1 | using System;
|
---|
2 | using System.Collections.Generic;
|
---|
3 | using System.Linq;
|
---|
4 | using System.Text;
|
---|
5 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
6 | using HeuristicLab.Core;
|
---|
7 | using HeuristicLab.Common;
|
---|
8 | using HeuristicLab.Hive.Contracts.Interfaces;
|
---|
9 | using HeuristicLab.Clients.Common;
|
---|
10 | using HeuristicLab.Hive.ExperimentManager;
|
---|
11 | using HeuristicLab.Hive.Contracts.BusinessObjects;
|
---|
12 | using HeuristicLab.PluginInfrastructure;
|
---|
13 | using HeuristicLab.Hive.Contracts.ResponseObjects;
|
---|
14 | using System.Threading;
|
---|
15 | using HeuristicLab.Random;
|
---|
16 | using System.Threading.Tasks;
|
---|
17 |
|
---|
18 | namespace HeuristicLab.HiveEngine {
|
---|
19 | /// <summary>
|
---|
20 | /// Represents an engine that executes operations which can be executed in parallel on the hive
|
---|
21 | /// </summary>
|
---|
22 | [StorableClass]
|
---|
23 | [Item("Hive Engine", "Engine for parallel execution on the hive. You need enable `Parallel` for at least one operator in your operator graph to have all childoperations parallelized. Also those childoperations must not have sideeffects on a higher scope.")]
|
---|
24 | public class HiveEngine : Engine {
|
---|
25 | private Semaphore maxConcurrentConnections = new Semaphore(4, 4); // avoid too many connections
|
---|
26 | private Semaphore maxSerializedJobsInMemory = new Semaphore(4, 4); // avoid memory problems
|
---|
27 | private CancellationToken cancellationToken;
|
---|
28 |
|
---|
29 | [Storable]
|
---|
30 | private IOperator currentOperator;
|
---|
31 |
|
---|
32 | [Storable]
|
---|
33 | public string ResourceIds { get; set; }
|
---|
34 |
|
---|
35 | [Storable]
|
---|
36 | private int priority;
|
---|
37 | public int Priority {
|
---|
38 | get { return priority; }
|
---|
39 | set { priority = value; }
|
---|
40 | }
|
---|
41 |
|
---|
42 | [Storable]
|
---|
43 | private TimeSpan executionTimeOnHive;
|
---|
44 | public TimeSpan ExecutionTimeOnHive {
|
---|
45 | get { return executionTimeOnHive; }
|
---|
46 | set {
|
---|
47 | if (value != executionTimeOnHive) {
|
---|
48 | executionTimeOnHive = value;
|
---|
49 | OnExecutionTimeOnHiveChanged();
|
---|
50 | }
|
---|
51 | }
|
---|
52 | }
|
---|
53 |
|
---|
54 | #region constructors and cloning
|
---|
55 | public HiveEngine() {
|
---|
56 | ResourceIds = "HEAL";
|
---|
57 | }
|
---|
58 |
|
---|
59 | [StorableConstructor]
|
---|
60 | protected HiveEngine(bool deserializing) : base(deserializing) { }
|
---|
61 | protected HiveEngine(HiveEngine original, Cloner cloner)
|
---|
62 | : base(original, cloner) {
|
---|
63 | this.ResourceIds = original.ResourceIds;
|
---|
64 | this.currentOperator = cloner.Clone(original.currentOperator);
|
---|
65 | this.priority = original.priority;
|
---|
66 | this.executionTimeOnHive = original.executionTimeOnHive;
|
---|
67 | }
|
---|
68 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
69 | return new HiveEngine(this, cloner);
|
---|
70 | }
|
---|
71 | #endregion
|
---|
72 |
|
---|
73 | #region Events
|
---|
74 | protected override void OnPrepared() {
|
---|
75 | base.OnPrepared();
|
---|
76 | this.ExecutionTimeOnHive = TimeSpan.Zero;
|
---|
77 | }
|
---|
78 |
|
---|
79 | public event EventHandler ExecutionTimeOnHiveChanged;
|
---|
80 | protected virtual void OnExecutionTimeOnHiveChanged() {
|
---|
81 | var handler = ExecutionTimeOnHiveChanged;
|
---|
82 | if (handler != null) handler(this, EventArgs.Empty);
|
---|
83 | }
|
---|
84 | #endregion
|
---|
85 |
|
---|
86 | protected override void Run(CancellationToken cancellationToken) {
|
---|
87 | this.cancellationToken = cancellationToken;
|
---|
88 | Run(ExecutionStack);
|
---|
89 | }
|
---|
90 |
|
---|
91 | private void Run(object state) {
|
---|
92 | Stack<IOperation> executionStack = (Stack<IOperation>)state;
|
---|
93 | IOperation next;
|
---|
94 | OperationCollection coll;
|
---|
95 | IAtomicOperation operation;
|
---|
96 | TaskScheduler.UnobservedTaskException += new EventHandler<UnobservedTaskExceptionEventArgs>(TaskScheduler_UnobservedTaskException);
|
---|
97 |
|
---|
98 | while (ExecutionStack.Count > 0) {
|
---|
99 | cancellationToken.ThrowIfCancellationRequested();
|
---|
100 |
|
---|
101 | next = ExecutionStack.Pop();
|
---|
102 | if (next is OperationCollection) {
|
---|
103 | coll = (OperationCollection)next;
|
---|
104 | if (coll.Parallel) {
|
---|
105 | // clone the parent scope here and reuse it for each operation. otherwise for each job the whole scope-tree first needs to be copied and then cleaned, which causes a lot of work for the Garbage Collector
|
---|
106 | IScope parentScopeClone = (IScope)((IAtomicOperation)coll.First()).Scope.Parent.Clone();
|
---|
107 | parentScopeClone.SubScopes.Clear();
|
---|
108 | parentScopeClone.ClearParentScopes();
|
---|
109 |
|
---|
110 | OperationJob[] jobs = new OperationJob[coll.Count];
|
---|
111 | for (int i = 0; i < coll.Count; i++) {
|
---|
112 | jobs[i] = new OperationJob(coll[i]);
|
---|
113 | }
|
---|
114 |
|
---|
115 | IScope[] scopes = ExecuteOnHive(jobs, parentScopeClone, cancellationToken);
|
---|
116 |
|
---|
117 | for (int i = 0; i < coll.Count; i++) {
|
---|
118 | if (coll[i] is IAtomicOperation) {
|
---|
119 | ExchangeScope(scopes[i], ((IAtomicOperation)coll[i]).Scope);
|
---|
120 | } else if (coll[i] is OperationCollection) {
|
---|
121 | // todo ??
|
---|
122 | }
|
---|
123 | }
|
---|
124 | } else {
|
---|
125 | for (int i = coll.Count - 1; i >= 0; i--)
|
---|
126 | if (coll[i] != null) executionStack.Push(coll[i]);
|
---|
127 | }
|
---|
128 | } else if (next is IAtomicOperation) {
|
---|
129 | operation = (IAtomicOperation)next;
|
---|
130 | try {
|
---|
131 | next = operation.Operator.Execute((IExecutionContext)operation, cancellationToken);
|
---|
132 | }
|
---|
133 | catch (Exception ex) {
|
---|
134 | ExecutionStack.Push(operation);
|
---|
135 | if (ex is OperationCanceledException) throw ex;
|
---|
136 | else throw new OperatorExecutionException(operation.Operator, ex);
|
---|
137 | }
|
---|
138 | if (next != null) ExecutionStack.Push(next);
|
---|
139 |
|
---|
140 | if (operation.Operator.Breakpoint) {
|
---|
141 | LogMessage(string.Format("Breakpoint: {0}", operation.Operator.Name != string.Empty ? operation.Operator.Name : operation.Operator.ItemName));
|
---|
142 | Pause();
|
---|
143 | }
|
---|
144 | }
|
---|
145 | }
|
---|
146 | }
|
---|
147 |
|
---|
148 | private void TaskScheduler_UnobservedTaskException(object sender, UnobservedTaskExceptionEventArgs e) {
|
---|
149 | e.SetObserved(); // avoid crash of process
|
---|
150 | }
|
---|
151 |
|
---|
152 | private IRandom FindRandomParameter(IExecutionContext ec) {
|
---|
153 | try {
|
---|
154 | if (ec == null)
|
---|
155 | return null;
|
---|
156 |
|
---|
157 | foreach (var p in ec.Parameters) {
|
---|
158 | if (p.Name == "Random" && p is IValueParameter)
|
---|
159 | return ((IValueParameter)p).Value as IRandom;
|
---|
160 | }
|
---|
161 | return FindRandomParameter(ec.Parent);
|
---|
162 | }
|
---|
163 | catch { return null; }
|
---|
164 | }
|
---|
165 |
|
---|
166 | private static void ReIntegrateScope(IAtomicOperation source, IAtomicOperation target) {
|
---|
167 | ExchangeScope(source.Scope, target.Scope);
|
---|
168 | }
|
---|
169 |
|
---|
170 | private static void ExchangeScope(IScope source, IScope target) {
|
---|
171 | target.Variables.Clear();
|
---|
172 | target.Variables.AddRange(source.Variables);
|
---|
173 | target.SubScopes.Clear();
|
---|
174 | target.SubScopes.AddRange(source.SubScopes);
|
---|
175 | // TODO: validate if parent scopes match - otherwise source is invalid
|
---|
176 | }
|
---|
177 |
|
---|
178 | /// <summary>
|
---|
179 | /// This method blocks until all jobs are finished
|
---|
180 | /// TODO: Cancelation needs to be refined; all tasks currently stay in Semaphore.WaitOne after cancelation
|
---|
181 | /// </summary>
|
---|
182 | /// <param name="jobs"></param>
|
---|
183 | private IScope[] ExecuteOnHive(OperationJob[] jobs, IScope parentScopeClone, CancellationToken cancellationToken) {
|
---|
184 | LogMessage(string.Format("Executing {0} operations on the hive.", jobs.Length));
|
---|
185 | IScope[] scopes = new Scope[jobs.Length];
|
---|
186 | object locker = new object();
|
---|
187 | IDictionary<Guid, int> jobIndices = new Dictionary<Guid, int>();
|
---|
188 |
|
---|
189 | try {
|
---|
190 | List<Guid> remainingJobIds = new List<Guid>();
|
---|
191 | JobResultList results;
|
---|
192 | var pluginsNeeded = ApplicationManager.Manager.Plugins.Select(x => new HivePluginInfoDto { Name = x.Name, Version = x.Version }).ToList();
|
---|
193 | int finishedCount = 0;
|
---|
194 | int uploadCount = 0;
|
---|
195 |
|
---|
196 | // create upload-tasks
|
---|
197 | var uploadTasks = new List<Task<JobDto>>();
|
---|
198 | for (int i = 0; i < jobs.Length; i++) {
|
---|
199 | var job = jobs[i];
|
---|
200 |
|
---|
201 | // shuffle random variable to avoid the same random sequence in each operation; todo: does not yet work (it cannot find the random variable)
|
---|
202 | IRandom random = FindRandomParameter(job.Operation as IExecutionContext);
|
---|
203 | if (random != null)
|
---|
204 | random.Reset(random.Next());
|
---|
205 |
|
---|
206 | uploadTasks.Add(Task.Factory.StartNew<JobDto>((keyValuePairObj) => {
|
---|
207 | return UploadJob(pluginsNeeded, keyValuePairObj, parentScopeClone, cancellationToken);
|
---|
208 | }, new KeyValuePair<int, OperationJob>(i, job), cancellationToken));
|
---|
209 | }
|
---|
210 |
|
---|
211 | Task processUploadedJobsTask = new Task(() => {
|
---|
212 | // process finished upload-tasks
|
---|
213 | int uploadTasksCount = uploadTasks.Count;
|
---|
214 | for (int i = 0; i < uploadTasksCount; i++) {
|
---|
215 | cancellationToken.ThrowIfCancellationRequested();
|
---|
216 |
|
---|
217 | var uploadTasksArray = uploadTasks.ToArray();
|
---|
218 | var task = uploadTasksArray[Task.WaitAny(uploadTasksArray)];
|
---|
219 | if (task.Status == TaskStatus.Faulted) {
|
---|
220 | LogException(task.Exception);
|
---|
221 | throw task.Exception;
|
---|
222 | }
|
---|
223 |
|
---|
224 | int key = ((KeyValuePair<int, OperationJob>)task.AsyncState).Key;
|
---|
225 | JobDto jobDto = task.Result;
|
---|
226 | lock (locker) {
|
---|
227 | uploadCount++;
|
---|
228 | jobIndices.Add(jobDto.Id, key);
|
---|
229 | remainingJobIds.Add(jobDto.Id);
|
---|
230 | }
|
---|
231 | jobs[key] = null; // relax memory
|
---|
232 | LogMessage(string.Format("Uploaded job #{0}", key + 1, jobDto.Id));
|
---|
233 | uploadTasks.Remove(task);
|
---|
234 | }
|
---|
235 | }, cancellationToken, TaskCreationOptions.PreferFairness);
|
---|
236 | processUploadedJobsTask.Start();
|
---|
237 |
|
---|
238 | // poll job-statuses and create tasks for those which are finished
|
---|
239 | var downloadTasks = new List<Task<OperationJob>>();
|
---|
240 | var executionTimes = new List<TimeSpan>();
|
---|
241 | var executionTimeOnHiveBefore = executionTimeOnHive;
|
---|
242 | while (processUploadedJobsTask.Status != TaskStatus.RanToCompletion || remainingJobIds.Count > 0) {
|
---|
243 | cancellationToken.ThrowIfCancellationRequested();
|
---|
244 |
|
---|
245 | Thread.Sleep(10000);
|
---|
246 | try {
|
---|
247 | using (Disposable<IClientFacade> service = ServiceLocator.Instance.ClientFacadePool.GetService()) {
|
---|
248 | results = service.Obj.GetJobResults(remainingJobIds).Obj;
|
---|
249 | }
|
---|
250 | var jobsFinished = results.Where(j => j.State == JobState.Finished || j.State == JobState.Failed || j.State == JobState.Aborted);
|
---|
251 | finishedCount += jobsFinished.Count();
|
---|
252 | if (jobsFinished.Count() > 0) LogMessage(string.Format("Finished: {0}/{1}", finishedCount, jobs.Length));
|
---|
253 | ExecutionTimeOnHive = executionTimeOnHiveBefore + executionTimes.Sum() + results.Select(x => x.ExecutionTime).Sum();
|
---|
254 |
|
---|
255 | foreach (var result in jobsFinished) {
|
---|
256 | if (result.State == JobState.Finished) {
|
---|
257 | downloadTasks.Add(Task.Factory.StartNew<OperationJob>((jobIdObj) => {
|
---|
258 | return DownloadJob(jobIndices, jobIdObj, cancellationToken);
|
---|
259 | }, result.Id, cancellationToken));
|
---|
260 | } else if (result.State == JobState.Aborted) {
|
---|
261 | LogMessage(string.Format("Job #{0} aborted (id: {1})", jobIndices[result.Id] + 1, result.Id));
|
---|
262 | } else if (result.State == JobState.Failed) {
|
---|
263 | LogMessage(string.Format("Job #{0} failed (id: {1}): {2}", jobIndices[result.Id] + 1, result.Id, result.Exception));
|
---|
264 | }
|
---|
265 | remainingJobIds.Remove(result.Id);
|
---|
266 | executionTimes.Add(result.ExecutionTime);
|
---|
267 | }
|
---|
268 | }
|
---|
269 | catch (Exception e) {
|
---|
270 | LogException(e);
|
---|
271 | }
|
---|
272 | }
|
---|
273 |
|
---|
274 | // process finished download-tasks
|
---|
275 | int downloadTasksCount = downloadTasks.Count;
|
---|
276 | for (int i = 0; i < downloadTasksCount; i++) {
|
---|
277 | cancellationToken.ThrowIfCancellationRequested();
|
---|
278 |
|
---|
279 | var downloadTasksArray = downloadTasks.ToArray();
|
---|
280 | var task = downloadTasksArray[Task.WaitAny(downloadTasksArray)];
|
---|
281 | var jobId = (Guid)task.AsyncState;
|
---|
282 | if (task.Status == TaskStatus.Faulted) {
|
---|
283 | LogException(task.Exception);
|
---|
284 | throw task.Exception;
|
---|
285 | }
|
---|
286 | scopes[jobIndices[(Guid)task.AsyncState]] = ((IAtomicOperation)task.Result.Operation).Scope;
|
---|
287 | downloadTasks.Remove(task);
|
---|
288 | }
|
---|
289 |
|
---|
290 | LogMessage(string.Format("All jobs finished (TotalExecutionTime: {0}).", executionTimes.Sum()));
|
---|
291 | DeleteJobs(jobIndices);
|
---|
292 |
|
---|
293 | return scopes;
|
---|
294 | }
|
---|
295 | catch (OperationCanceledException e) {
|
---|
296 | lock (locker) {
|
---|
297 | if (jobIndices != null) DeleteJobs(jobIndices);
|
---|
298 | }
|
---|
299 | throw e;
|
---|
300 | }
|
---|
301 | catch (Exception e) {
|
---|
302 | lock (locker) {
|
---|
303 | if (jobIndices != null) DeleteJobs(jobIndices);
|
---|
304 | }
|
---|
305 | LogException(e);
|
---|
306 | throw e;
|
---|
307 | }
|
---|
308 | }
|
---|
309 |
|
---|
310 | private void DeleteJobs(IDictionary<Guid, int> jobIndices) {
|
---|
311 | if (jobIndices.Count > 0) {
|
---|
312 | using (Disposable<IClientFacade> service = ServiceLocator.Instance.ClientFacadePool.GetService()) {
|
---|
313 | foreach (Guid jobId in jobIndices.Keys) {
|
---|
314 | service.Obj.DeleteJob(jobId);
|
---|
315 | }
|
---|
316 | LogMessage(string.Format("Deleted {0} jobs on hive.", jobIndices.Count));
|
---|
317 | jobIndices.Clear();
|
---|
318 | }
|
---|
319 | }
|
---|
320 | }
|
---|
321 |
|
---|
322 | private static object locker = new object();
|
---|
323 | private JobDto UploadJob(List<HivePluginInfoDto> pluginsNeeded, object keyValuePairObj, IScope parentScopeClone, CancellationToken cancellationToken) {
|
---|
324 | var keyValuePair = (KeyValuePair<int, OperationJob>)keyValuePairObj;
|
---|
325 | var groups = ResourceIds.Split(';');
|
---|
326 | ResponseObject<JobDto> response = null;
|
---|
327 | try {
|
---|
328 | maxSerializedJobsInMemory.WaitOne();
|
---|
329 | SerializedJob serializedJob = null;
|
---|
330 | while (serializedJob == null) { // repeat until success; rare race-conditions occur at serializations (enumeration was changed-exceptions); maybe this is because all the parent-scopes and execution-contexts at some point contain the hiveengine and the Log in here
|
---|
331 | cancellationToken.ThrowIfCancellationRequested();
|
---|
332 | try {
|
---|
333 | lock (Log) {
|
---|
334 | serializedJob = new SerializedJob();
|
---|
335 | }
|
---|
336 | }
|
---|
337 | catch (Exception e) {
|
---|
338 | LogException(e);
|
---|
339 | }
|
---|
340 | }
|
---|
341 | // clone operation and remove unnecessary scopes; don't do this earlier to avoid memory problems
|
---|
342 | lock (locker) {
|
---|
343 | ((IAtomicOperation)keyValuePair.Value.Operation).Scope.Parent = parentScopeClone;
|
---|
344 | keyValuePair.Value.Operation = (IOperation)keyValuePair.Value.Operation.Clone();
|
---|
345 | if (keyValuePair.Value.Operation is IAtomicOperation)
|
---|
346 | ((IAtomicOperation)keyValuePair.Value.Operation).Scope.ClearParentScopes();
|
---|
347 | serializedJob.SerializedJobData = SerializedJob.Serialize(keyValuePair.Value);
|
---|
348 | }
|
---|
349 | serializedJob.JobInfo = new JobDto();
|
---|
350 | serializedJob.JobInfo.State = JobState.Offline;
|
---|
351 | serializedJob.JobInfo.CoresNeeded = 1;
|
---|
352 | serializedJob.JobInfo.PluginsNeeded = pluginsNeeded;
|
---|
353 | serializedJob.JobInfo.Priority = priority;
|
---|
354 | try {
|
---|
355 | maxConcurrentConnections.WaitOne();
|
---|
356 | while (response == null) { // repeat until success
|
---|
357 | cancellationToken.ThrowIfCancellationRequested();
|
---|
358 | try {
|
---|
359 | using (Disposable<IClientFacade> service = ServiceLocator.Instance.StreamedClientFacadePool.GetService()) {
|
---|
360 | response = service.Obj.AddJobWithGroupStrings(serializedJob, groups);
|
---|
361 | serializedJob = null;
|
---|
362 | }
|
---|
363 | }
|
---|
364 | catch (Exception e) {
|
---|
365 | LogException(e);
|
---|
366 | }
|
---|
367 | }
|
---|
368 | }
|
---|
369 | finally {
|
---|
370 | maxConcurrentConnections.Release();
|
---|
371 | }
|
---|
372 | }
|
---|
373 | finally {
|
---|
374 | maxSerializedJobsInMemory.Release();
|
---|
375 | }
|
---|
376 | return response.Obj;
|
---|
377 | }
|
---|
378 |
|
---|
379 | private OperationJob DownloadJob(IDictionary<Guid, int> jobIndices, object jobIdObj, CancellationToken cancellationToken) {
|
---|
380 | Guid jobId = (Guid)jobIdObj;
|
---|
381 | SerializedJob serializedJob = null;
|
---|
382 | OperationJob operationJob = null;
|
---|
383 | try {
|
---|
384 | maxSerializedJobsInMemory.WaitOne();
|
---|
385 | maxConcurrentConnections.WaitOne();
|
---|
386 | while (serializedJob == null) { // repeat until success
|
---|
387 | cancellationToken.ThrowIfCancellationRequested();
|
---|
388 | try {
|
---|
389 | using (Disposable<IClientFacade> service = ServiceLocator.Instance.StreamedClientFacadePool.GetService()) {
|
---|
390 | serializedJob = service.Obj.GetLastSerializedResult(jobId).Obj;
|
---|
391 | }
|
---|
392 | }
|
---|
393 | catch (Exception e) {
|
---|
394 | LogException(e);
|
---|
395 | }
|
---|
396 | }
|
---|
397 | operationJob = SerializedJob.Deserialize<OperationJob>(serializedJob.SerializedJobData);
|
---|
398 | serializedJob = null;
|
---|
399 | LogMessage(string.Format("Downloaded job #{0}", jobIndices[jobId] + 1, jobId));
|
---|
400 | }
|
---|
401 | finally {
|
---|
402 | maxConcurrentConnections.Release();
|
---|
403 | maxSerializedJobsInMemory.Release();
|
---|
404 | }
|
---|
405 | return operationJob;
|
---|
406 | }
|
---|
407 |
|
---|
408 | /// <summary>
|
---|
409 | /// Threadsafe message logging
|
---|
410 | /// </summary>
|
---|
411 | private void LogMessage(string message) {
|
---|
412 | lock (Log) {
|
---|
413 | Log.LogMessage(message);
|
---|
414 | }
|
---|
415 | }
|
---|
416 |
|
---|
417 | /// <summary>
|
---|
418 | /// Threadsafe exception logging
|
---|
419 | /// </summary>
|
---|
420 | private void LogException(Exception exception) {
|
---|
421 | lock (Log) {
|
---|
422 | Log.LogException(exception);
|
---|
423 | }
|
---|
424 | }
|
---|
425 | }
|
---|
426 |
|
---|
427 | public static class ScopeExtensions {
|
---|
428 | public static void ClearParentScopes(this IScope scope) {
|
---|
429 | scope.ClearParentScopes(null);
|
---|
430 | }
|
---|
431 |
|
---|
432 | public static void ClearParentScopes(this IScope scope, IScope childScope) {
|
---|
433 | if (childScope != null) {
|
---|
434 | scope.SubScopes.Clear();
|
---|
435 | scope.SubScopes.Add(childScope);
|
---|
436 | }
|
---|
437 | if (scope.Parent != null)
|
---|
438 | scope.Parent.ClearParentScopes(scope);
|
---|
439 | }
|
---|
440 | }
|
---|
441 |
|
---|
442 | public static class EnumerableExtensions {
|
---|
443 | public static TimeSpan Sum(this IEnumerable<TimeSpan> times) {
|
---|
444 | return TimeSpan.FromMilliseconds(times.Select(e => e.TotalMilliseconds).Sum());
|
---|
445 | }
|
---|
446 | }
|
---|
447 | }
|
---|