public static async Task MainAsync()
{
const string poolId = "MultiInstanceSamplePool";
const string jobId = "MultiInstanceSampleJob";
const string taskId = "MultiInstanceSampleTask";
const int numberOfNodes = 3;
// The application package and version to deploy to the compute nodes.
// It should contain your MPIHelloWorld sample MS-MPI program:
// https://blogs.technet.microsoft.com/windowshpc/2015/02/02/how-to-compile-and-run-a-simple-ms-mpi-program/
// And the MSMpiSetup.exe installer:
// https://www.microsoft.com/download/details.aspx?id=52981
// Then upload it as an application package:
// https://azure.microsoft.com/documentation/articles/batch-application-packages/
const string appPackageId = "MPIHelloWorld";
const string appPackageVersion = "1.0";
TimeSpan timeout = TimeSpan.FromMinutes(30);
// Configure your AccountSettings in the Microsoft.Azure.Batch.Samples.Common project within this solution
BatchSharedKeyCredentials cred = new BatchSharedKeyCredentials(AccountSettings.Default.BatchServiceUrl,
AccountSettings.Default.BatchAccountName,
AccountSettings.Default.BatchAccountKey);
using (BatchClient batchClient = BatchClient.Open(cred))
{
// Create the pool of compute nodes and the job to which we add the multi-instance task.
await CreatePoolAsync(batchClient, poolId, numberOfNodes, appPackageId, appPackageVersion);
await CreateJobAsync(batchClient, jobId, poolId);
// Create the multi-instance task. The MultiInstanceSettings property (configured
// below) tells Batch to create one primary and several subtasks, the total number
// of which matches the number of instances you specify in the MultiInstanceSettings.
// This main task's command line is the "application command," and is executed *only*
// by the primary, and only after the primary and all subtasks have executed the
// "coordination command" (the MultiInstanceSettings.CoordinationCommandLine).
CloudTask multiInstanceTask = new CloudTask(id: taskId,
commandline: $"cmd /c mpiexec.exe -c 1 -wdir %AZ_BATCH_TASK_SHARED_DIR% %AZ_BATCH_APP_PACKAGE_{appPackageId.ToUpper()}#{appPackageVersion}%\\MPIHelloWorld.exe");
// Configure the task's MultiInstanceSettings. Specify the number of nodes
// to allocate to the multi-instance task, and the "coordination command".
// The CoordinationCommandLine is run by the primary and subtasks, and is
// used in this sample to start SMPD on the compute nodes.
multiInstanceTask.MultiInstanceSettings =
new MultiInstanceSettings(@"cmd /c start cmd /c smpd.exe -d", numberOfNodes);
// Submit the task to the job. Batch will take care of creating one primary and
// enough subtasks to match the total number of nodes allocated to the task,
// and schedule them for execution on the nodes.
Console.WriteLine($"Adding task [{taskId}] to job [{jobId}]...");
await batchClient.JobOperations.AddTaskAsync(jobId, multiInstanceTask);
// Get the "bound" version of the multi-instance task.
CloudTask mainTask = await batchClient.JobOperations.GetTaskAsync(jobId, taskId);
// We use a TaskStateMonitor to monitor the state of our tasks. In this case,
// we will wait for the task to reach the Completed state.
Console.WriteLine($"Awaiting task completion, timeout in {timeout}...");
TaskStateMonitor taskStateMonitor = batchClient.Utilities.CreateTaskStateMonitor();
await taskStateMonitor.WhenAll(new List <CloudTask> {
mainTask
}, TaskState.Completed, timeout);
// Refresh the task to obtain up-to-date property values from Batch, such as
// its current state and information about the node on which it executed.
await mainTask.RefreshAsync();
string stdOut = mainTask.GetNodeFile(Constants.StandardOutFileName).ReadAsString();
string stdErr = mainTask.GetNodeFile(Constants.StandardErrorFileName).ReadAsString();
Console.WriteLine();
Console.WriteLine($"Main task [{mainTask.Id}] is in state [{mainTask.State}] and ran on compute node [{mainTask.ComputeNodeInformation.ComputeNodeId}]:");
Console.WriteLine("---- stdout.txt ----");
Console.WriteLine(stdOut);
Console.WriteLine("---- stderr.txt ----");
Console.WriteLine(stdErr);
// Need to delay a bit to allow the Batch service to mark the subtasks as Complete
TimeSpan subtaskTimeout = TimeSpan.FromSeconds(10);
Console.WriteLine($"Main task completed, waiting {subtaskTimeout} for subtasks to complete...");
System.Threading.Thread.Sleep(subtaskTimeout);
Console.WriteLine();
Console.WriteLine("---- Subtask information ----");
// Obtain the collection of subtasks for the multi-instance task, and print
// some information about each.
IPagedEnumerable <SubtaskInformation> subtasks = mainTask.ListSubtasks();
await subtasks.ForEachAsync(async (subtask) =>
{
Console.WriteLine("subtask: " + subtask.Id);
Console.WriteLine("\texit code: " + subtask.ExitCode);
if (subtask.State == SubtaskState.Completed)
{
// Obtain the file from the node on which the subtask executed. For normal CloudTasks,
// we could simply call CloudTask.GetNodeFile(Constants.StandardOutFileName), but the
// subtasks are not "normal" tasks in Batch, and thus must be handled differently.
ComputeNode node =
await batchClient.PoolOperations.GetComputeNodeAsync(subtask.ComputeNodeInformation.PoolId,
subtask.ComputeNodeInformation.ComputeNodeId);
string outPath = subtask.ComputeNodeInformation.TaskRootDirectory + "\\" + Constants.StandardOutFileName;
string errPath = subtask.ComputeNodeInformation.TaskRootDirectory + "\\" + Constants.StandardErrorFileName;
NodeFile stdOutFile = await node.GetNodeFileAsync(outPath.Trim('\\'));
NodeFile stdErrFile = await node.GetNodeFileAsync(errPath.Trim('\\'));
stdOut = await stdOutFile.ReadAsStringAsync();
stdErr = await stdErrFile.ReadAsStringAsync();
Console.WriteLine($"\tnode: " + node.Id);
Console.WriteLine("\tstdout.txt: " + stdOut);
Console.WriteLine("\tstderr.txt: " + stdErr);
}
else
{
Console.WriteLine($"\tSubtask {subtask.Id} is in state {subtask.State}");
}
});
// Clean up the resources we've created in the Batch account
Console.WriteLine();
Console.Write("Delete job? [yes] no: ");
string response = Console.ReadLine().ToLower();
if (response != "n" && response != "no")
{
await batchClient.JobOperations.DeleteJobAsync(jobId);
}
Console.Write("Delete pool? [yes] no: ");
response = Console.ReadLine().ToLower();
if (response != "n" && response != "no")
{
await batchClient.PoolOperations.DeletePoolAsync(poolId);
}
}
}
public static async Task MainAsync()
{
const string poolId = "MultiInstanceSamplePool";
const string jobId = "MultiInstanceSampleJob";
const string taskId = "MultiInstanceSampleTask";
const int numberOfNodes = 5;
//jmeno package kterou uploaduju na azure s polu s MSMpiSetup
const string appPackageId = "Parallel";
const string appPackageVersion = "1.0";
TimeSpan timeout = TimeSpan.FromMinutes(15);
AccountSettings accountSettings = SampleHelpers.LoadAccountSettings();
//nakonfigurované batch accounty abych se mohl připojit ke svému účtu
BatchSharedKeyCredentials cred = new BatchSharedKeyCredentials(
accountSettings.BatchServiceUrl,
accountSettings.BatchAccountName,
accountSettings.BatchAccountKey);
using (BatchClient batchClient = BatchClient.Open(cred))
{
// Vytvoření fondu výpočetních uzlů a úlohu, do které přidáme úlohu s více instancemi.
await CreatePoolAsync(batchClient, poolId, numberOfNodes, appPackageId, appPackageVersion);
await CreateJobAsync(batchClient, jobId, poolId);
//batch vytvoří jednu hlavní a několik dílčích úkolů
CloudTask multiInstanceTask = new CloudTask(id: taskId,
commandline: $"cmd /c mpiexec.exe -c 1 -wdir %AZ_BATCH_TASK_SHARED_DIR% %AZ_BATCH_APP_PACKAGE_{appPackageId.ToUpper()}#{appPackageVersion}%\\ParallelMpiApp.exe");
// příkaz SPMD = více samostatných procesorů současně spouští stejný program
multiInstanceTask.MultiInstanceSettings =
new MultiInstanceSettings(@"cmd /c start cmd /c smpd.exe -d", numberOfNodes);
//zadání úkolů, vytvoří se jeden primární a několik dílčích,
//aby odpovídaly počtu uzlů a naplánuje se jejich provedení v uzlech
Console.WriteLine($"Adding task [{taskId}] to job [{jobId}]...");
await batchClient.JobOperations.AddTaskAsync(jobId, multiInstanceTask);
//verze úlohy
CloudTask mainTask = await batchClient.JobOperations.GetTaskAsync(jobId, taskId);
// sledování stavu úkolů,čekáme až bude úloha dokončena
Console.WriteLine($"Awaiting task completion, timeout in {timeout}...");
TaskStateMonitor taskStateMonitor = batchClient.Utilities.CreateTaskStateMonitor();
await taskStateMonitor.WhenAll(new List <CloudTask> {
mainTask
}, TaskState.Completed, timeout);
//aktualizace úlohy
await mainTask.RefreshAsync();
string stdOut = mainTask.GetNodeFile(Constants.StandardOutFileName).ReadAsString();
string stdErr = mainTask.GetNodeFile(Constants.StandardErrorFileName).ReadAsString();
Console.WriteLine();
Console.WriteLine($"Main task [{mainTask.Id}] is in state [{mainTask.State}] and ran on compute node [{mainTask.ComputeNodeInformation.ComputeNodeId}]:");
Console.WriteLine("---- stdout.txt ----");
Console.WriteLine(stdOut);
Console.WriteLine("---- stderr.txt ----");
Console.WriteLine(stdErr);
// par sekund čas aby se stačily dílčí úlohy dokončit
TimeSpan subtaskTimeout = TimeSpan.FromSeconds(10);
Console.WriteLine($"Main task completed, waiting {subtaskTimeout} for subtasks to complete...");
System.Threading.Thread.Sleep(subtaskTimeout);
Console.WriteLine();
Console.WriteLine("---- Subtask information ----");
//kolekce dílčích úlohů a tisk informací o každém
IPagedEnumerable <SubtaskInformation> subtasks = mainTask.ListSubtasks();
await subtasks.ForEachAsync(async (subtask) =>
{
Console.WriteLine("subtask: " + subtask.Id);
Console.WriteLine("\texit code: " + subtask.ExitCode);
if (subtask.State == SubtaskState.Completed)
{
//získání souborů z uzlů
ComputeNode node =
await batchClient.PoolOperations.GetComputeNodeAsync(subtask.ComputeNodeInformation.PoolId,
subtask.ComputeNodeInformation.ComputeNodeId);
string outPath = subtask.ComputeNodeInformation.TaskRootDirectory + "\\" + Constants.StandardOutFileName;
string errPath = subtask.ComputeNodeInformation.TaskRootDirectory + "\\" + Constants.StandardErrorFileName;
NodeFile stdOutFile = await node.GetNodeFileAsync(outPath.Trim('\\'));
NodeFile stdErrFile = await node.GetNodeFileAsync(errPath.Trim('\\'));
stdOut = await stdOutFile.ReadAsStringAsync();
stdErr = await stdErrFile.ReadAsStringAsync();
Console.WriteLine($"\tnode: " + node.Id);
Console.WriteLine("\tstdout.txt: " + stdOut);
Console.WriteLine("\tstderr.txt: " + stdErr);
}
else
{
Console.WriteLine($"\tSubtask {subtask.Id} is in state {subtask.State}");
}
});
// vymazání zdrojů které jsme vytvořili, abychom to nemuseli dělat manuálně(fondy,úlohy)
Console.WriteLine();
Console.Write("Delete job? [yes] no: ");
string response = Console.ReadLine().ToLower();
if (response != "n" && response != "no")
{
await batchClient.JobOperations.DeleteJobAsync(jobId);
}
Console.Write("Delete pool? [yes] no: ");
response = Console.ReadLine().ToLower();
if (response != "n" && response != "no")
{
await batchClient.PoolOperations.DeletePoolAsync(poolId);
}
}
}