Changeset 15630 for branches/HiveProjectManagement/HeuristicLab.Services.Hive/3.3/Manager/HeartbeatManager.cs

Timestamp:

01/18/18 15:08:25 (7 years ago)

Author:

jzenisek

Message:

#2839

• updated Heartbeat processing (regarding: checking against AssignedJobResources and handling of the updated Job deletion routine)
• updated Job deletion routine(still in progress at GenerateStatistics)

File:

• branches/HiveProjectManagement/HeuristicLab.Services.Hive/3.3/Manager/HeartbeatManager.cs (modified) (1 diff)

branches/HiveProjectManagement/HeuristicLab.Services.Hive/3.3/Manager/HeartbeatManager.cs

@@ -141 +141 @@
     /// </summary>
     private IEnumerable<MessageContainer> UpdateTasks(IPersistenceManager pm, Heartbeat heartbeat, bool isAllowedToCalculate) {
+      var taskDao = pm.TaskDao;
+      var jobDao = pm.JobDao;
+      var assignedJobResourceDao = pm.AssignedJobResourceDao;
+      var actions = new List<MessageContainer>();
+      if (heartbeat.JobProgress == null || !heartbeat.JobProgress.Any())
+        return actions;
+
+      var jobIdsWithStatisticsPending = jobDao.GetJobIdsByState(DA.JobState.StatisticsPending).ToList();
+
+      // select all tasks and statelogs with one query
+      var taskIds = heartbeat.JobProgress.Select(x => x.Key).ToList();
+      var taskInfos = pm.UseTransaction(() =>
+        (from task in taskDao.GetAll()
+          where taskIds.Contains(task.TaskId)
+          let lastStateLog = task.StateLogs.OrderByDescending(x => x.DateTime).FirstOrDefault()
+          select new {
+            TaskId = task.TaskId,
+            JobId = task.JobId,
+            State = task.State,
+            Command = task.Command,
+            SlaveId = lastStateLog != null ? lastStateLog.SlaveId : default(Guid)
+          }).ToList()
+      );
+
+      // process the jobProgresses
+      foreach (var jobProgress in heartbeat.JobProgress) {
+        var progress = jobProgress;
+        var curTask = taskInfos.SingleOrDefault(x => x.TaskId == progress.Key);
+        if (curTask == null) {
+          actions.Add(new MessageContainer(MessageContainer.MessageType.AbortTask, progress.Key));
+          LogFactory.GetLogger(this.GetType().Namespace).Log("Task on slave " + heartbeat.SlaveId + " does not exist in DB: " + jobProgress.Key);
+        } else if (jobIdsWithStatisticsPending.Contains(curTask.JobId)) {
+          // parenting job of current task has been requested for deletion (indicated by job state "Statistics Pending")
+          // update task execution time
+          pm.UseTransaction(() => {
+            taskDao.UpdateExecutionTime(curTask.TaskId, progress.Value.TotalMilliseconds);
+          });
+          actions.Add(new MessageContainer(MessageContainer.MessageType.AbortTask, curTask.TaskId));
+          LogFactory.GetLogger(this.GetType().Namespace).Log("Abort task " + curTask.TaskId + " on slave " + heartbeat.SlaveId + ". The parenting job " + curTask.JobId + " was requested to be deleted.");
+        } else if (curTask.SlaveId == Guid.Empty || curTask.SlaveId != heartbeat.SlaveId) {
+          // assigned slave does not match heartbeat
+          actions.Add(new MessageContainer(MessageContainer.MessageType.AbortTask, curTask.TaskId));
+          LogFactory.GetLogger(this.GetType().Namespace).Log("The slave " + heartbeat.SlaveId + " is not supposed to calculate task: " + curTask.TaskId);
+        } else if (!isAllowedToCalculate) {
+          actions.Add(new MessageContainer(MessageContainer.MessageType.PauseTask, curTask.TaskId));
+          LogFactory.GetLogger(this.GetType().Namespace).Log("The slave " + heartbeat.SlaveId + " is not allowed to calculate any tasks tue to a downtime. The task is paused.");
+        } else if (assignedJobResourceDao.CheckJobGrantedForResource(curTask.JobId, heartbeat.SlaveId)) { 
+          // assigned resources ids of task do not match with slaveId (and parent resourceGroupIds); this might happen when slave is moved to different group
+          actions.Add(new MessageContainer(MessageContainer.MessageType.PauseTask, curTask.TaskId));
+          LogFactory.GetLogger(this.GetType().Namespace).Log("The slave " + heartbeat.SlaveId + " is not granted to calculate task: " + curTask.TaskId);
+        } else {
+          // update task execution time
+          pm.UseTransaction(() => {
+            taskDao.UpdateExecutionTime(curTask.TaskId, progress.Value.TotalMilliseconds);
+          });
+          switch (curTask.Command) {
+            case DA.Command.Stop:
+              actions.Add(new MessageContainer(MessageContainer.MessageType.StopTask, curTask.TaskId));
+              break;
+            case DA.Command.Pause:
+              actions.Add(new MessageContainer(MessageContainer.MessageType.PauseTask, curTask.TaskId));
+              break;
+            case DA.Command.Abort:
+              actions.Add(new MessageContainer(MessageContainer.MessageType.AbortTask, curTask.TaskId));
+              break;
+          }
+        }
+        
+      } 
+      return actions;
+    }
+
+    /// <summary>
+    /// Update the progress of each task
+    /// Checks if all the task sent by heartbeat are supposed to be calculated by this slave
+    /// </summary>
+    private IEnumerable<MessageContainer> UpdateTasks_Old(IPersistenceManager pm, Heartbeat heartbeat, bool isAllowedToCalculate) {
       var taskDao = pm.TaskDao;
       var assignedResourceDao = pm.AssignedTaskResourceDao;