我该如何在WHERE子句中编写带有子查询的Django查询？

Question

我正在使用Django和Python 3.7。我在弄清楚如何编写Django查询（其中有一个子查询作为where子句的一部分）时遇到了麻烦。这是模型...

class Article(models.Model):
    objects = ArticleManager()
    title = models.TextField(default='', null=False)
    created_on = models.DateTimeField(auto_now_add=True)


class ArticleStat(models.Model):
    objects = ArticleStatManager()
    article = models.ForeignKey(Article, on_delete=models.CASCADE, related_name='articlestats')
    elapsed_time_in_seconds = models.IntegerField(default=0, null=False)
    votes = models.FloatField(default=0, null=False)


class StatByHour(models.Model):
    index = models.FloatField(default=0)
    # this tracks the hour when the article came out
    hour_of_day = IntegerField(
        null=False,
        validators=[
            MaxValueValidator(23),
            MinValueValidator(0)
        ]
    )

在PostGres中，查询看起来类似于

SELECT *
FROM article a,
     articlestat ast
WHERE a.id = ast.article_id
  AND ast.votes > 100 * (
    SELECT "index" 
    FROM statbyhour 
    WHERE hour_of_day = extract(hour from (a.created_on + 1000 * interval '1 second')))

注意子查询是WHERE子句的一部分

ast.votes > 100 * (select index from statbyhour where hour_of_day = extract(hour from (a.created_on + 1000 * interval '1 second'))) 

所以我认为我可以做这样的事情...

hour_filter = Func(
    Func(
        (F("article__created_on") + avg_fp_time_in_seconds * "interval '1 second'"),
        function='HOUR FROM'),
    function='EXTRACT')
...
votes_criterion2 = Q(votes__gte=F("article__website__stats__total_score") / F(
    "article__website__stats__num_articles") * settings.TRENDING_PCT_FLOOR *
                                StatByHour.objects.get(hour_of_day=hour_filter) * day_of_week_index)
qset = ArticleStat.objects.filter(votes_criterion1 & votes_criterion2,
                                  comments__lte=25)

但这会导致“无法将关键字'article'解析为字段。选择包括：hour_of_day，id，index，num_articles，total_score”错误。我认为这是因为Django在运行其中的较大查询之前就对我的“ StatByHour.objects”查询进行了评估，但是我不知道该如何重写以使子查询同时运行。

编辑： K，将我的子查询移到实际的“子查询”函数中，并引用了我使用OuterRef创建的过滤器...

hour_filter = Func(
    Func(
        (F("article__created_on") + avg_fp_time_in_seconds * "interval '1 second'"),
        function='HOUR FROM'),
    function='EXTRACT')
query = StatByHour.objects.get(hour_of_day=OuterRef(hour_filter))


...
votes_criterion2 = Q(votes__gte=F("article__website__stats__total_score") / F(
    "article__website__stats__num_articles") * settings.TRENDING_PCT_FLOOR *
                                Subquery(query) * 
                 day_of_week_index)
qset = ArticleStat.objects.filter(votes_criterion1 & votes_criterion2,
                                  comments__lte=25)

这将导致

This queryset contains a reference to an outer query and may only be used in a subquery.

这很奇怪，因为我在子查询中使用它。

编辑＃2：即使根据给定的答案更改了查询后...

hour_filter = Func(
    Func(
        (F("article__created_on") + avg_fp_time_in_seconds * "interval '1 second'"),
        function='HOUR FROM'),
    function='EXTRACT')
query = StatByHour.objects.filter(hour_of_day=OuterRef(hour_filter))[:1]

...
votes_criterion2 = Q(votes__gte=F("article__website__stats__total_score") / F(
    "article__website__stats__num_articles") * settings.TRENDING_PCT_FLOOR *
                                Subquery(query) *
                                day_of_week_index)
qset = ArticleStat.objects.filter(et_criterion1 & et_criterion2 & et_criterion3,
                                  votes_criterion1 & votes_criterion2,
                                  article__front_page_first_appeared_date__isnull=True,
                                  comments__lte=25)

我仍然收到错误消息

'Func' object has no attribute 'split'

Answer 1

Subqueries必须是未立即评估的查询，以便可以将其评估推迟到运行外部查询之前。 get()不符合要求，因为它会立即执行并返回一个对象实例，而不是Queryset。

但是，用filter代替get然后取[:1]片应该可以工作：

StatByHour.objects.filter(hour_of_day=OuterRef('hour_filter')).values('hour_of_day')[:1]

请注意，OuterRef中的字段引用是字符串文字而不是变量。

此外，子查询需要返回单列和单行（因为它们被分配给单个字段），因此需要返回values()和上面的切片。

此外，我还没有在Q对象中使用子查询；我不确定是否可以。您可能必须先将子查询输出保存在注释中，然后将其用于过滤器计算。

Answer 2

我发现将尽可能多地移入批注以阐明正在发生的事情很有帮助。

您可以使用Extract函数来获取小时。如果您想合并更复杂的avg_fp_time_in_seconds内容，则需要定义自己的Func，我没有尝试过重复，因为它值得我们发表（这可能是'Func' object has no attribute 'split'错误来自）。

# First, add a field for the hour 
articles_with_hour = Article.objects.annotate(created_on_hour=ExtractHour('created_on'))

# Set up the subquery, referencing the annotated field
for_this_hour = StatByHour.objects.filter(hour_of_day=OuterRef('created_on_hour'))

# Add the subquery, making sure to slice down to one value
articles_with_hour_index = articles_with_hour.annotate(
    index_for_this_hour=Subquery(for_this_hour.values('index')[:1]),
)

# Add the website averages for later calculations 
#  (note if total_score and num_articles are different field types
#  you may need an ExpressionWrapper)
articles_with_avg_website_score = articles_with_hour_index.annotate(
    average_article_score_for_website=(
        F("website__stats__total_score") / F("website__stats__num_articles")
    )
)

# Use the averages to calculate the trending floor for each article
articles_with_trending_floor = articles_with_avg_website_score.annotate(
    trending_floor=F('average_article_score_for_website') * settings.TRENDING_PCT_FLOOR,
)

# Set up the criteria, referencing fields that are already annotated on the qs
# ...
votes_gte_trending_floor_for_this_hour_criterion = Q(articlestats__votes__gte=(
    F('trending_floor')
    * F('index_for_this_hour')
    * day_of_week_index  # not sure where this comes from?
))
# ...

# Then just filter down (note this is an Article QuerySet, not ArticleStat)
qset = articles_with_trending_floor.filter(
    votes_gte_trending_floor_for_this_hour_criterion,
    # other criteria
    front_page_first_appeared_date__isnull=True,
    articlestats__comments__lte=25,
)

这些计算中的许多都可以简化，甚至可以使用多个kwarg在一个annotate调用中完成所有操作，但是我认为将其全部布置起来更容易理解。

Answer 3

这肯定是一个Subquery解决方案。

Django> = 1.11

  

作为警告，我确实测试了代码，但是仅使用模型，我就没有任何数据，因此，此答案只是为了向正确的方向指出您的努力

x = pd.DataFrame(df.iloc[:, [1,0]].values, columns=df.columns)
pd.concat([df, x], sort=False)

    col_1   col_2
0   a       1
1   b       2
2   c       3
3   d       4
0   1       a
1   2       b
2   3       c
3   4       d

乍一看，看起来您可能需要在子查询中执行# Query that references an outer field from another model, in this case created_on. # On wich we are performing a lookup in order to "extract" the hour (assuming here) # a DateTimeField or a TimeField. stat_by_hour = StatByHour.objects.filter(hour_of_day=OuterRef('created_on__hour')) # Then filter articles, that have articlestats.votes # greater than 100 * stat_by_hour.index result = Article.objects.filter( articlestats__votes__gt=100 * Subquery(stat_by_hour.values('index')[:1], output_field=FloatField()) ) 或order_by('index')，这样切片order_by('-index')将获得最小值或最大值（取决于您的需求。）

我确定您可以使用此功能（或非常相似的功能）来实现所需的功能。

Answer 4

使用由hour_of_day=ExtractHour(OuterRef('article__created_on') + timedelta(seconds=avg_fp_time_in_seconds))过滤的子查询进行过滤。实际代码将需要一个额外的ExpressionWrapper，并且仅适用于Django >= 2.1.0：

import datetime

from django.db import models
from django.db.models import F, OuterRef, Subquery, Value
from django.db.models.functions import ExtractHour, Coalesce
from django.db.models.expressions import ExpressionWrapper


relevant_hour_stats = (
    StatByHour.objects
    .filter(
        hour_of_day=ExtractHour(ExpressionWrapper(
            OuterRef('article__created_on')  # NOTE: `OuterRef()+Expression` works only on Django >= 2.1.0
            +
            datetime.timedelta(seconds=avg_fp_time_in_seconds),
            output_field=models.DateTimeField()
        )),
    )
    .annotate(
        votes_threshold=Coalesce(
            100.0 * F('index'),
            0.0,
            output_field=models.FloatField(),
        ),
    )
    .order_by('-votes_threshold')
    # NOTE: your StatByHour model does not have unique=True on hour_of_day
    # field, so there may be several stat for same hour.
    # And from your SQL example it's unclear how should they be handled. So I
    # assume that "greatest" threshold is needed.
)

article_stats = (
    ArticleStat.objects
    .all()
    .filter(
        votes__gt=Coalesce(
            Subquery(relevant_hour_stats.values('votes_threshold')[:1]),
            Value(0.0),
            output_field=models.FloatField(),
        ),
    )
)

P.S。如果您在github上建立一些“演示项目”，那么任何人都可以克隆它并在本地检查他们的想法，那会容易得多。

P.P.S。经过测试，该代码可以正常运行，但是可以在不同的模型/字段上使用：

In [15]: relevant_something = (ModelOne.objects.filter(index=ExtractHour(ExpressionWrapper(OuterRef('due_date') + datetime.timedelta(seconds=1000), output_field=models.DateTimeField()))).annotate(votes_threshold=100*F('indent')).order_by('-votes_threshold'))

In [16]: ts = ModelTwo.objects.all().filter(votes__gt=Subquery(relevant_notes.values('votes_threshold')[:1], output_field=models.IntegerField()))

In [17]: print(ts.query)

SELECT 
    ...
FROM 
    "some_app_model_two" 
WHERE 
    "some_app_model_two"."votes" > (
        SELECT 
            (100 * U0."indent") AS "votes_threshold" 
        FROM 
            "some_app_model_one" U0 
        WHERE 
            U0."index" = (
                EXTRACT(
                    'hour' 
                    FROM ("some_app_model_two"."due_date" + 0:16:40) 
                    AT TIME ZONE 'America/Los_Angeles'
                )
            ) 
        ORDER BY "votes_threshold" DESC 
        LIMIT 1
    )
ORDER BY 
    "some_app_model_two"."due_date" ASC, 
    "some_app_model_two"."priority" ASC, 
    "some_app_model_two"."updated_at" DESC

因此，如果您遇到任何错误，请显示您正在运行的ACTUAL代码

Answer 5

看看Django queries。我认为您可以通过将SQL基本查询更改为Django提供的内容来解决问题。

如果它不起作用，则可以perform raw SQL queries。