在Django的2.2文件，我使用的，给出了下面的例子中使用select_for_update：

from django.db import transaction

entries = Entry.objects.select_for_update().filter(author=request.user)
with transaction.atomic():
    for entry in entries:
        ...

Using this approach, one would presumably mutate the model instances assigned to entry and call save on these.

There are cases where I'd prefer the alternative approach below, but I'm unsure whether it would work (or even make sense) with select_for_update.

with transaction.atomic():
    Entry.objects.select_for_update().filter(author=request.user).update(foo="bar", wobble="wibble")

The documentation states that the lock is created when the queryset is evaluated, so I doubt the update method would work. As far as I'm aware update just performs an UPDATE ... WHERE query, with no SELECT before it. However, I would appreciate it if someone more experienced with this aspect of the Django ORM could confirm this.

A secondary question is whether a lock even adds any protection against race conditions if one makes a single UPDATE query against the locked rows. (I've entered this train of thought because I'm refactoring code that uses a lock when updating the values of two columns of a single row.)

回答

with transaction.atomic():
    qs = polls.models.Question.objects.select_for_update().all()
    qs.update(question_text='test')

print(connection.queries)
# {'sql': 'UPDATE "polls_question" SET "question_text" = 'test'', 'time': '0.008'}

with transaction.atomic():
    qs = polls.models.Question.objects.select_for_update().all()
    list(qs) # cause evaluation, locking the selected rows
    qs.update(question_text='test')

print(connection.queries)
#[...
# {'sql': 'SELECT "polls_question"."id", "polls_question"."question_text", "polls_question"."pub_date" FROM "polls_question" FOR UPDATE', 'time': '0.003'},
# {'sql': 'UPDATE "polls_question" SET "question_text" = 'test'', 'time': '0.001'}
#]