python - Django 和 Postgres 中为 select_for_update 生成的查询顺序的差异

Question

我遇到了一个奇怪的情况，当在 transaction.atomic() block 中使用 select_for_update() 时，Django 和 Postgres 中记录的查询顺序不同。

基本上，我有一个 ModelForm，我在其中针对数据库验证 cleaned_data 是否存在重复请求。然后在创建 View 的 form_valid() 方法中，我正在保存实例。为了在同一个事务中执行这两个操作，我覆盖了 post() 方法，并将这两个方法调用包装在 transaction.atomic() 中。

这是我上面所说的代码:

# Form
class MenuForm(forms.ModelForm):
    def __init__(self, *args, **kwargs):
        user_id = kwargs.pop('user_id', None)
        super(MenuForm, self).__init__(*args, **kwargs)

    def clean(self):
        cleaned_data = super(MenuForm, self).clean()
        dish_name = cleaned_data.get('dish_name')
        menus = Menu.objects.select_for_update().filter(user_id=self.user_id)

        for menu in menus:
            if menu.dish_name == dish_name:
                self.add_error('dish_name', 'Dish already exists')
                return cleaned_data
        return cleaned_data

# CreateView
class MenuCreateView(CreateView):
    form_class = MenuForm

    def get_form_kwargs(self):
        kwargs = super(MenuCreateView, self).get_form_kwargs()
        kwargs.update({'user_id': self.request.session.get('user_id')})
        return kwargs

    def form_valid(self, form):
        user = User.objects.get(id=self.request.session.get('user_id'))
        form.instance.user = user
        return super(MenuCreateView, self).form_valid(form)

    def post(self, request, *args, **kwargs):
        form = self.get_form()

        with transaction.atomic():
            if form.is_valid():
                return self.form_valid(form)
            else:
                return self.form_invalid(form)

现在假设我同时发出两个请求，以创建包含同一道菜的菜单。我预计第二个请求会失败。但是，他们两个都在过去。看起来，第二个交易没有看到前一个交易中所做的更改。因此，在 select_for_update() 返回的两个事务中，总的 menus 保持不变。

鉴于 Postgres 默认隔离级别是 READ COMMITTED，我希望这些更改是可见的。因此，我尝试记录查询以查看 COMMIT；在正确的时间被解雇。这是 django 和 postgres 中的查询日志:

Django 日志:

SELECT "menu"."id", "menu"."dish_id", "menu"."dish_name" FROM "menu" WHERE ("menu"."dish_name" = "Test Dish") FOR UPDATE; args=("Test Dish")    
INSERT INTO "menu" ("dish_id", "dish_name") VALUES (2, "Test Dish") RETURNING "menu"."id"; args=(2, "Test Dish")
SELECT "menu"."id", "menu"."dish_id", "menu"."dish_name" FROM "menu" WHERE ("menu"."dish_name" = "Test Dish") FOR UPDATE; args=("Test Dish")
INSERT INTO "menu" ("dish_id", "dish_name") VALUES (2, "Test Dish") RETURNING "menu"."id"; args=(2, "Test Dish")

Postgres 日志:

<2016-03-18 17:55:46.176 IST  0  2/31  56ebf3ca.aac0>LOG:  statement: SHOW default_transaction_isolation
<2016-03-18 17:55:46.177 IST  0  2/32  56ebf3ca.aac0>LOG:  statement: SET TIME ZONE 'UTC'
<2016-03-18 17:55:46.178 IST  0  2/33  56ebf3ca.aac0>LOG:  statement: SELECT t.oid, typarray
    FROM pg_type t JOIN pg_namespace ns
        ON typnamespace = ns.oid
    WHERE typname = 'hstore';

<2016-03-18 17:55:46.182 IST  0  2/34  56ebf3ca.aac0>LOG:  statement: BEGIN
<2016-03-18 17:55:46.301 IST  0  3/2  56ebf3ca.aac1>LOG:  statement: SHOW default_transaction_isolation
<2016-03-18 17:55:46.302 IST  0  3/3  56ebf3ca.aac1>LOG:  statement: SET TIME ZONE 'UTC'
<2016-03-18 17:55:46.302 IST  0  3/4  56ebf3ca.aac1>LOG:  statement: SELECT t.oid, typarray
    FROM pg_type t JOIN pg_namespace ns
        ON typnamespace = ns.oid
    WHERE typname = 'hstore';

<2016-03-18 17:55:46.312 IST  0  3/5  56ebf3ca.aac1>LOG:  statement: BEGIN
<2016-03-18 17:55:46.963 IST  0  3/5  56ebf3ca.aac1>LOG:  statement: SELECT "menu"."id", "menu"."dish_id", "menu"."dish_name" FROM "menu" 
WHERE ("menu"."dish_name" = "Test Dish") FOR UPDATE
<2016-03-18 17:55:46.964 IST  0  2/34  56ebf3ca.aac0>LOG:  statement: SELECT "menu"."id", "menu"."dish_id", "menu"."dish_name" FROM "menu" 
WHERE ("menu"."dish_name" = "Test Dish") FOR UPDATE
<2016-03-18 17:55:47.040 IST  23712  3/5  56ebf3ca.aac1>LOG:  statement: INSERT INTO "menu" ("dish_id", "dish_name") VALUES (2, "Test Dish")RETURNING "menu"."id"
<2016-03-18 17:55:47.061 IST  23712  3/5  56ebf3ca.aac1>LOG:  statement: COMMIT
<2016-03-18 17:55:47.229 IST  23713  2/34  56ebf3ca.aac0>LOG:  statement: INSERT INTO "menu" ("dish_id", "dish_name") VALUES (2, "Test Dish")RETURNING "menu"."id"
<2016-03-18 17:55:47.231 IST  23713  2/34  56ebf3ca.aac0>LOG:  statement: COMMIT

Postgres.conf:

max_connections = 100
log_destination = 'stderr'
logging_collector = on
log_directory = 'pg_log'
log_line_prefix = '<%m  %x  %v  %c>'
log_statement = 'all'

如您所见，两个日志中 SELECT 和 INSERT 查询的顺序不同。我无法理解为什么会这样。此外，如果您注意到，Postgres 日志中 SELECT 查询的 session_id 是不同的。这能说明什么吗？

如果这是预期的行为，我该如何解决这里的核心问题？避免基于现有记录的并发 INSERT 查询。

更新:

我没有提到忽略重复菜单的实际逻辑不仅仅基于菜名。上面一个是简化的例子。

将菜单模型考虑为:

class Menu:
    user_id = models.IntegerField()
    dish = models.ForeignKey(Dish)
    order_start_time = models.DateTimeField()
    order_end_time = models.DateTimeField()

实际逻辑是这样的:

• 从数据库中获取所有带有 dish_name 的菜单。
• 检查所有这些菜单的 order_start_time 和 order_end_time，看看它们是否与 order_start_time 和 order_end_time 重叠 用于新菜单。如果发现冲突，请避免添加。

因此，我们可以为菜肴添加两个菜单 - d1，具有订购窗口 - [9am-10am] 和 [2pm-3pm].

Answer 1

已编辑:

如何让 Django 验证重叠保留？

可以为模型validate_unique添加特殊方法:

    from django.db import models
    from django.core.validators import ValidationError
    from django.forms.forms import NON_FIELD_ERRORS

    class Dish(models.Model):
        name = models.CharField('Dish name', max_length=200)

    class Menu(models.Model):
        user_id = models.IntegerField()
        dish = models.ForeignKey(Dish)
        order_start_time = models.DateTimeField()
        order_end_time = models.DateTimeField()

        def validate_unique(self, *args, **kwargs):
            # call inherited unique validators
            super().validate_unique(*args, **kwargs)  # or super(Menu, self) for Python2.7
            # query if DB already has object with same dish 
            # and overlapping reservation 
            # [order_start_time, order_end_time]
            qs = self.__class__._default_manager.filter(
                    order_start_time__lte=self.order_end_time,
                    order_end_time__gte=self.order_start_time,
                    dish=self.dish,
                )
            # and this object is not the same we are working with
            if not self._state.adding and self.pk is not None:
                qs = qs.exclude(pk=self.pk)
            if qs.exists():
                raise ValidationError({
                    NON_FIELD_ERRORS: ['Overlapping order dates for dish'],
                    })

让我们在控制台中尝试一下:

    from core.models import *
    m=Menu(user_id=1, dish_id=1, order_start_time='2016-03-22 10:00', order_end_time='2016-03-22 15:00')
    m.validate_unique()
    # no output here - all is ok
    m.save()
    print(m.id)
    8

    # lets add duplicate
    m=Menu(user_id=1, dish_id=1, order_start_time='2016-03-22 12:00', order_end_time='2016-03-22 13:00')
    m.validate_unique()
    Traceback (most recent call last):
       File "<console>", line 1, in <module>
       File "/Users/el/tmp/hypothesis_test/menu/core/models.py", line 29, in validate_unique
           NON_FIELD_ERRORS: ['Overlapping order dates for dish'],
    django.core.exceptions.ValidationError: {'__all__': ['Overlapping order dates for dish']}

    # excellent!  dup is found!

    # But! Django helps you find dups but allows you to add them to db if you want it!
    # It's responsibility of your application not to add duplicates.

    m.save()
    print(m.id)
    9

如何确定noboby可以添加重复项？

在这种情况下，您需要在数据库级别进行约束。

在 PostgreSQL 控制台中:

    CREATE EXTENSION btree_gist;

    -- our table:
    SELECT * FROM core_menu;
    id | user_id |    order_start_time    |     order_end_time     | dish_id
   ----+---------+------------------------+------------------------+---------
     8 |       1 | 2016-03-22 13:00:00+03 | 2016-03-22 18:00:00+03 |       1
     9 |       1 | 2016-03-22 15:00:00+03 | 2016-03-22 16:00:00+03 |       1

    DELETE FROM core_menu WHERE id=9; -- we should remove dups before adding unique constraint

    ALTER TABLE core_menu 
        ADD CONSTRAINT core_menu_exclude_dish_same_tstzrange_constr 
            EXCLUDE USING gist (dish_id WITH =, tstzrange(order_start_time, order_end_time)  WITH &&);

现在让我们创建复制对象并将其添加到数据库:

    m=Menu(user_id=1, dish_id=1, order_start_time='2016-03-22 13:00', order_end_time='2016-03-22 14:00')

    m.save()
    Traceback (most recent call last):
       File "/Users/el/tmp/hypothesis_test/venv/lib/python3.5/site-packages/django/db/backends/utils.py", line 64, in execute
         return self.cursor.execute(sql, params)
    psycopg2.IntegrityError: ОШИБКА:  конфликтующее значение ключа нарушает ограничение-исключение "core_menu_exclude_dish_same_tstzrange_constr"
    DETAIL:  Key (dish_id, tstzrange(order_start_time, order_end_time))=(1, ["2016-03-22 13:00:00+00","2016-03-22 14:00:00+00")) conflicts with existing key (dish_id, tstzrange(order_start_time, order_end_time))=(1, ["2016-03-22 10:00:00+00","2016-03-22 15:00:00+00")).

太棒了!现在，数据在程序和数据库级别得到验证。