/// <summary>
/// Sums the given TaskResult's points processed data with the totals.
/// </summary>
/// <param name="TaskResult"></param>
private void AddDataFromTaskResult(KMeansTaskResult taskResult)
{
TotalNumPointsChanged += taskResult.NumPointsChanged;
foreach (KeyValuePair <Guid, PointsProcessedData> pointsProcessedDataForCentroid in taskResult.PointsProcessedDataByCentroid)
{
AddPointsProcessedDataForCentroid(pointsProcessedDataForCentroid.Key, pointsProcessedDataForCentroid.Value);
}
}
public void SavePointsProcessedDataByCentroidTest()
{
KMeansTaskData task = new KMeansTaskData(Guid.NewGuid(), Guid.NewGuid(), 1, null, 2, 3, 10, 0, null, DateTime.Now, DateTime.Now, 0, null);
KMeansTaskResult target = new KMeansTaskResult(task);
target.PointsProcessedDataByCentroid[Guid.NewGuid()] = new PointsProcessedData()
{
NumPointsProcessed = 100,
PartialPointSum = new Point(10, -10)
};
target.PointsProcessedDataByCentroid[Guid.NewGuid()] = new PointsProcessedData()
{
NumPointsProcessed = 100,
PartialPointSum = new Point(10, -10)
};
target.SavePointsProcessedDataByCentroid();
foreach (KeyValuePair<Guid, PointsProcessedData> pair in target.PointsProcessedDataByCentroid)
{
Assert.IsTrue(target.PointsProcessedDataByCentroidList.Contains(pair));
}
}
/// <summary>
/// Handles a worker's TaskResult from a running k-means job. Adds up the partial sums from the TaskResult.
/// </summary>
/// <param name="message"></param>
/// <returns>False if the given taskData result has already been counted, true otherwise.</returns>
public bool ProcessWorkerResponse(KMeansTaskResult taskResult, IEnumerable <Worker> workers)
{
// Make sure we're actually still waiting for a result for this taskData
// If not, this might be a duplicate queue message
if (!TaskResultMatchesRunningTask(taskResult))
{
return(true);
}
AzureHelper.LogPerformance(() =>
{
KMeansTask task = TaskResultWithTaskID(taskResult.TaskID);
task.Running = false; // The task has returned a response, which means that it has stopped running
// Add the worker's updated points blocks
if (taskResult.PointsBlockListBlob != null)
{
using (Stream stream = AzureHelper.GetBlob(taskResult.PointsBlockListBlob).OpenRead())
{
BinaryFormatter bf = new BinaryFormatter();
List <string> pointsBlockList = bf.Deserialize(stream) as List <string>;
pointsBlockIDs.AddRange(pointsBlockList);
}
}
// Copy out and integrate the data from the worker response
AddDataFromTaskResult(taskResult);
}, jobData.JobID.ToString(), methodName: "ProcessWorkerResponse", iterationCount: IterationCount, points: Points.Uri.ToString(), centroids: Centroids.Uri.ToString(), machineID: MachineID);
// If this is the last worker to return, this iteration is done and we should start the next one
if (NoMoreRunningTasks())
{
NextIteration(workers);
}
return(true);
}
/// <summary>
/// Handles a worker response as part of a running k-means job. Looks up the appropriate job and passes the worker's response to it.
/// </summary>
/// <param name="message">The worker response. Must be of type KMeansTaskResult.</param>
private bool ProcessWorkerResponse(KMeansTaskResult taskResult)
{
taskResult.RestorePointsProcessedDataByCentroid();
// Make sure the job belongs to this server
if (!jobs.ContainsKey(taskResult.JobID))
return false;
System.Diagnostics.Trace.TraceInformation("[ServerRole] ProcessWorkerResponse(jobID={0}, taskID={1}, iterationCount={2})", taskResult.JobID, taskResult.TaskID, jobs[taskResult.JobID].IterationCount);
return jobs[taskResult.JobID].ProcessWorkerResponse(taskResult, workers.Values);
}
public KMeansTaskProcessor(KMeansTaskData task)
{
this.task = task;
this.TaskResult = new KMeansTaskResult(task);
}
private bool TaskResultMatchesRunningTask(KMeansTaskResult taskResult)
{
KMeansTask task = TaskResultWithTaskID(taskResult.TaskID);
return task != null && task.Running;
}
/// <summary>
/// Sums the given TaskResult's points processed data with the totals.
/// </summary>
/// <param name="TaskResult"></param>
private void AddDataFromTaskResult(KMeansTaskResult taskResult)
{
TotalNumPointsChanged += taskResult.NumPointsChanged;
foreach (KeyValuePair<Guid, PointsProcessedData> pointsProcessedDataForCentroid in taskResult.PointsProcessedDataByCentroid)
{
AddPointsProcessedDataForCentroid(pointsProcessedDataForCentroid.Key, pointsProcessedDataForCentroid.Value);
}
}
/// <summary>
/// Handles a worker's TaskResult from a running k-means job. Adds up the partial sums from the TaskResult.
/// </summary>
/// <param name="message"></param>
/// <returns>False if the given taskData result has already been counted, true otherwise.</returns>
public bool ProcessWorkerResponse(KMeansTaskResult taskResult, IEnumerable<Worker> workers)
{
// Make sure we're actually still waiting for a result for this taskData
// If not, this might be a duplicate queue message
if (!TaskResultMatchesRunningTask(taskResult))
return true;
AzureHelper.LogPerformance(() =>
{
KMeansTask task = TaskResultWithTaskID(taskResult.TaskID);
task.Running = false; // The task has returned a response, which means that it has stopped running
// Add the worker's updated points blocks
if (taskResult.PointsBlockListBlob != null)
{
using (Stream stream = AzureHelper.GetBlob(taskResult.PointsBlockListBlob).OpenRead())
{
BinaryFormatter bf = new BinaryFormatter();
List<string> pointsBlockList = bf.Deserialize(stream) as List<string>;
pointsBlockIDs.AddRange(pointsBlockList);
}
}
// Copy out and integrate the data from the worker response
AddDataFromTaskResult(taskResult);
}, jobData.JobID.ToString(), methodName: "ProcessWorkerResponse", iterationCount: IterationCount, points: Points.Uri.ToString(), centroids: Centroids.Uri.ToString(), machineID: MachineID);
// If this is the last worker to return, this iteration is done and we should start the next one
if (NoMoreRunningTasks())
{
NextIteration(workers);
}
return true;
}
public void ProcessWorkerResponseTest()
{
KMeansJobData jobData = new KMeansJobData(Guid.NewGuid(), 4, null, 2, 10, DateTime.Now);
KMeansJob_Accessor target = new KMeansJob_Accessor(jobData, "server");
target.InitializeStorage();
// Upload a block with an arbitrary ClusterPoint, so we can verify it gets copied
ClusterPoint arbitraryPoint = new ClusterPoint(1, 2, Guid.NewGuid());
List<string> blockList;
using (ObjectCachedBlockWriter<ClusterPoint> pointPartitionWriteStream = new ObjectCachedBlockWriter<ClusterPoint>(target.Points, point => point.ToByteArray(), ClusterPoint.Size,
Environment.GetEnvironmentVariable("TEMP") + @"\" + Guid.NewGuid().ToString()))
{
pointPartitionWriteStream.Write(arbitraryPoint);
pointPartitionWriteStream.FlushBlock();
blockList = pointPartitionWriteStream.BlockList;
}
KMeansTaskData taskData = new KMeansTaskData(jobData, Guid.NewGuid(), 0, 1, target.Centroids.Uri, DateTime.Now, 0, null);
target.tasks.Clear();
target.tasks.Add(new KMeansTask(taskData));
KMeansTaskResult taskResult = new KMeansTaskResult(taskData);
CloudBlob pointsBlockListBlob = AzureHelper.CreateBlob(jobData.JobID.ToString(), Guid.NewGuid().ToString());
using (Stream stream = pointsBlockListBlob.OpenWrite())
{
BinaryFormatter bf = new BinaryFormatter();
bf.Serialize(stream, blockList);
}
taskResult.PointsBlockListBlob = pointsBlockListBlob.Uri;
taskResult.NumPointsChanged = 2;
Guid centroidID = Guid.NewGuid();
taskResult.PointsProcessedDataByCentroid = new Dictionary<Guid, PointsProcessedData> {
{ centroidID, new PointsProcessedData() {
NumPointsProcessed = 2,
PartialPointSum = new Point(1, 2)
}
}
};
target.ProcessWorkerResponse(taskResult, new List<Worker>());
// Verify that the first ClusterPoint in Points is indeed equal to arbitraryPoint
using (ObjectStreamReader<ClusterPoint> pointsStream = new ObjectStreamReader<ClusterPoint>(target.Points, ClusterPoint.FromByteArray, ClusterPoint.Size))
{
ClusterPoint point = pointsStream.First();
Assert.AreEqual(arbitraryPoint.X, point.X);
Assert.AreEqual(arbitraryPoint.Y, point.Y);
Assert.AreEqual(arbitraryPoint.CentroidID, point.CentroidID);
}
}
public void MultiIterationJobTest()
{
KMeansJobData jobData = new KMeansJobData(Guid.NewGuid(), 4, null, 2, 2, DateTime.Now);
List<Worker> workers = new List<Worker> {
new Worker("a", "", 1),
new Worker("b", "", 1)
};
KMeansJob_Accessor job = new KMeansJob_Accessor(jobData, "server");
// First iteration
job.InitializeStorage();
job.EnqueueTasks(workers);
for (int i = 0; i < jobData.MaxIterationCount; i++)
{
CheckWorkerRequests(job,
(from task in job.tasks
where task.Running
select task.TaskData),
workers.Count, job.Points);
// Create the worker results and send them to the job
List<KMeansTaskResult> results = new List<KMeansTaskResult>();
foreach (var task in job.tasks)
{
var taskResult = new KMeansTaskResult(task.TaskData);
taskResult.NumPointsChanged = 1;
results.Add(taskResult);
}
foreach (var result in results)
{
job.ProcessWorkerResponse(result, workers);
}
}
}
private bool TaskResultMatchesRunningTask(KMeansTaskResult taskResult)
{
KMeansTask task = TaskResultWithTaskID(taskResult.TaskID);
return(task != null && task.Running);
}
