Concurrency and django scaling horizontal - django

Question

Hello I would like to ask you for advice for an application, I currently use Django cookie cutter for the project, I use as a library, Django MPTT in this model it must respect a certain hierarchy and that some objects do not find themselves in the same place as another to do this I need to isolate the objects to then perform my calculation and create my new objects. First of all since it could take a while I used Cellery to make my additions to the database I tried several concurrent accesses and I never had a rollback, I thought instead of using Cellery since I know how to scale the docker into several Django, I would integrate this creation into my view and use the #atomic decorator. The problem is that when we create objects at the same time on each different server (django1,django2,django3) I have a lot of rollback while with celeriac I have no problem. I don't understand why competition is more difficult in Django than in celery? Do you have any advice for me? For a better management of different Django servers at the same time, thank you in advance example :
Work in celery norollback :
#celery_app.task()
def create_user_in_matrix(user_id):
.....
new_human = HumanFaction.objects.select_for_update().filter(id=int(user_id))
with transaction.atomic():
new_human = new_human.first()
alderol = HumanFaction.objects.get(user_id=alderol_user_id)
print('selection utilisateur', new_human)
level = 1
level_max = HumanFaction.objects.aggregate(Max('level'))['level__max']
if level_max == 0:
level_max = 1
print('level maximum', level_max)
while level <= level_max and user_whole:
users = HumanFaction.objects.select_for_update().filter(level=level)
HumanFaction.objects.create(user=new_human, parent=aledol)
.....
Operates every third time in a django view during a colision test frequent rollback (django,1,2,3)
#transaction.atomic
def human_group(request):
......
new_human = HumanFaction.objects.select_for_update().filter(id=int(user_id))
with transaction.atomic():
new_human = new_human.first()
alderol = HumanFaction.objects.get(user_id=alderol_user_id)
print('selection utilisateur', new_human)
level = 1
level_max = HumanFaction.objects.aggregate(Max('level'))['level__max']
if level_max == 0:
level_max = 1
print('level maximum', level_max)
while level <= level_max and user_whole:
users = HumanFaction.objects.select_for_update().filter(level=level)
HumanFaction.objects.create(user=new_human, parent=aledol)
.....