncarnatePath(node.DataFolders.ExchangeUrlFromSystem, taskId, CopyPhase.Out);
string ftpOutFolderFromResource = IncarnationParams.IncarnatePath(node.DataFolders.ExchangeUrlFromResource, taskId, CopyPhase.Out);
string gridFolder = IncarnationParams.IncarnatePath(node.DataFolders.LocalFolder, taskId, CopyPhase.None);
IOProxy.Ftp.MakePath(ftpOutFolderFromSystem);
SshExec(PilotCommands.CopyFilesToGridFtp, gridFolder + " " + ftpOutFolderFromResource);
if (state.Contains("is finished"))
{
return(new TaskStateInfo(TaskState.Completed, state));
}
else
{
return(new TaskStateInfo(TaskState.Failed, state));
}
}
}
public void CopyOutputsToExchange(TaskRunContext task, string farmId)
{
ulong taskId = task.TaskId;
var node = GetNode(task);
var pack = PackageByName(node, task.PackageName);
// temporary hack: files are not pushed from resource => using path from resource for scp copying
string outFolderFromSystem = IncarnationParams.IncarnatePath(node.DataFolders.ExchangeUrlFromResource, taskId, CopyPhase.Out);
//string outFolderFromSystem = IncarnationParams.IncarnatePath(node.DataFolders.ExchangeUrlFromSystem, taskId, CopyPhase.Out);
bool copyingOutsToFtp = outFolderFromSystem.StartsWith("ftp://");
if (copyingOutsToFtp && !outFolderFromSystem.EndsWith("/"))
{
outFolderFromSystem += '/';
}
if (!copyingOutsToFtp && !outFolderFromSystem.EndsWith("\\"))
{
outFolderFromSystem += '\\';
}
string clusterFolder = IncarnationParams.IncarnatePath((!String.IsNullOrEmpty(pack.LocalDir)) ? String.Format(pack.LocalDir, task.TaskId) : node.DataFolders.LocalFolder, taskId, CopyPhase.Out);
if (!clusterFolder.EndsWith("\\"))
{
clusterFolder += "\\";
}
var exClient = GetExecuteServiceClient(node);
string[] fileNames = exClient.GetAllFileNames(farmId, taskId);
foreach (var output in task.ExpectedOutputFileNames)
{
Log.Info(output);
}
//IOProxy.Ftp.MakePath(ftpOutFolderFromSystem);
var dirStructure = fileNames
.Where(name => name.Contains('/') || name.Contains('\\')) // inside subdir
.Select(name => name.Remove(name.LastIndexOfAny(new[] { '\\', '/' })))
.Distinct()
.Select(file => outFolderFromSystem + file)
.Union(new[] { outFolderFromSystem });
foreach (string dir in dirStructure)
{
if (copyingOutsToFtp)
{
IOProxy.Ftp.MakePath(dir);
}
else
{
Log.Debug("Creating dir " + dir);
Directory.CreateDirectory(dir);
}
}
Log.Info("Copying output files");
//System.Threading.Tasks.Parallel.ForEach(fileNames, (fileName) =>
foreach (string fileName in fileNames)
{
//if (files.Contains(fileName))
{
string tmpFile = Path.GetTempFileName();
try
{
Log.Info("Copying file " + clusterFolder + fileName);
//ScpGet( node, clusterFolder + fileName, tmpFile, false);
DownloadFile(node, clusterFolder + fileName, tmpFile, taskId, farmId);
if (copyingOutsToFtp)
{
IOProxy.Ftp.UploadLocalFile(tmpFile, outFolderFromSystem, fileName, shouldCreatePath: false);
}
else
{
File.Copy(tmpFile, outFolderFromSystem + fileName);
}
File.Delete(tmpFile);
Log.Info("File copied " + fileName);
}
catch (Exception e)
{
Log.Warn(String.Format("Exception on file '{0}' copy: {1}", clusterFolder + fileName, e));
}
}
}//);
}
public object Run(TaskRunContext task)
{
lock (_gridLock)
{
RefreshCertificate();
//var incarnation = task.Incarnation;
string tmpFileName = null;
if (task.UserCert != null)
{
Log.Info("Using user's certificate");
tmpFileName = Path.GetTempFileName();
IOProxy.Storage.Download(task.UserCert, tmpFileName);
var scpForCert = new SSH.Scp(HELPER_SSH_HOST, HELPER_SSH_USER, HELPER_SSH_PASS);
scpForCert.Connect();
scpForCert.Recursive = true;
scpForCert.Put(tmpFileName, "/tmp/x509up_u500");
scpForCert.Close();
File.Delete(tmpFileName);
SshExec(PilotCommands.SetPermissionsOnProxyCertFile);
}
else
{
Log.Info("Using system's certificate");
}
try
{
long coresToUse = task.NodesConfig.Sum(cfg => cfg.Cores);
var node = GetNode(task);
var pack = node.PackageByName(task.PackageName);
// todo : remove
string commandLine = task.CommandLine;
commandLine = commandLine.Replace("java -jar ", "");
if (task.PackageName.ToLowerInvariant() == "cnm")
{
commandLine = commandLine.Replace("{0}", "ru.ifmo.hpc.main.ExtendedModel");
}
else
if (task.PackageName.ToLowerInvariant() == "ism")
{
commandLine = commandLine.Replace("{0}", "ru.ifmo.hpc.main.SpreadModel");
}
else
{
//if (task.PackageName.ToLowerInvariant() == "orca")
commandLine = commandLine.Replace("{0}", "");
}
string ftpFolderFromSystem = IncarnationParams.IncarnatePath(node.DataFolders.ExchangeUrlFromSystem, task.TaskId, CopyPhase.In);
string ftpFolderFromResource = IncarnationParams.IncarnatePath(node.DataFolders.ExchangeUrlFromResource, task.TaskId, CopyPhase.In);
string gridFtpFolder = IncarnationParams.IncarnatePath(node.DataFolders.LocalFolder, task.TaskId, CopyPhase.None);
SshExec(PilotCommands.MakeFolderOnGridFtp, gridFtpFolder);
string endl = "\n";
// Сначала дописываем недостающий входной файл (скрипт запуска пакета на кластере)
string scriptName = pack.AppPath;
//if (pack.EnvVars.Any())
{
// Файл с установкой переменных окружения, если пакет их использует
scriptName = "run.sh";
var scriptContent = new StringBuilder();
scriptContent.Append("#!/bin/bash" + endl);
foreach (var pair in pack.EnvVars)
{
scriptContent.AppendFormat("export {0}={1}" + endl, pair.Key, pair.Value);
}
scriptContent.Append(pack.AppPath);
/*
* if (task.PackageName.ToLowerInvariant() == "orca")
* {
* string[] args = commandLine.Split(new[] { ' ' }, StringSplitOptions.RemoveEmptyEntries);
* for (int i = 0; i < args.Length; i++)
* {
* if (args[i] == "orca.out")
* scriptContent.Append(" >");
*
* scriptContent.Append(" $" + (i + 1).ToString());
* }
* }
* else*/
{
scriptContent.Append(" " + commandLine);
}
string scriptLocalPath = Path.GetTempFileName();
File.WriteAllText(scriptLocalPath, scriptContent.ToString());
IOProxy.Ftp.UploadLocalFile(scriptLocalPath, ftpFolderFromSystem, scriptName);
File.Delete(scriptLocalPath);
}
//IOProxy.Ftp.UploadLocalFile(DEFAULT_JOB_LAUNCHER_PATH, GetFtpInputFolder(taskId), Path.GetFileName(DEFAULT_JOB_LAUNCHER_PATH));
// Копируем входные файлы с ФТП на ГридФТП
SshExec(PilotCommands.CopyFilesToGridFtp, ftpFolderFromResource + " " + gridFtpFolder);
SshExec(PilotCommands.MakeFilesExecutableOnGridFtp, gridFtpFolder + "*");
// Формируем описание задания для грида
var jobFileContent = new StringBuilder();
jobFileContent.AppendFormat(@"{{ ""version"": 2, ""description"": ""{0}""," + endl, task.TaskId);
jobFileContent.AppendFormat(@" ""default_storage_base"": ""{0}""," + endl, gridFtpFolder);
jobFileContent.AppendFormat(@" ""tasks"": [ {{ ""id"": ""a"", ""description"": ""task"", ""definition"": {{ ""version"": 2," + endl);
jobFileContent.AppendFormat(@" ""executable"": ""{0}""," + endl, scriptName);
//jobFileContent.AppendFormat(@" ""arguments"": [ ""{0}"" ]," + endl, String.Join(@""", """, args));
jobFileContent.AppendFormat(@" ""input_files"": {{" + endl);
if (scriptName == "run.sh") // todo : if no input files?
{
jobFileContent.AppendFormat(@" ""run.sh"": ""run.sh""," + endl);
}
jobFileContent.AppendFormat(@" " + String.Join(
"," + endl + " ",
task.InputFiles.Select(
file => String.Format(@"""{0}"": ""{0}""", file.FileName)
)
));
jobFileContent.AppendFormat(endl + @" }}," + endl);
jobFileContent.AppendFormat(@" ""output_files"": {{" + endl);
//if (task.PackageName.ToLowerInvariant() == "cnm")
// jobFileContent.AppendFormat(@" ""output.dat"": ""output.dat""" + endl);
//else
if (task.PackageName.ToLowerInvariant() == "ism")
{
jobFileContent.AppendFormat(@" ""output.dat"": ""output.dat""" + endl);
}
else
if (task.PackageName.ToLowerInvariant() == "orca")
{
jobFileContent.AppendFormat(@" ""orca.out"": ""orca.out""," + endl);
jobFileContent.AppendFormat(@" ""eldens.cube"": ""eldens.cube""" + endl);
}
else
{
jobFileContent.AppendFormat(@" " + String.Join(
"," + endl + " ",
task.ExpectedOutputFileNames
.Where(name => name != "std.out" && name != "std.err")
.Select(
name => String.Format(@"""{0}"": ""{0}""", name)
)
) + endl);
}
jobFileContent.AppendFormat(@" }}," + endl);
jobFileContent.AppendFormat(@" ""stdout"": ""std.out"", ""stderr"": ""std.err"", " + endl);
jobFileContent.AppendFormat(@" ""count"": {0}" + endl, coresToUse);
if (pack.Params.ContainsKey("requirements"))
{
jobFileContent.AppendFormat(@" ,""requirements"": {0}" + endl, pack.Params["requirements"]);
}
jobFileContent.AppendFormat(@" }} }} ]," + endl);
jobFileContent.AppendFormat(@" ""requirements"": {{ ""hostname"": [""{0}""]", node.NodeAddress);
//if (pack.Params.ContainsKey("requirements"))
// jobFileContent.AppendFormat(@", {0}" + endl, pack.Params["requirements"]);
jobFileContent.AppendFormat(@"}}" + endl + "}}", node.NodeAddress);
Log.Debug(String.Format("Task's '{0}' grid job JSON: ", task.TaskId, jobFileContent));
string jobFileName = "job_" + task.TaskId.ToString() + ".js";
string jobFilePathOnHelper = JOBS_FOLDER_ON_HELPER + jobFileName;
//string jobFileContent = File.ReadAllText(DEFAULT_JOB_DESCR_PATH).Replace(GRIDFTP_PATH_TOKEN, taskFolderOnGridFtp);
string jobFilePathLocal = Path.GetTempFileName();
File.WriteAllText(jobFilePathLocal, jobFileContent.ToString());
// Записываем его на сервер с Пилотом
var scp = new SSH.Scp(HELPER_SSH_HOST, HELPER_SSH_USER, HELPER_SSH_PASS);
/*
* var notifier = new JobDescriptionUploadNotifier(TaskId, Cluster, RunParams);
* scp.OnTransferEnd += new SSH.FileTransferEvent(notifier.OnFinish); // todo : необязательно
*/
scp.Connect();
scp.Recursive = true;
scp.Put(jobFilePathLocal, jobFilePathOnHelper);
scp.Close();
File.Delete(jobFilePathLocal); // todo : remove files on helper and gridftp
// Запускаем
Log.Info(String.Format(
"Trying to exec task {0} on grid cluster {1}",
task.TaskId, node.NodeName
));
string launchResult = SshExec(PilotCommands.SubmitJob, jobFilePathOnHelper, pilotUrl: node.Services.ExecutionUrl);
int urlPos = launchResult.IndexOf("https://");
string jobUrl = launchResult.Substring(urlPos).Trim() + "a";
Log.Debug(jobUrl);
Log.Info(String.Format(
"Task {0} launched on grid with jobUrl = {1}",
task.TaskId, jobUrl
));
return(jobUrl);
}
catch (Exception e)
{
Log.Error(String.Format(
"Error while starting task {0} in grid: {1}\n{2}",
task.TaskId, e.Message, e.StackTrace
));
throw;
}
finally
{
if (task.UserCert != null)
{
Log.Info("Wiping user's certificate");
tmpFileName = Path.GetTempFileName();
File.WriteAllText(tmpFileName, "Wiped by Easis system");
var scpForCert = new SSH.Scp(HELPER_SSH_HOST, HELPER_SSH_USER, HELPER_SSH_PASS);
scpForCert.Connect();
scpForCert.Recursive = true;
scpForCert.Put(tmpFileName, "/tmp/x509up_u500");
scpForCert.Close();
File.Delete(tmpFileName);
SshExec(PilotCommands.SetPermissionsOnProxyCertFile);
}
}
}
}
public void CopyOutputsToExchange(TaskRunContext task)
{
ulong taskId = task.TaskId;
var node = GetNode(task);
var pack = PackageByName(node, task.PackageName);
// temporary hack: files are not pushed from resource => using path from resource for scp copying
string outFolderFromSystem = IncarnationParams.IncarnatePath(node.DataFolders.ExchangeUrlFromResource, taskId, CopyPhase.Out);
//string outFolderFromSystem = IncarnationParams.IncarnatePath(node.DataFolders.ExchangeUrlFromSystem, taskId, CopyPhase.Out);
bool copyingOutsToFtp = outFolderFromSystem.StartsWith("ftp://");
if (copyingOutsToFtp && !outFolderFromSystem.EndsWith("/"))
{
outFolderFromSystem += '/';
}
if (!copyingOutsToFtp && !outFolderFromSystem.EndsWith("\\"))
{
outFolderFromSystem += '\\';
}
string clusterFolder = IncarnationParams.IncarnatePath((!String.IsNullOrEmpty(pack.LocalDir)) ? String.Format(pack.LocalDir, task.TaskId) : node.DataFolders.LocalFolder, taskId, CopyPhase.Out);
if (!clusterFolder.EndsWith("/"))
{
clusterFolder += "/";
}
//var files = ImproveFiles(task.Incarnation.ExpectedOutputFileNames);
/* var fileNames =
* SshExec(node, SshPbsCommands.Find, clusterFolder)
* .Split(new[] { ", ", "," }, StringSplitOptions.RemoveEmptyEntries)
* .Where(st => !st.Contains("/"))
* .Select(st => st.Replace("*", "").Replace("|", "").Replace("\n",""))
* .Where(st => !st.Contains(".rst") && !st.Contains(".err") && !st.Contains(".esav"));*/
var fileNames = SshExec(node, "cd " + clusterFolder + "; " + SSH_FIND_COMMAND, "")
.Replace("./", "/").Split(new[] { "\n" }, StringSplitOptions.RemoveEmptyEntries)
.Where(st => !st.Contains(".rst") /*&& !st.Contains(".err")*/ && !st.Contains(".esav"))
.Select(st => st.Trim(new[] { '/', '\\' }));
//IOProxy.Ftp.MakePath(ftpOutFolderFromSystem);
var dirStructure = fileNames
.Where(name => name.Contains('/') || name.Contains('\\')) // inside subdir
.Select(name => name.Remove(name.LastIndexOfAny(new[] { '\\', '/' })))
.Distinct()
.Select(file => outFolderFromSystem + file)
.Union(new[] { outFolderFromSystem });
foreach (string dir in dirStructure)
{
if (copyingOutsToFtp)
{
IOProxy.Ftp.MakePath(dir);
}
else
{
Log.Debug("Creating dir " + dir);
Directory.CreateDirectory(dir);
}
}
Log.Info("Copying output files");
//System.Threading.Tasks.Parallel.ForEach(fileNames, (fileName) =>
foreach (string fileName in fileNames)
{
//if (files.Contains(fileName))
{
string tmpFile = Path.GetTempFileName();
try
{
Log.Info("Copying file " + clusterFolder + fileName);
//ScpGet(node, clusterFolder + fileName, tmpFile, false);
DownloadFile(node, clusterFolder + fileName, tmpFile);
if (copyingOutsToFtp)
{
IOProxy.Ftp.UploadLocalFile(tmpFile, outFolderFromSystem, fileName, shouldCreatePath: false);
}
else
{
File.Copy(tmpFile, outFolderFromSystem + fileName);
}
File.Delete(tmpFile);
Log.Info("File copied " + fileName);
}
catch (Ssh.SshTransferException e)
{
Log.Warn(String.Format("During coping file {0} for task {1} from error was happend: {2}", fileName, taskId, e)); // todo : lolwut?
}
catch (Exception e)
{
Log.Warn(String.Format("Exception on file '{0}' copy: {1}", clusterFolder + fileName, e));
}
}
}//);
}
