/// <summary>
/// Creates a job and adds a task to it. The task is a
/// custom executable which has a resource file associated with it.
/// </summary>
/// <param name="batchClient">The BatchClient to use when interacting with the Batch service.</param>
/// <param name="storageAccount">The cloud storage account to upload files to.</param>
/// <param name="jobId">The ID of the job.</param>
/// <returns>An asynchronous <see cref="Task"/> representing the operation.</returns>
private async Task SubmitJobAsync(BatchClient batchClient, CloudStorageAccount storageAccount, string jobId)
{
// create an empty unbound Job
CloudJob unboundJob = batchClient.JobOperations.CreateJob();
unboundJob.Id = jobId;
unboundJob.PoolInformation = new PoolInformation()
{
PoolId = this.jobManagerSettings.PoolId
};
// Upload the required files for the job manager task
await SampleHelpers.UploadResourcesAsync(storageAccount, this.jobManagerSettings.BlobContainer, JobManagerRequiredFiles);
List <ResourceFile> jobManagerResourceFiles = await SampleHelpers.UploadResourcesAndCreateResourceFileReferencesAsync(
storageAccount,
this.jobManagerSettings.BlobContainer,
JobManagerRequiredFiles);
// Set up the JobManager environment settings
List <EnvironmentSetting> jobManagerEnvironmentSettings = new List <EnvironmentSetting>()
{
// No need to pass the batch account name as an environment variable since the batch service provides
// an environment variable for each task which contains the account name
new EnvironmentSetting("SAMPLE_BATCH_KEY", this.accountSettings.BatchAccountKey),
new EnvironmentSetting("SAMPLE_BATCH_URL", this.accountSettings.BatchServiceUrl),
new EnvironmentSetting("SAMPLE_STORAGE_ACCOUNT", this.accountSettings.StorageAccountName),
new EnvironmentSetting("SAMPLE_STORAGE_KEY", this.accountSettings.StorageAccountKey),
new EnvironmentSetting("SAMPLE_STORAGE_URL", this.accountSettings.StorageServiceUrl),
};
unboundJob.JobManagerTask = new JobManagerTask()
{
Id = JobManagerTaskId,
CommandLine = JobManagerTaskExe,
ResourceFiles = jobManagerResourceFiles,
KillJobOnCompletion = true,
EnvironmentSettings = jobManagerEnvironmentSettings
};
// Commit Job to create it in the service
await unboundJob.CommitAsync();
}
/// <summary>
/// Populates Azure Storage with the required files, and
/// submits the job to the Azure Batch service.
/// </summary>
public async Task RunAsync()
{
Console.WriteLine("Running with the following settings: ");
Console.WriteLine("----------------------------------------");
Console.WriteLine(this.textSearchSettings.ToString());
Console.WriteLine(this.accountSettings.ToString());
CloudStorageAccount cloudStorageAccount = new CloudStorageAccount(
new StorageCredentials(
this.accountSettings.StorageAccountName,
this.accountSettings.StorageAccountKey),
this.accountSettings.StorageServiceUrl,
useHttps: true);
//Upload resources if required.
if (this.textSearchSettings.ShouldUploadResources)
{
Console.WriteLine("Splitting file: {0} into {1} subfiles",
Constants.TextFilePath,
this.textSearchSettings.NumberOfMapperTasks);
//Split the text file into the correct number of files for consumption by the mapper tasks.
FileSplitter splitter = new FileSplitter();
List <string> mapperTaskFiles = await splitter.SplitAsync(
Constants.TextFilePath,
this.textSearchSettings.NumberOfMapperTasks);
List <string> files = Constants.RequiredExecutableFiles.Union(mapperTaskFiles).ToList();
await SampleHelpers.UploadResourcesAsync(
cloudStorageAccount,
this.textSearchSettings.BlobContainer,
files);
}
//Generate a SAS for the container.
string containerSasUrl = SampleHelpers.ConstructContainerSas(
cloudStorageAccount,
this.textSearchSettings.BlobContainer);
//Set up the Batch Service credentials used to authenticate with the Batch Service.
BatchSharedKeyCredentials credentials = new BatchSharedKeyCredentials(
this.accountSettings.BatchServiceUrl,
this.accountSettings.BatchAccountName,
this.accountSettings.BatchAccountKey);
using (BatchClient batchClient = await BatchClient.OpenAsync(credentials))
{
//
// Construct the job properties in local memory before commiting them to the Batch Service.
//
//Allow enough compute nodes in the pool to run each mapper task, and 1 extra to run the job manager.
int numberOfPoolComputeNodes = 1 + this.textSearchSettings.NumberOfMapperTasks;
//Define the pool specification for the pool which the job will run on.
PoolSpecification poolSpecification = new PoolSpecification()
{
TargetDedicated = numberOfPoolComputeNodes,
VirtualMachineSize = "small",
//You can learn more about os families and versions at:
//http://azure.microsoft.com/documentation/articles/cloud-services-guestos-update-matrix
OSFamily = "4",
TargetOSVersion = "*"
};
//Use the auto pool feature of the Batch Service to create a pool when the job is created.
//This creates a new pool for each job which is added.
AutoPoolSpecification autoPoolSpecification = new AutoPoolSpecification()
{
AutoPoolIdPrefix = "TextSearchPool",
KeepAlive = false,
PoolLifetimeOption = PoolLifetimeOption.Job,
PoolSpecification = poolSpecification
};
//Define the pool information for this job -- it will run on the pool defined by the auto pool specification above.
PoolInformation poolInformation = new PoolInformation()
{
AutoPoolSpecification = autoPoolSpecification
};
//Define the job manager for this job. This job manager will run first and will submit the tasks for
//the job. The job manager is the executable which manages the lifetime of the job
//and all tasks which should run for the job. In this case, the job manager submits the mapper and reducer tasks.
List <ResourceFile> jobManagerResourceFiles = SampleHelpers.GetResourceFiles(containerSasUrl, Constants.RequiredExecutableFiles);
const string jobManagerTaskId = "JobManager";
JobManagerTask jobManagerTask = new JobManagerTask()
{
ResourceFiles = jobManagerResourceFiles,
CommandLine = Constants.JobManagerExecutable,
//Determines if the job should terminate when the job manager process exits.
KillJobOnCompletion = true,
Id = jobManagerTaskId
};
//Create the unbound job in local memory. An object which exists only in local memory (and not on the Batch Service) is "unbound".
string jobId = Environment.GetEnvironmentVariable("USERNAME") + DateTime.UtcNow.ToString("yyyyMMdd-HHmmss");
CloudJob unboundJob = batchClient.JobOperations.CreateJob(jobId, poolInformation);
unboundJob.JobManagerTask = jobManagerTask; //Assign the job manager task to this job
try
{
//Commit the unbound job to the Batch Service.
Console.WriteLine("Adding job: {0} to the Batch Service.", unboundJob.Id);
await unboundJob.CommitAsync(); //Issues a request to the Batch Service to add the job which was defined above.
//
// Wait for the job manager task to complete.
//
//An object which is backed by a corresponding Batch Service object is "bound."
CloudJob boundJob = await batchClient.JobOperations.GetJobAsync(jobId);
CloudTask boundJobManagerTask = await boundJob.GetTaskAsync(jobManagerTaskId);
TimeSpan maxJobCompletionTimeout = TimeSpan.FromMinutes(30);
// Monitor the current tasks to see when they are done.
// Occasionally a task may get killed and requeued during an upgrade or hardware failure, including the job manager
// task. The job manager will be re-run in this case. Robustness against this was not added into the sample for
// simplicity, but should be added into any production code.
Console.WriteLine("Waiting for job's tasks to complete");
TaskStateMonitor taskStateMonitor = batchClient.Utilities.CreateTaskStateMonitor();
bool timedOut = await taskStateMonitor.WaitAllAsync(new List <CloudTask> {
boundJobManagerTask
}, TaskState.Completed, maxJobCompletionTimeout);
Console.WriteLine("Done waiting for job manager task.");
await boundJobManagerTask.RefreshAsync();
//Check to ensure the job manager task exited successfully.
await Helpers.CheckForTaskSuccessAsync(boundJobManagerTask, dumpStandardOutOnTaskSuccess : false);
if (timedOut)
{
throw new TimeoutException(string.Format("Timed out waiting for job manager task to complete."));
}
//
// Download and write out the reducer tasks output
//
string reducerText = await SampleHelpers.DownloadBlobTextAsync(cloudStorageAccount, this.textSearchSettings.BlobContainer, Constants.ReducerTaskResultBlobName);
Console.WriteLine("Reducer reuslts:");
Console.WriteLine(reducerText);
}
finally
{
//Delete the job.
//This will delete the auto pool associated with the job as long as the pool
//keep alive property is set to false.
if (this.textSearchSettings.ShouldDeleteJob)
{
Console.WriteLine("Deleting job {0}", jobId);
batchClient.JobOperations.DeleteJob(jobId);
}
//Note that there were files uploaded to a container specified in the
//configuration file. This container will not be deleted or cleaned up by this sample.
}
}
}
public async static Task JobMain(string[] args)
{
//Load the configuration
Settings topNWordsConfiguration = Settings.Default;
AccountSettings accountSettings = AccountSettings.Default;
CloudStorageAccount cloudStorageAccount = new CloudStorageAccount(
new StorageCredentials(
accountSettings.StorageAccountName,
accountSettings.StorageAccountKey),
accountSettings.StorageServiceUrl,
useHttps: true);
StagingStorageAccount stagingStorageAccount = new StagingStorageAccount(
accountSettings.StorageAccountName,
accountSettings.StorageAccountKey,
cloudStorageAccount.BlobEndpoint.ToString());
using (BatchClient client = BatchClient.Open(new BatchSharedKeyCredentials(accountSettings.BatchServiceUrl, accountSettings.BatchAccountName, accountSettings.BatchAccountKey)))
{
string stagingContainer = null;
//OSFamily 4 == OS 2012 R2. You can learn more about os families and versions at:
//http://msdn.microsoft.com/en-us/library/azure/ee924680.aspx
CloudPool pool = client.PoolOperations.CreatePool(
topNWordsConfiguration.PoolId,
targetDedicatedComputeNodes: topNWordsConfiguration.PoolNodeCount,
virtualMachineSize: "small",
cloudServiceConfiguration: new CloudServiceConfiguration(osFamily: "4"));
List <string> files = new List <string>
{
Path.Combine(BatchStartTaskFolderName, BatchStartTaskTelemetryRunnerName),
};
files.AddRange(AIFilesToUpload);
var resourceHelperTask = SampleHelpers.UploadResourcesAndCreateResourceFileReferencesAsync(
cloudStorageAccount,
AIBlobConatinerName,
files);
List <ResourceFile> resourceFiles = resourceHelperTask.Result;
pool.StartTask = new StartTask()
{
CommandLine = string.Format("cmd /c {0}", BatchStartTaskTelemetryRunnerName),
ResourceFiles = resourceFiles
};
Console.WriteLine("Adding pool {0}", topNWordsConfiguration.PoolId);
try
{
await GettingStartedCommon.CreatePoolIfNotExistAsync(client, pool);
}
catch (AggregateException ae)
{
// Go through all exceptions and dump useful information
ae.Handle(x =>
{
Console.Error.WriteLine("Creating pool ID {0} failed", topNWordsConfiguration.PoolId);
if (x is BatchException)
{
BatchException be = x as BatchException;
Console.WriteLine(be.ToString());
Console.WriteLine();
}
else
{
Console.WriteLine(x);
}
// can't continue without a pool
return(false);
});
}
catch (BatchException be)
{
Console.Error.WriteLine("Creating pool ID {0} failed", topNWordsConfiguration.PoolId);
Console.WriteLine(be.ToString());
Console.WriteLine();
}
try
{
Console.WriteLine("Creating job: " + topNWordsConfiguration.JobId);
// get an empty unbound Job
CloudJob unboundJob = client.JobOperations.CreateJob();
unboundJob.Id = topNWordsConfiguration.JobId;
unboundJob.PoolInformation = new PoolInformation()
{
PoolId = topNWordsConfiguration.PoolId
};
// Commit Job to create it in the service
await unboundJob.CommitAsync();
// create file staging objects that represent the executable and its dependent assembly to run as the task.
// These files are copied to every node before the corresponding task is scheduled to run on that node.
FileToStage topNWordExe = new FileToStage(TopNWordsExeName, stagingStorageAccount);
FileToStage storageDll = new FileToStage(StorageClientDllName, stagingStorageAccount);
// Upload application insights assemblies
List <FileToStage> aiStagedFiles = new List <FileToStage>();
foreach (string aiFile in AIFilesToUpload)
{
aiStagedFiles.Add(new FileToStage(aiFile, stagingStorageAccount));
}
// In this sample, the input data is copied separately to Storage and its URI is passed to the task as an argument.
// This approach is appropriate when the amount of input data is large such that copying it to every node via FileStaging
// is not desired and the number of tasks is small since a large number of readers of the blob might get throttled
// by Storage which will lengthen the overall processing time.
//
// You'll need to observe the behavior and use published techniques for finding the right balance of performance versus
// complexity.
string[] documents = Directory.GetFiles(topNWordsConfiguration.DocumentsRootPath);
await SampleHelpers.UploadResourcesAsync(cloudStorageAccount, BooksContainerName, documents);
// initialize a collection to hold the tasks that will be submitted in their entirety
List <CloudTask> tasksToRun = new List <CloudTask>(documents.Length);
for (int i = 0; i < documents.Length; i++)
{
CloudTask task = new CloudTask("task_no_" + i, String.Format("{0} --Task {1} {2} {3} {4}",
TopNWordsExeName,
"https://onbehalfoutput.blob.core.windows.net/" + documents[i],
topNWordsConfiguration.TopWordCount,
accountSettings.StorageAccountName,
accountSettings.StorageAccountKey));
//This is the list of files to stage to a container -- for each job, one container is created and
//files all resolve to Azure Blobs by their name (so two tasks with the same named file will create just 1 blob in
//the container).
task.FilesToStage = new List <IFileStagingProvider>
{
topNWordExe,
storageDll,
};
foreach (FileToStage stagedFile in aiStagedFiles)
{
task.FilesToStage.Add(stagedFile);
}
tasksToRun.Add(task);
}
// Commit all the tasks to the Batch Service. Ask AddTask to return information about the files that were staged.
// The container information is used later on to remove these files from Storage.
ConcurrentBag <ConcurrentDictionary <Type, IFileStagingArtifact> > fsArtifactBag = new ConcurrentBag <ConcurrentDictionary <Type, IFileStagingArtifact> >();
client.JobOperations.AddTask(topNWordsConfiguration.JobId, tasksToRun, fileStagingArtifacts: fsArtifactBag);
// loop through the bag of artifacts, looking for the one that matches our staged files. Once there,
// capture the name of the container holding the files so they can be deleted later on if that option
// was configured in the settings.
foreach (var fsBagItem in fsArtifactBag)
{
IFileStagingArtifact fsValue;
if (fsBagItem.TryGetValue(typeof(FileToStage), out fsValue))
{
SequentialFileStagingArtifact stagingArtifact = fsValue as SequentialFileStagingArtifact;
if (stagingArtifact != null)
{
stagingContainer = stagingArtifact.BlobContainerCreated;
Console.WriteLine(
"Uploaded files to container: {0} -- you will be charged for their storage unless you delete them.",
stagingArtifact.BlobContainerCreated);
}
}
}
//Get the job to monitor status.
CloudJob job = client.JobOperations.GetJob(topNWordsConfiguration.JobId);
Console.Write("Waiting for tasks to complete ... ");
// Wait 20 minutes for all tasks to reach the completed state. The long timeout is necessary for the first
// time a pool is created in order to allow nodes to be added to the pool and initialized to run tasks.
IPagedEnumerable <CloudTask> ourTasks = job.ListTasks(new ODATADetailLevel(selectClause: "id"));
client.Utilities.CreateTaskStateMonitor().WaitAll(ourTasks, TaskState.Completed, TimeSpan.FromMinutes(20));
Console.WriteLine("tasks are done.");
foreach (CloudTask t in ourTasks)
{
Console.WriteLine("Task " + t.Id);
Console.WriteLine("stdout:" + Environment.NewLine + t.GetNodeFile(Batch.Constants.StandardOutFileName).ReadAsString());
Console.WriteLine();
Console.WriteLine("stderr:" + Environment.NewLine + t.GetNodeFile(Batch.Constants.StandardErrorFileName).ReadAsString());
}
}
finally
{
//Delete the pool that we created
if (topNWordsConfiguration.ShouldDeletePool)
{
Console.WriteLine("Deleting pool: {0}", topNWordsConfiguration.PoolId);
client.PoolOperations.DeletePool(topNWordsConfiguration.PoolId);
}
//Delete the job that we created
if (topNWordsConfiguration.ShouldDeleteJob)
{
Console.WriteLine("Deleting job: {0}", topNWordsConfiguration.JobId);
client.JobOperations.DeleteJob(topNWordsConfiguration.JobId);
}
//Delete the containers we created
if (topNWordsConfiguration.ShouldDeleteContainer)
{
DeleteContainers(accountSettings, stagingContainer);
}
}
}
}
/// <summary>
/// Populates Azure Storage with the required files, and
/// submits the job to the Azure Batch service.
/// </summary>
public async Task RunAsync()
{
Console.WriteLine("Running with the following settings: ");
Console.WriteLine("----------------------------------------");
Console.WriteLine(this.textSearchSettings.ToString());
Console.WriteLine(this.accountSettings.ToString());
CloudStorageAccount cloudStorageAccount = new CloudStorageAccount(
new StorageCredentials(
this.accountSettings.StorageAccountName,
this.accountSettings.StorageAccountKey),
this.accountSettings.StorageServiceUrl,
useHttps: true);
//Upload resources if required
Console.WriteLine($"Creating container {this.textSearchSettings.OutputBlobContainer} if it doesn't exist...");
var blobClient = cloudStorageAccount.CreateCloudBlobClient();
var outputContainer = blobClient.GetContainerReference(this.textSearchSettings.OutputBlobContainer);
await outputContainer.CreateIfNotExistsAsync();
if (this.textSearchSettings.ShouldUploadResources)
{
Console.WriteLine("Splitting file: {0} into {1} subfiles",
Constants.TextFilePath,
this.textSearchSettings.NumberOfMapperTasks);
//Split the text file into the correct number of files for consumption by the mapper tasks.
FileSplitter splitter = new FileSplitter();
List <string> mapperTaskFiles = await splitter.SplitAsync(
Constants.TextFilePath,
this.textSearchSettings.NumberOfMapperTasks);
List <string> files = Constants.RequiredExecutableFiles.Union(mapperTaskFiles).ToList();
await SampleHelpers.UploadResourcesAsync(
cloudStorageAccount,
this.textSearchSettings.InputBlobContainer,
files);
}
//Generate a SAS for the container.
string inputContainerSasUrl = SampleHelpers.ConstructContainerSas(
cloudStorageAccount,
this.textSearchSettings.InputBlobContainer,
permissions: WindowsAzure.Storage.Blob.SharedAccessBlobPermissions.Read);
string outputContainerSasUrl = SampleHelpers.ConstructContainerSas(
cloudStorageAccount,
this.textSearchSettings.OutputBlobContainer,
permissions: WindowsAzure.Storage.Blob.SharedAccessBlobPermissions.Read |
WindowsAzure.Storage.Blob.SharedAccessBlobPermissions.Write);
//Set up the Batch Service credentials used to authenticate with the Batch Service.
BatchSharedKeyCredentials credentials = new BatchSharedKeyCredentials(
this.accountSettings.BatchServiceUrl,
this.accountSettings.BatchAccountName,
this.accountSettings.BatchAccountKey);
using (BatchClient batchClient = BatchClient.Open(credentials))
{
//
// Construct the job properties in local memory before commiting them to the Batch Service.
//
//Allow enough compute nodes in the pool to run each mapper task
int numberOfPoolComputeNodes = this.textSearchSettings.NumberOfMapperTasks;
//Define the pool specification for the pool which the job will run on.
PoolSpecification poolSpecification = new PoolSpecification()
{
TargetDedicatedComputeNodes = numberOfPoolComputeNodes,
VirtualMachineSize = "standard_d1_v2",
//You can learn more about os families and versions at:
//http://azure.microsoft.com/documentation/articles/cloud-services-guestos-update-matrix
CloudServiceConfiguration = new CloudServiceConfiguration(osFamily: "5")
};
//Use the auto pool feature of the Batch Service to create a pool when the job is created.
//This creates a new pool for each job which is added.
AutoPoolSpecification autoPoolSpecification = new AutoPoolSpecification()
{
AutoPoolIdPrefix = "TextSearchPool",
KeepAlive = false,
PoolLifetimeOption = PoolLifetimeOption.Job,
PoolSpecification = poolSpecification
};
//Define the pool information for this job -- it will run on the pool defined by the auto pool specification above.
PoolInformation poolInformation = new PoolInformation()
{
AutoPoolSpecification = autoPoolSpecification
};
//Create the unbound job in local memory. An object which exists only in local memory (and not on the Batch Service) is "unbound".
string jobId = Environment.GetEnvironmentVariable("USERNAME") + DateTime.UtcNow.ToString("yyyyMMdd-HHmmss");
CloudJob unboundJob = batchClient.JobOperations.CreateJob(jobId, poolInformation);
unboundJob.UsesTaskDependencies = true;
try
{
//Commit the unbound job to the Batch Service.
Console.WriteLine($"Adding job: {unboundJob.Id} to the Batch Service.");
await unboundJob.CommitAsync(); //Issues a request to the Batch Service to add the job which was defined above.
// Add tasks to the job
var mapperTasks = CreateMapperTasks(inputContainerSasUrl, outputContainerSasUrl);
var reducerTask = CreateReducerTask(inputContainerSasUrl, outputContainerSasUrl, mapperTasks);
var tasksToAdd = Enumerable.Concat(mapperTasks, new[] { reducerTask });
//Submit the unbound task collection to the Batch Service.
//Use the AddTask method which takes a collection of CloudTasks for the best performance.
Console.WriteLine("Submitting {0} mapper tasks", this.textSearchSettings.NumberOfMapperTasks);
Console.WriteLine("Submitting 1 reducer task");
await batchClient.JobOperations.AddTaskAsync(jobId, tasksToAdd);
//An object which is backed by a corresponding Batch Service object is "bound."
CloudJob boundJob = await batchClient.JobOperations.GetJobAsync(jobId);
// Update the job now that we've added tasks so that when all of the tasks which we have added
// are complete, the job will automatically move to the completed state.
boundJob.OnAllTasksComplete = OnAllTasksComplete.TerminateJob;
boundJob.Commit();
boundJob.Refresh();
//
// Wait for the tasks to complete.
//
List <CloudTask> tasks = await batchClient.JobOperations.ListTasks(jobId).ToListAsync();
TimeSpan maxJobCompletionTimeout = TimeSpan.FromMinutes(30);
// Monitor the current tasks to see when they are done.
// Occasionally a task may get killed and requeued during an upgrade or hardware failure,
// Robustness against this was not added into the sample for
// simplicity, but should be added into any production code.
Console.WriteLine("Waiting for job's tasks to complete");
TaskStateMonitor taskStateMonitor = batchClient.Utilities.CreateTaskStateMonitor();
try
{
await taskStateMonitor.WhenAll(tasks, TaskState.Completed, maxJobCompletionTimeout);
}
finally
{
Console.WriteLine("Done waiting for all tasks to complete");
// Refresh the task list
tasks = await batchClient.JobOperations.ListTasks(jobId).ToListAsync();
//Check to ensure the job manager task exited successfully.
foreach (var task in tasks)
{
await Helpers.CheckForTaskSuccessAsync(task, dumpStandardOutOnTaskSuccess : false);
}
}
//
// Download and write out the reducer tasks output
//
string reducerText = await SampleHelpers.DownloadBlobTextAsync(cloudStorageAccount, this.textSearchSettings.OutputBlobContainer, Constants.ReducerTaskResultBlobName);
Console.WriteLine("Reducer reuslts:");
Console.WriteLine(reducerText);
}
finally
{
//Delete the job.
//This will delete the auto pool associated with the job as long as the pool
//keep alive property is set to false.
if (this.textSearchSettings.ShouldDeleteJob)
{
Console.WriteLine($"Deleting job {jobId}");
await batchClient.JobOperations.DeleteJobAsync(jobId);
}
if (this.textSearchSettings.ShouldDeleteContainers)
{
Console.WriteLine("Deleting containers");
var inputContainer = blobClient.GetContainerReference(this.textSearchSettings.InputBlobContainer);
await inputContainer.DeleteIfExistsAsync();
await outputContainer.DeleteIfExistsAsync();
}
}
}
}
