Changeset 5786 for branches/HeuristicLab.Hive-3.4/sources/HeuristicLab.Clients.Hive/3.4/ExperimentManager/HiveExperimentClient.cs

Timestamp:

03/22/11 11:36:53 (13 years ago)

Author:

cneumuel

Message:

#1233

• implemented correct numbering of BatchRuns
• improvements in ExperimentManager
• fixed bug in server (jobs were scheduled multiple times)
• added exception handling for task in slave
• improved timeout handling of jobs (LifecycleManager)

File:

• branches/HeuristicLab.Hive-3.4/sources/HeuristicLab.Clients.Hive/3.4/ExperimentManager/HiveExperimentClient.cs (modified) (4 diffs)

branches/HeuristicLab.Hive-3.4/sources/HeuristicLab.Clients.Hive/3.4/ExperimentManager/HiveExperimentClient.cs

@@ -22 +22 @@
 using System;
 using System.Collections.Generic;
+using System.Configuration;
+using System.IO;
 using System.Linq;
 using System.Threading;
 using HeuristicLab.Clients.Hive.Jobs;
+using HeuristicLab.Collections;
 using HeuristicLab.Common;
 using HeuristicLab.Core;
 using HeuristicLab.Optimization;
+using HeuristicLab.PluginInfrastructure;
 
 namespace HeuristicLab.Clients.Hive {
-  using System.Configuration;
-  using System.IO;
-  using HeuristicLab.Collections;
-  using HeuristicLab.PluginInfrastructure;
-
   /// <summary>
   /// An experiment which contains multiple batch runs of algorithms.
@@ -378 +377 @@
 
     #region HiveJob Events
-    void HiveJob_JobStateChanged(object sender, EventArgs e) {
+    private void HiveJob_JobStateChanged(object sender, EventArgs e) {
       if (HiveJob != null) {
         rootJobId = HiveJob.Job.Id;
@@ -520 +519 @@
         HiveJob hj = hiveJob.GetHiveJobByJobId(lightweightJob.Id);
         if (hj != null) {
+          DateTime lastJobDataUpdate = hj.Job.LastJobDataUpdate;
           hj.UpdateFromLightweightJob(lightweightJob);
-          if ((hj.Job.State == JobState.Aborted ||
-               hj.Job.State == JobState.Failed ||
-               hj.Job.State == JobState.Finished) &&
-              !hj.IsFinishedOptimizerDownloaded) {
+
+          // lastJobDataUpdate equals DateTime.MinValue right after it was uploaded. When the first results are polled, this value is updated
+          if (lastJobDataUpdate != DateTime.MinValue && lastJobDataUpdate < hj.Job.LastJobDataUpdate) {
             LogMessage(hj.Job.Id, "Downloading optimizer for job");
             OptimizerJob optimizerJob = LoadOptimizerJob(hj.Job.Id);
             if (optimizerJob == null) {
               // something bad happened to this job. set to finished to allow the rest beeing downloaded
-              hj.IsFinishedOptimizerDownloaded = true;
+              //hj.IsFinishedOptimizerDownloaded = true;
             } else {
-              if (lightweightJob.ParentJobId.HasValue) {
-                HiveJob parentHiveJob = HiveJob.GetHiveJobByJobId(lightweightJob.ParentJobId.Value);
-                parentHiveJob.UpdateChildOptimizer(optimizerJob, hj.Job.Id);
+              // if the job is paused, download but don't integrate into parent optimizer (to avoid Prepare)
+              if (hj.Job.State == JobState.Paused) {
+                
               } else {
-                this.HiveJob.IsFinishedOptimizerDownloaded = true;
+                if (lightweightJob.ParentJobId.HasValue) {
+                  HiveJob parentHiveJob = HiveJob.GetHiveJobByJobId(lightweightJob.ParentJobId.Value);
+                  parentHiveJob.UpdateChildOptimizer(optimizerJob, hj.Job.Id);
+                } else {
+                  //this.HiveJob.IsFinishedOptimizerDownloaded = true;
+                }
               }
             }
@@ -550 +554 @@
 
     private bool AllJobsFinished() {
-      return HiveJob.GetAllHiveJobs().All(hj => hj.IsFinishedOptimizerDownloaded);
+      //return HiveJob.GetAllHiveJobs().All(hj => hj.IsFinishedOptimizerDownloaded);
+      return HiveJob.GetAllHiveJobs().All(j => j.Job.State == JobState.Finished
+                                            || j.Job.State == JobState.Aborted
+                                            || j.Job.State == JobState.Failed);
     }