moved logs connected to unknown undeployable to Devel

AliceO2Group · Oct 28, 2024 · f03e42c · f03e42c
1 parent ddb067b
commit f03e42c
Show file tree

Hide file tree

Showing 3 changed files with 15 additions and 9 deletions.
diff --git a/core/environment/transition_deploy.go b/core/environment/transition_deploy.go
@@ -245,6 +245,7 @@ func (t DeployTransition) do(env *Environment) (err error) {
 								WithField("partition", env.Id().String()).
 								WithField("timeout", deploymentTimeout).
 								WithField("detector", detector).
+								WithField("level", infologger.IL_Devel).
 								Error("role failed to deploy within timeout")
 							undeployableTaskRoles = append(undeployableTaskRoles, role.GetPath())
 						}
@@ -275,6 +276,7 @@ func (t DeployTransition) do(env *Environment) (err error) {
 							WithField("partition", env.Id().String()).
 							WithField("timeout", deploymentTimeout).
 							WithField("detector", detector).
+							WithField("level", infologger.IL_Devel).
 							Error("role failed to deploy because of timeout")
 						undeployableTaskRoles = append(undeployableTaskRoles, role.GetPath())
 					} else if roleStatus != task.ACTIVE {
@@ -288,6 +290,7 @@ func (t DeployTransition) do(env *Environment) (err error) {
 							WithField("partition", env.Id().String()).
 							WithField("timeout", deploymentTimeout).
 							WithField("detector", detector).
+							WithField("level", infologger.IL_Devel).
 							Error("role failed to deploy because of timeout")
 						inactiveTaskRoles = append(inactiveTaskRoles, role.GetPath())
 					}

diff --git a/core/task/manager.go b/core/task/manager.go
@@ -596,16 +596,21 @@ func (m *Manager) acquireTasks(envId uid.ID, taskDescriptors Descriptors) (err e
 				break DEPLOYMENT_ATTEMPTS_LOOP
 			}
 
-			log.WithField("partition", envId).Errorf("Deployment failed %d/%d attempts. Check messages in IL to figure out why. Retrying...", attemptCount+1, MAX_ATTEMPTS_PER_DEPLOY_REQUEST)
+			log.WithField("partition", envId).
+				WithField("level", infologger.IL_Devel).
+				Errorf("Deployment failed %d/%d attempts. Check messages in IL to figure out why. Retrying...", attemptCount+1, MAX_ATTEMPTS_PER_DEPLOY_REQUEST)
 			time.Sleep(time.Second * SLEEP_LENGTH_BETWEEN_PER_DEPLOY_REQUESTS)
 		}
 	}
 
-	logDescriptors("critical task deployment impossible: ", logWithId.Errorf, undeployableCriticalDescriptors)
-	logDescriptors("critical task deployment failure: ", logWithId.Errorf, undeployedCriticalDescriptors)
+	{
+		logWithIdDev := logWithId.WithField("level", infologger.IL_Devel)
+		logDescriptors("critical task deployment impossible: ", logWithIdDev.Errorf, undeployableCriticalDescriptors)
+		logDescriptors("critical task deployment failure: ", logWithIdDev.Errorf, undeployedCriticalDescriptors)
 
-	logDescriptors("non-critical task deployment failure: ", logWithId.Warningf, undeployedNonCriticalDescriptors)
-	logDescriptors("non-critical task deployment impossible: ", logWithId.Warningf, undeployableNonCriticalDescriptors)
+		logDescriptors("non-critical task deployment failure: ", logWithIdDev.Warningf, undeployedNonCriticalDescriptors)
+		logDescriptors("non-critical task deployment impossible: ", logWithIdDev.Warningf, undeployableNonCriticalDescriptors)
+	}
 
 	// After retries notify environment about failed critical tasks
 	for _, desc := range undeployableDescriptors {
@@ -617,11 +622,8 @@ func (m *Manager) acquireTasks(envId uid.ID, taskDescriptors Descriptors) (err e
 	m.deployMu.Unlock()
 
 	if !deploymentSuccess {
-		// While all the required roles are running, for some reason we
-		// can't lock some of them, so we must roll back and keep them
-		// unlocked in the roster.
 		var deployedTaskIds []string
-		for taskPtr, _ := range deployedTasks {
+		for taskPtr := range deployedTasks {
 			taskPtr.SetParent(nil)
 			deployedTaskIds = append(deployedTaskIds, taskPtr.taskId)
 		}

diff --git a/core/task/scheduler.go b/core/task/scheduler.go
@@ -592,6 +592,7 @@ func (state *schedulerState) resourceOffers(fidStore store.Singleton) events.Han
 						descriptorsUndeployable = append(descriptorsUndeployable, descriptor)
 						descriptorsStillToDeploy = append(descriptorsStillToDeploy[:i], descriptorsStillToDeploy[i+1:]...)
 						log.WithField("partition", envId.String()).
+							WithField("level", infologger.IL_Devel).
 							WithField("descriptor", descriptor.TaskClassName).
 							Errorf("no resource offer for required host %s, deployment will be aborted", requiredMachineId)
 					}