Free cookie consent management tool by TermsFeed Policy Generator

source: trunk/sources/HeuristicLab.Clients.Hive.Slave/3.3/Core.cs @ 7166

Last change on this file since 7166 was 7166, checked in by ascheibe, 12 years ago

#1672

  • don't serialize the results 2 times before uploading
  • made slave a little bit more robust
File size: 17.9 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 2002-2011 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
4 *
5 * This file is part of HeuristicLab.
6 *
7 * HeuristicLab is free software: you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation, either version 3 of the License, or
10 * (at your option) any later version.
11 *
12 * HeuristicLab is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
19 */
20#endregion
21
22using System;
23using System.Diagnostics;
24using System.ServiceModel;
25using System.Threading;
26using System.Threading.Tasks;
27using HeuristicLab.Clients.Hive.SlaveCore.Properties;
28using HeuristicLab.Clients.Hive.SlaveCore.ServiceContracts;
29using HeuristicLab.Common;
30using HeuristicLab.Core;
31using TS = System.Threading.Tasks;
32
33
34namespace HeuristicLab.Clients.Hive.SlaveCore {
35  /// <summary>
36  /// The core component of the Hive Slave.
37  /// Handles commands sent from the Hive Server and does all webservice calls for jobs.
38  /// </summary>
39  public class Core : MarshalByRefObject {
40    private static HeartbeatManager heartbeatManager;
41    public static HeartbeatManager HeartbeatManager {
42      get { return heartbeatManager; }
43    }
44
45    public EventLog ServiceEventLog { get; set; }
46
47    private Semaphore waitShutdownSem = new Semaphore(0, 1);
48    private bool abortRequested;
49    private ISlaveCommunication clientCom;
50    private ServiceHost slaveComm;
51    private WcfService wcfService;
52    private TaskManager taskManager;
53    private ConfigManager configManager;
54    private PluginManager pluginManager;
55
56    public Core() {
57      var log = new ThreadSafeLog(Settings.Default.MaxLogCount);
58      this.pluginManager = new PluginManager(WcfService.Instance, log);
59      this.taskManager = new TaskManager(pluginManager, log);
60      log.MessageAdded += new EventHandler<EventArgs<string>>(log_MessageAdded);
61
62      RegisterTaskManagerEvents();
63
64      this.configManager = new ConfigManager(taskManager);
65      ConfigManager.Instance = this.configManager;
66    }
67
68    /// <summary>
69    /// Main method for the client
70    /// </summary>
71    public void Start() {
72      abortRequested = false;
73      EventLogManager.ServiceEventLog = ServiceEventLog;
74
75      try {
76        //start the client communication service (pipe between slave and slave gui)
77        slaveComm = new ServiceHost(typeof(SlaveCommunicationService));
78        slaveComm.Open();
79        clientCom = SlaveClientCom.Instance.ClientCom;
80
81        // delete all left over task directories
82        pluginManager.CleanPluginTemp();
83        clientCom.LogMessage("Hive Slave started");
84
85        wcfService = WcfService.Instance;
86        RegisterServiceEvents();
87
88        StartHeartbeats(); // Start heartbeats thread       
89        DispatchMessageQueue(); // dispatch messages until abortRequested
90      }
91      catch (Exception ex) {
92        if (ServiceEventLog != null) {
93          EventLogManager.LogException(ex);
94        } else {
95          //try to log with clientCom. if this works the user sees at least a message,
96          //else an exception will be thrown anyways.
97          clientCom.LogMessage(string.Format("Uncaught exception: {0} {1} Core is going to shutdown.", ex.ToString(), Environment.NewLine));
98        }
99        ShutdownCore();
100      } finally {
101        DeregisterServiceEvents();
102        waitShutdownSem.Release();
103      }
104    }
105
106    private void StartHeartbeats() {
107      //Initialize the heartbeat     
108      if (heartbeatManager == null) {
109        heartbeatManager = new HeartbeatManager();
110        heartbeatManager.StartHeartbeat();
111      }
112    }
113
114    private void DispatchMessageQueue() {
115      MessageQueue queue = MessageQueue.GetInstance();
116      while (!abortRequested) {
117        MessageContainer container = queue.GetMessage();
118        DetermineAction(container);
119        if (!abortRequested) {
120          clientCom.StatusChanged(configManager.GetStatusForClientConsole());
121        }
122      }
123    }
124
125    private void RegisterServiceEvents() {
126      WcfService.Instance.Connected += new EventHandler(WcfService_Connected);
127      WcfService.Instance.ExceptionOccured += new EventHandler<EventArgs<Exception>>(WcfService_ExceptionOccured);
128    }
129
130    private void DeregisterServiceEvents() {
131      WcfService.Instance.Connected -= WcfService_Connected;
132      WcfService.Instance.ExceptionOccured -= WcfService_ExceptionOccured;
133    }
134
135    private void WcfService_ExceptionOccured(object sender, EventArgs<Exception> e) {
136      clientCom.LogMessage(string.Format("Connection to server interruped with exception: {0}", e.Value.Message));
137    }
138
139    private void WcfService_Connected(object sender, EventArgs e) {
140      clientCom.LogMessage("Connected successfully to Hive server");
141    }
142
143    /// <summary>
144    /// Reads and analyzes the Messages from the MessageQueue and starts corresponding actions
145    /// </summary>
146    /// <param name="container">The container, containing the message</param>
147    private void DetermineAction(MessageContainer container) {
148      clientCom.LogMessage(string.Format("Message: {0} for task: {1} ", container.Message.ToString(), container.TaskId));
149
150      switch (container.Message) {
151        case MessageContainer.MessageType.CalculateTask:
152          CalculateTaskAsync(container.TaskId);
153          break;
154        case MessageContainer.MessageType.AbortTask:
155          AbortTaskAsync(container.TaskId);
156          break;
157        case MessageContainer.MessageType.StopTask:
158          StopTaskAsync(container.TaskId);
159          break;
160        case MessageContainer.MessageType.PauseTask:
161          PauseTaskAsync(container.TaskId);
162          break;
163        case MessageContainer.MessageType.StopAll:
164          DoStopAll();
165          break;
166        case MessageContainer.MessageType.PauseAll:
167          DoPauseAll();
168          break;
169        case MessageContainer.MessageType.AbortAll:
170          DoAbortAll();
171          break;
172        case MessageContainer.MessageType.ShutdownSlave:
173          ShutdownCore();
174          break;
175        case MessageContainer.MessageType.Restart:
176          DoStartSlave();
177          break;
178        case MessageContainer.MessageType.Sleep:
179          Sleep();
180          break;
181        case MessageContainer.MessageType.SayHello:
182          wcfService.Connect(configManager.GetClientInfo());
183          break;
184        case MessageContainer.MessageType.NewHBInterval:
185          int interval = wcfService.GetNewHeartbeatInterval(ConfigManager.Instance.GetClientInfo().Id);
186          if (interval != -1) {
187            HeartbeatManager.Interval = TimeSpan.FromSeconds(interval);
188          }
189          break;
190      }
191    }
192
193    private void CalculateTaskAsync(Guid jobId) {
194      TS.Task.Factory.StartNew(HandleCalculateTask, jobId)
195      .ContinueWith((t) => {
196        SlaveStatusInfo.IncrementExceptionOccured();
197        clientCom.LogMessage(t.Exception.ToString());
198      }, TaskContinuationOptions.OnlyOnFaulted);
199    }
200
201    private void StopTaskAsync(Guid jobId) {
202      TS.Task.Factory.StartNew(HandleStopTask, jobId)
203       .ContinueWith((t) => {
204         SlaveStatusInfo.IncrementExceptionOccured();
205         clientCom.LogMessage(t.Exception.ToString());
206       }, TaskContinuationOptions.OnlyOnFaulted);
207    }
208
209    private void PauseTaskAsync(Guid jobId) {
210      TS.Task.Factory.StartNew(HandlePauseTask, jobId)
211       .ContinueWith((t) => {
212         SlaveStatusInfo.IncrementExceptionOccured();
213         clientCom.LogMessage(t.Exception.ToString());
214       }, TaskContinuationOptions.OnlyOnFaulted);
215    }
216
217    private void AbortTaskAsync(Guid jobId) {
218      TS.Task.Factory.StartNew(HandleAbortTask, jobId)
219       .ContinueWith((t) => {
220         SlaveStatusInfo.IncrementExceptionOccured();
221         clientCom.LogMessage(t.Exception.ToString());
222       }, TaskContinuationOptions.OnlyOnFaulted);
223    }
224
225    private void HandleCalculateTask(object taskIdObj) {
226      Guid taskId = (Guid)taskIdObj;
227      Task task = null;
228      int usedCores = 0;
229      try {
230        task = wcfService.GetTask(taskId);
231        if (task == null) throw new TaskNotFoundException(taskId);
232        if (ConfigManager.Instance.GetFreeCores() < task.CoresNeeded) throw new OutOfCoresException();
233        if (ConfigManager.Instance.GetFreeMemory() < task.MemoryNeeded) throw new OutOfMemoryException();
234        SlaveStatusInfo.IncrementUsedCores(task.CoresNeeded); usedCores = task.CoresNeeded;
235        TaskData taskData = wcfService.GetTaskData(taskId);
236        if (taskData == null) throw new TaskDataNotFoundException(taskId);
237        task = wcfService.UpdateJobState(taskId, TaskState.Calculating, null);
238        if (task == null) throw new TaskNotFoundException(taskId);
239        taskManager.StartTaskAsync(task, taskData);
240      }
241      catch (TaskNotFoundException) {
242        SlaveStatusInfo.DecrementUsedCores(usedCores);
243        throw;
244      }
245      catch (TaskDataNotFoundException) {
246        SlaveStatusInfo.DecrementUsedCores(usedCores);
247        throw;
248      }
249      catch (TaskAlreadyRunningException) {
250        SlaveStatusInfo.DecrementUsedCores(usedCores);
251        throw;
252      }
253      catch (OutOfCoresException) {
254        wcfService.UpdateJobState(taskId, TaskState.Waiting, "No more cores available");
255        throw;
256      }
257      catch (OutOfMemoryException) {
258        wcfService.UpdateJobState(taskId, TaskState.Waiting, "No more memory available");
259        throw;
260      }
261      catch (Exception e) {
262        SlaveStatusInfo.DecrementUsedCores(usedCores);
263        wcfService.UpdateJobState(taskId, TaskState.Waiting, e.ToString()); // unknown internal error - report and set waiting again
264        throw;
265      }
266    }
267
268    private void HandleStopTask(object taskIdObj) {
269      Guid taskId = (Guid)taskIdObj;
270      try {
271        Task task = wcfService.GetTask(taskId);
272        if (task == null) throw new TaskNotFoundException(taskId);
273        taskManager.StopTaskAsync(taskId);
274      }
275      catch (TaskNotFoundException) {
276        throw;
277      }
278      catch (TaskNotRunningException) {
279        throw;
280      }
281      catch (AppDomainNotCreatedException) {
282        throw;
283      }
284    }
285
286    private void HandlePauseTask(object taskIdObj) {
287      Guid taskId = (Guid)taskIdObj;
288      try {
289        Task task = wcfService.GetTask(taskId);
290        if (task == null) throw new TaskNotFoundException(taskId);
291        taskManager.PauseTaskAsync(taskId);
292      }
293      catch (TaskNotFoundException) {
294        throw;
295      }
296      catch (TaskNotRunningException) {
297        throw;
298      }
299      catch (AppDomainNotCreatedException) {
300        throw;
301      }
302    }
303
304    private void HandleAbortTask(object taskIdObj) {
305      Guid taskId = (Guid)taskIdObj;
306      try {
307        taskManager.AbortTask(taskId);
308      }
309      catch (TaskNotFoundException) {
310        throw;
311      }
312    }
313
314    #region TaskManager Events
315    private void RegisterTaskManagerEvents() {
316      this.taskManager.TaskStarted += new EventHandler<EventArgs<SlaveTask>>(taskManager_TaskStarted);
317      this.taskManager.TaskPaused += new EventHandler<EventArgs<SlaveTask, TaskData>>(taskManager_TaskPaused);
318      this.taskManager.TaskStopped += new EventHandler<EventArgs<SlaveTask, TaskData>>(taskManager_TaskStopped);
319      this.taskManager.TaskFailed += new EventHandler<EventArgs<Tuple<SlaveTask, TaskData, Exception>>>(taskManager_TaskFailed);
320      this.taskManager.ExceptionOccured += new EventHandler<EventArgs<SlaveTask, Exception>>(taskManager_ExceptionOccured);
321      this.taskManager.TaskAborted += new EventHandler<EventArgs<SlaveTask>>(taskManager_TaskAborted);
322    }
323
324    private void taskManager_TaskStarted(object sender, EventArgs<SlaveTask> e) {
325      // successfully started, everything is good
326    }
327
328    private void taskManager_TaskPaused(object sender, EventArgs<SlaveTask, TaskData> e) {
329      try {
330        SlaveStatusInfo.DecrementUsedCores(e.Value.CoresNeeded);
331        heartbeatManager.AwakeHeartBeatThread();
332        Task task = wcfService.GetTask(e.Value.TaskId);
333        if (task == null) throw new TaskNotFoundException(e.Value.TaskId);
334        task.ExecutionTime = e.Value.ExecutionTime;
335        TaskData taskData = e.Value2;
336        wcfService.UpdateTaskData(task, taskData, configManager.GetClientInfo().Id, TaskState.Paused);
337      }
338      catch (TaskNotFoundException ex) {
339        clientCom.LogMessage(ex.ToString());
340      }
341      catch (Exception ex) {
342        clientCom.LogMessage(ex.ToString());
343      }
344    }
345
346    private void taskManager_TaskStopped(object sender, EventArgs<SlaveTask, TaskData> e) {
347      try {
348        SlaveStatusInfo.DecrementUsedCores(e.Value.CoresNeeded);
349        heartbeatManager.AwakeHeartBeatThread();
350        Task task = wcfService.GetTask(e.Value.TaskId);
351        if (task == null) throw new TaskNotFoundException(e.Value.TaskId);
352        task.ExecutionTime = e.Value.ExecutionTime;
353        TaskData taskData = e.Value2;
354        wcfService.UpdateTaskData(task, taskData, configManager.GetClientInfo().Id, TaskState.Finished);
355      }
356      catch (TaskNotFoundException ex) {
357        clientCom.LogMessage(ex.ToString());
358      }
359      catch (Exception ex) {
360        clientCom.LogMessage(ex.ToString());
361      }
362    }
363
364    private void taskManager_TaskFailed(object sender, EventArgs<Tuple<SlaveTask, TaskData, Exception>> e) {
365      try {
366        SlaveStatusInfo.DecrementUsedCores(e.Value.Item1.CoresNeeded);
367        heartbeatManager.AwakeHeartBeatThread();
368        SlaveTask slaveTask = e.Value.Item1;
369        TaskData taskData = e.Value.Item2;
370        Exception exception = e.Value.Item3;
371
372        Task task = wcfService.GetTask(slaveTask.TaskId);
373        if (task == null) throw new TaskNotFoundException(slaveTask.TaskId);
374        task.ExecutionTime = slaveTask.ExecutionTime;
375        if (taskData != null) {
376          wcfService.UpdateTaskData(task, taskData, configManager.GetClientInfo().Id, TaskState.Failed, exception.ToString());
377        } else {
378          wcfService.UpdateJobState(task.Id, TaskState.Failed, exception.ToString());
379        }
380        clientCom.LogMessage(exception.Message);
381      }
382      catch (TaskNotFoundException ex) {
383        SlaveStatusInfo.IncrementExceptionOccured();
384        clientCom.LogMessage(ex.ToString());
385      }
386      catch (Exception ex) {
387        SlaveStatusInfo.IncrementExceptionOccured();
388        clientCom.LogMessage(ex.ToString());
389      }
390    }
391
392    private void taskManager_ExceptionOccured(object sender, EventArgs<SlaveTask, Exception> e) {
393      SlaveStatusInfo.DecrementUsedCores(e.Value.CoresNeeded);
394      SlaveStatusInfo.IncrementExceptionOccured();
395      heartbeatManager.AwakeHeartBeatThread();
396      clientCom.LogMessage(string.Format("Exception occured for task {0}: {1}", e.Value.TaskId, e.Value2.ToString()));
397      wcfService.UpdateJobState(e.Value.TaskId, TaskState.Waiting, e.Value2.ToString());
398    }
399
400    private void taskManager_TaskAborted(object sender, EventArgs<SlaveTask> e) {
401      SlaveStatusInfo.DecrementUsedCores(e.Value.CoresNeeded);
402    }
403    #endregion
404
405    #region Log Events
406    private void log_MessageAdded(object sender, EventArgs<string> e) {
407      try {
408        clientCom.LogMessage(e.Value.Split('\t')[1]);
409      }
410      catch { }
411    }
412    #endregion
413
414    /// <summary>
415    /// aborts all running jobs, no results are sent back
416    /// </summary>
417    private void DoAbortAll() {
418      clientCom.LogMessage("Aborting all jobs.");
419      foreach (Guid taskId in taskManager.TaskIds) {
420        AbortTaskAsync(taskId);
421      }
422    }
423
424    /// <summary>
425    /// wait for jobs to finish, then pause client
426    /// </summary>
427    private void DoPauseAll() {
428      clientCom.LogMessage("Pausing all jobs.");
429      foreach (Guid taskId in taskManager.TaskIds) {
430        PauseTaskAsync(taskId);
431      }
432    }
433
434    /// <summary>
435    /// pause slave immediately
436    /// </summary>
437    private void DoStopAll() {
438      clientCom.LogMessage("Stopping all jobs.");
439      foreach (Guid taskId in taskManager.TaskIds) {
440        StopTaskAsync(taskId);
441      }
442    }
443
444    #region Slave Lifecycle Methods
445    /// <summary>
446    /// completly shudown slave
447    /// </summary>
448    public void Shutdown() {
449      MessageContainer mc = new MessageContainer(MessageContainer.MessageType.ShutdownSlave);
450      MessageQueue.GetInstance().AddMessage(mc);
451      waitShutdownSem.WaitOne();
452    }
453
454    /// <summary>
455    /// complete shutdown, should be called before the the application is exited
456    /// </summary>
457    private void ShutdownCore() {
458      clientCom.LogMessage("Shutdown signal received");
459      clientCom.LogMessage("Stopping heartbeat");
460      heartbeatManager.StopHeartBeat();
461      abortRequested = true;
462
463      DoAbortAll();
464
465      clientCom.LogMessage("Logging out");
466      WcfService.Instance.Disconnect();
467      clientCom.Shutdown();
468      SlaveClientCom.Close();
469
470      if (slaveComm.State != CommunicationState.Closed)
471        slaveComm.Close();
472    }
473
474    /// <summary>
475    /// reinitializes everything and continues operation,
476    /// can be called after Sleep()
477    /// </summary> 
478    private void DoStartSlave() {
479      clientCom.LogMessage("Restart received");
480      configManager.Asleep = false;
481    }
482
483    /// <summary>
484    /// stop slave, except for client gui communication,
485    /// primarily used by gui if core is running as windows service
486    /// </summary>   
487    private void Sleep() {
488      clientCom.LogMessage("Sleep received - not accepting any new jobs");
489      configManager.Asleep = true;
490      DoPauseAll();
491    }
492    #endregion
493  }
494}
Note: See TracBrowser for help on using the repository browser.