fixup! [core] deployment retry properly retries only failed tasks

Author: Michal Tichák

core/task/manager.go

@@ -497,8 +497,8 @@ func (m *Manager) acquireTasks(envId uid.ID, taskDescriptors Descriptors) (err e
 
 	// we are retrying deployment multiple times in case of failure and we don't want
 	// to rerun tasks already running
-	tasksToRunOneRound := make(Descriptors, len(tasksToRun))
-	copy(tasksToRunOneRound, tasksToRun)
+	tasksToRunThisAttempt := make(Descriptors, len(tasksToRun))
+	copy(tasksToRunThisAttempt, tasksToRun)
 
 	allDeployedTasks := make(DeploymentMap)
 	if len(tasksToRun) > 0 {
@@ -532,26 +532,26 @@ func (m *Manager) acquireTasks(envId uid.ID, taskDescriptors Descriptors) (err e
 
 			outcomeCh := make(chan ResourceOffersOutcome)
 			m.tasksToDeploy <- &ResourceOffersDeploymentRequest{
-				tasksToDeploy: tasksToRunOneRound,
+				tasksToDeploy: tasksToRunThisAttempt,
 				envId:         envId,
 				outcomeCh:     outcomeCh,
 			} // buffered channel, does not block
 
 			log.WithField("partition", envId).
-				Debugf("scheduler has been sent request to deploy %d tasks", len(tasksToRunOneRound))
+				Debugf("scheduler has been sent request to deploy %d tasks", len(tasksToRunThisAttempt))
 
 			timeReviveOffers := time.Now()
 			timeDeployMu := time.Now()
 			m.reviveOffersTrg <- struct{}{} // signal scheduler to revive offers
 			<-m.reviveOffersTrg             // we only continue when it's done
 			utils.TimeTrack(timeReviveOffers, "acquireTasks: revive offers",
-				log.WithField("tasksToRunOneRound", len(tasksToRunOneRound)).
+				log.WithField("tasksToRunThisAttempt", len(tasksToRunThisAttempt)).
 					WithField("partition", envId))
 
 			roOutcome := <-outcomeCh // blocks until a verdict from resourceOffers comes in
 
 			utils.TimeTrack(timeDeployMu, "acquireTasks: deployment critical section",
-				log.WithField("tasksToRunOneRound", len(tasksToRunOneRound)).
+				log.WithField("tasksToRunThisAttempt", len(tasksToRunThisAttempt)).
 					WithField("partition", envId))
 
 			deployedThisRound := roOutcome.deployed
@@ -561,21 +561,18 @@ func (m *Manager) acquireTasks(envId uid.ID, taskDescriptors Descriptors) (err e
 			logWithId.WithField("tasks", deployedThisRound).
 				Debugf("resourceOffers is done, %d new tasks running", len(deployedThisRound))
 
-			for deployedKey, deployedTask := range deployedThisRound {
-				allDeployedTasks[deployedKey] = deployedTask
+			for deployedTask, deployedDescriptor := range deployedThisRound {
+				allDeployedTasks[deployedTask] = deployedDescriptor
+				// add deployed tasks to roster, so updates can be distributed properly
+				m.roster.append(deployedTask)
 			}
 
-			// add deployed tasks to roster, so updates can be distributed properly
-			for taskPtr := range deployedThisRound {
-				m.roster.append(taskPtr)
-			}
-
-			if len(deployedThisRound) != len(tasksToRunOneRound) {
+			if len(deployedThisRound) != len(tasksToRunThisAttempt) {
 				// ↑ Not all roles could be deployed. If some were critical,
 				//   we cannot proceed with running this environment. Either way,
 				//   we keep the roles running since they might be useful in the future.
 				logWithId.WithField("level", infologger.IL_Devel).
-					Errorf("environment deployment failure: %d tasks requested for deployment, but %d deployed", len(tasksToRunOneRound), len(deployedThisRound))
+					Errorf("environment deployment failure: %d tasks requested for deployment, but %d deployed", len(tasksToRunThisAttempt), len(deployedThisRound))
 
 				for _, desc := range undeployedDescriptors {
 					if desc.TaskRole.GetTaskTraits().Critical == true {
@@ -609,9 +606,9 @@ func (m *Manager) acquireTasks(envId uid.ID, taskDescriptors Descriptors) (err e
 				}
 				break DEPLOYMENT_ATTEMPTS_LOOP
 			}
-			tasksToRunOneRound = make(Descriptors, 0, len(undeployableDescriptors)+len(undeployedDescriptors))
-			tasksToRunOneRound = append(tasksToRunOneRound, undeployedDescriptors...)
-			tasksToRunOneRound = append(tasksToRunOneRound, undeployableDescriptors...)
+			tasksToRunThisAttempt = make(Descriptors, 0, len(undeployableDescriptors)+len(undeployedDescriptors))
+			tasksToRunThisAttempt = append(tasksToRunThisAttempt, undeployedDescriptors...)
+			tasksToRunThisAttempt = append(tasksToRunThisAttempt, undeployableDescriptors...)
 
 			log.WithField("partition", envId).
 				WithField("level", infologger.IL_Devel).