Changeset 6444 for branches/HeuristicLab.Hive-3.4/sources/HeuristicLab.Services.Hive/3.4

Timestamp:

06/19/11 23:21:21 (13 years ago)

Author:

cneumuel

Message:

#1233

• stability improvements for HiveExperiment and HiveEngine
• parallelized upload of jobs
• enabled cancellation of job upload
• reduced the amount of double-assignment of jobs by an additional check in HeartbeatManager
• tried to tackle the amount of deadlocks by automatically rerunning transactions
• some fixes

File:

• branches/HeuristicLab.Hive-3.4/sources/HeuristicLab.Services.Hive/3.4/HeartbeatManager.cs (modified) (2 diffs)

branches/HeuristicLab.Hive-3.4/sources/HeuristicLab.Services.Hive/3.4/HeartbeatManager.cs

@@ -62 +62 @@
           if (availableJobs.Count() > 0) {
             var job = availableJobs.First();
-            actions.Add(new MessageContainer(MessageContainer.MessageType.CalculateJob, job.Id));
-            AssignJob(slave, job);
+            if (AssignJob(slave, job))
+              actions.Add(new MessageContainer(MessageContainer.MessageType.CalculateJob, job.Id));
           }
         }
@@ -70 +70 @@
     }
 
-    private void AssignJob(Slave slave, Job job) {
+    // returns true if assignment was successful
+    private bool AssignJob(Slave slave, Job job) {
+      // load job again and check if it is still available (this is an attempt to reduce the race condition which causes multiple heartbeats to get the same job assigned)
+      if (dao.GetJob(job.Id).State != JobState.Waiting) return false;
+
       job = dao.UpdateJobState(job.Id, JobState.Transferring, slave.Id, null, null);
-      dao.UpdateSlave(slave);
 
       // from now on the job has some time to send the next heartbeat (ApplicationConstants.TransferringJobHeartbeatTimeout)
       job.LastHeartbeat = DateTime.Now;
       dao.UpdateJob(job);
+      return true;
     }