Backport of scheduler: ensure dup alloc names are fixed before plan submit. into release/1.5.x

.changelog/18873.txt

@@ -0,0 +1,3 @@
+```release-note:bug
+scheduler: Ensure duplicate allocation IDs are tracked and fixed when performing job updates
+```

nomad/structs/structs.go

@@ -10204,13 +10204,19 @@ func (a *Allocation) JobNamespacedID() NamespacedID {
 // Index returns the index of the allocation. If the allocation is from a task
 // group with count greater than 1, there will be multiple allocations for it.
 func (a *Allocation) Index() uint {
-	l := len(a.Name)
-	prefix := len(a.JobID) + len(a.TaskGroup) + 2
+	return AllocIndexFromName(a.Name, a.JobID, a.TaskGroup)
+}
+
+// AllocIndexFromName returns the index of an allocation given its name, the
+// jobID and the task group name.
+func AllocIndexFromName(allocName, jobID, taskGroup string) uint {
+	l := len(allocName)
+	prefix := len(jobID) + len(taskGroup) + 2
 	if l <= 3 || l <= prefix {
 		return uint(0)
 	}
 
-	strNum := a.Name[prefix : len(a.Name)-1]
+	strNum := allocName[prefix : len(allocName)-1]
 	num, _ := strconv.Atoi(strNum)
 	return uint(num)
 }

scheduler/generic_sched.go

@@ -463,7 +463,7 @@ func (s *GenericScheduler) computeJobAllocs() error {
 		s.queuedAllocs[p.placeTaskGroup.Name] += 1
 		destructive = append(destructive, p)
 	}
-	return s.computePlacements(destructive, place)
+	return s.computePlacements(destructive, place, results.taskGroupAllocNameIndexes)
 }
 
 // downgradedJobForPlacement returns the job appropriate for non-canary placement replacement
@@ -505,7 +505,8 @@ func (s *GenericScheduler) downgradedJobForPlacement(p placementResult) (string,
 
 // computePlacements computes placements for allocations. It is given the set of
 // destructive updates to place and the set of new placements to place.
-func (s *GenericScheduler) computePlacements(destructive, place []placementResult) error {
+func (s *GenericScheduler) computePlacements(destructive, place []placementResult, nameIndex map[string]*allocNameIndex) error {
+
 	// Get the base nodes
 	byDC, err := s.setNodes(s.job)
 	if err != nil {
@@ -528,6 +529,12 @@ func (s *GenericScheduler) computePlacements(destructive, place []placementResul
 			// Get the task group
 			tg := missing.TaskGroup()
 
+			// This is populated from the reconciler via the compute results,
+			// therefore we cannot have an allocation belonging to a task group
+			// that has not generated and been through allocation name index
+			// tracking.
+			taskGroupNameIndex := nameIndex[tg.Name]
+
 			var downgradedJob *structs.Job
 
 			if missing.DowngradeNonCanary() {
@@ -624,12 +631,34 @@ func (s *GenericScheduler) computePlacements(destructive, place []placementResul
 					resources.Shared.Ports = option.AllocResources.Ports
 				}
 
+				// Pull the allocation name as a new variables, so we can alter
+				// this as needed without making changes to the original
+				// object.
+				newAllocName := missing.Name()
+
+				// Identify the index from the name, so we can check this
+				// against the allocation name index tracking for duplicates.
+				allocIndex := structs.AllocIndexFromName(newAllocName, s.job.ID, tg.Name)
+
+				// If the allocation index is a duplicate, we cannot simply
+				// create a new allocation with the same name. We need to
+				// generate a new index and use this. The log message is useful
+				// for debugging and development, but could be removed in a
+				// future version of Nomad.
+				if taskGroupNameIndex.IsDuplicate(allocIndex) {
+					oldAllocName := newAllocName
+					newAllocName = taskGroupNameIndex.Next(1)[0]
+					taskGroupNameIndex.UnsetIndex(allocIndex)
+					s.logger.Debug("duplicate alloc index found and changed",
+						"old_alloc_name", oldAllocName, "new_alloc_name", newAllocName)
+				}
+
 				// Create an allocation for this
 				alloc := &structs.Allocation{
 					ID:                 uuid.Generate(),
 					Namespace:          s.job.Namespace,
 					EvalID:             s.eval.ID,
-					Name:               missing.Name(),
+					Name:               newAllocName,
 					JobID:              s.job.ID,
 					TaskGroup:          tg.Name,
 					Metrics:            s.ctx.Metrics(),