Issue with ManyToMany Relationships not updating immediately after save

I have a general solution to this that seems a bit cleaner than monkey-patching the core or even using celery (although I'm sure someone could find areas where it fails). Basically I add a clean() method in the admin for the form that has the m2m relationships, and set the instance relations to the cleaned_data version. This make the correct data available to the instance's save method, even though it's not "on the books" yet. Try it and see how it goes:

def clean(self, *args, **kwargs):
    # ... actual cleaning here
    # then find the m2m fields and copy from cleaned_data to the instance
    for f in self.instance._meta.get_all_field_names():
        if f in self.cleaned_data:
            field = self.instance._meta.get_field_by_name(f)[0]
            if isinstance(field, ManyToManyField):
                setattr(self.instance,f,self.cleaned_data[f])

When you are trying to access the ManyToMany fields in the post_save signal of the model, the related objects have already been removed and will not be added again until after the signal is finished.

To access this data you have to tie into the save_related method in your ModelAdmin. Unfortunately you'll also have to include the code in the post_save signal for non-admin requests that require customization.

see: https://docs.djangoproject.com/en/1.7/ref/contrib/admin/#django.contrib.admin.ModelAdmin.save_related

Example:

# admin.py
Class GroupAdmin(admin.ModelAdmin):
    ...
    def save_related(self, request, form, formsets, change):
        super(GroupAdmin, self).save_related(request, form, formsets, change)
        # do something with the manytomany data from the admin
        form.instance.users.add(some_user)

Then in your signals you can make the same changes that you want to execute on a save:

# signals.py
@receiver(post_save, sender=Group)
def group_post_save(sender, instance, created, **kwargs):
    # do somethign with the manytomany data from non-admin
    instance.users.add(some_user)
    # note that instance.users.all() will be empty from the admin: []

See http://gterzian.github.io/Django-Cookbook/signals/2013/09/07/manipulating-m2m-with-signals.html

problem: When you manipulate the m2m of a model within a post or pre_save signal receiver, your changes get wiped out in the subsequent 'clearing' of the m2m by Django.

solution: In you post or pre_save signal handler, register another handler to the m2m_changed signal on the m2m intermediary model of the model whose m2m you want to update.

Please note that this second handler will receive several m2m_changed signals, and it is key to test for the value of the 'action' arguments passed along with them.

Within this second handler, check for the 'post_clear' action. When you receive a signal with the post_clear action, the m2m has been cleared by Django and you have a chance to successfully manipulate it.

an example:

def save_handler(sender, instance, *args, **kwargs):
    m2m_changed.connect(m2m_handler, sender=sender.m2mfield.through, weak=False)


def m2m_handler(sender, instance, action, *args, **kwargs):
    if action =='post_clear':
        succesfully_manipulate_m2m(instance)


pre_save.connect(save_handler, sender=YouModel, weak=False)

see https://docs.djangoproject.com/en/1.5/ref/signals/#m2m-changed

When you save a model via admin forms it's not an atomic transaction. The main object gets saved first (to make sure it has a PK), then the M2M is cleared and the new values set to whatever came out of the form. So if you are in the save() of the main object you are in a window of opportunity where the M2M hasn't been updated yet. In fact, if you try to do something to the M2M, the change will get wiped out by the clear(). I ran into this about a year ago.

The code has changed somewhat from the pre-ORM refactor days, but it boils down to code in django.db.models.fields.ManyRelatedObjectsDescriptor and ReverseManyRelatedObjectsDescriptor. Look at their __set__() methods and you'll see manager.clear(); manager.add(*value) That clear() complete cleans out any M2M references for the current main object in that table. The add() then sets the new values.

So to answer your question: yes, this is a transaction issue.

Is there a signal thrown when the transaction ends? Nothing official, but read on:

There was a related thread a few months ago and MonkeyPatching was one method proposed. Grégoire posted a MonkeyPatch for this. I haven't tried it, but it looks like it should work.