我有这个简单的模型:

模型.py

class Ping(models.Model):
    online = models.BooleanField()
    created = models.DateTimeField(db_index=True, default=timezone.now)

    def __str__(self):
        return f'{self.online}, {self.created}'

它给了我以下结果:

mysql [lab]> SELECT * FROM myapp_ping;
+----+--------+----------------------------+
| id | online | created                    |
+----+--------+----------------------------+
|  1 |      1 | 2018-08-02 13:34:09.435292 |
|  2 |      1 | 2018-08-02 13:35:09.520200 |
|  3 |      0 | 2018-08-02 13:36:09.540638 |
|  4 |      0 | 2018-08-02 13:37:10.529783 |
|  5 |      1 | 2018-08-02 13:38:09.779012 |
|  6 |      1 | 2018-08-02 13:39:09.650365 |
|  7 |      1 | 2018-08-02 13:40:09.625543 |
|  8 |      1 | 2018-08-02 13:41:09.892196 |
|  9 |      1 | 2018-08-02 13:42:09.802186 |
| 10 |      1 | 2018-08-02 13:43:09.864551 |
| 11 |      1 | 2018-08-02 13:44:09.960962 |
| 12 |      1 | 2018-08-02 13:45:09.891947 |
| 13 |      0 | 2018-08-02 13:46:09.141727 |
| 14 |      0 | 2018-08-02 13:47:09.142030 |
| 15 |      0 | 2018-08-02 13:48:09.160942 |
| 16 |      0 | 2018-08-02 13:49:09.152879 |
| 17 |      0 | 2018-08-02 13:50:09.280246 |
| 18 |      1 | 2018-08-02 13:51:09.363184 |
| 19 |      1 | 2018-08-02 13:52:09.405863 |
| 20 |      1 | 2018-08-02 13:53:09.403251 |
+----+--------+----------------------------+
20 rows in set (0.00 sec)

有没有办法得到类似这样的输出(online 为 false 的范围):

停机时间:

from                | to                  | duration
2018-08-02 13:36:09 | 2018-08-02 13:37:10 | 1 minute and 1 second
2018-08-02 13:46:09 | 2018-08-02 13:50:09 | 4 minutes and 0 seconds

我不确定这是否可以通过 Django ORM 完成，或者它是否需要原始 MySQL 查询来使用类似 CASE 或 IF 语句？

更新:UTC 2018 年 8 月 8 日星期三 15:13:15

所以我从 @AKX answer 获得了两种解决方案的概念证明:

模型.py

class PingManager(models.Manager):
    def downtime_python(self):
        queryset = super().get_queryset().filter(created__gt=timezone.now() - timezone.timedelta(days=30))
        offline = False
        ret = []
        for entry in queryset:
            if not entry.online and not offline:
                offline = True
                _ret = {'start': str(entry.created)}
            if entry.online and offline:
                _ret.update({'end': str(entry.created)})
                ret.append(_ret)
                offline = False
        return ret

    def downtime_sql(self):
        queryset = super().get_queryset().filter(created__gt=timezone.now() - timezone.timedelta(days=30))
        offline = queryset.filter(online=False).order_by('created').first()
        last = queryset.order_by('created').last()
        ret = []
        if offline:
            online = queryset.filter(created__gt=offline.created, online=True).order_by('created').first()
            ret.append({'start': str(offline.created), 'end': str(online.created)})
            while True:
                offline = queryset.filter(created__gt=online.created, online=False).order_by('created').first()
                if offline:
                    online = queryset.filter(created__gt=offline.created, online=True).order_by('created').first()
                if (online and offline) and online.created < last.created:
                    ret.append({'start': str(offline.created), 'end': str(online.created)})
                    continue
                else:
                    break
        return ret

class Ping(models.Model):
    online = models.BooleanField()
    created = models.DateTimeField(db_index=True, default=timezone.now)
    objects = PingManager()

    def __str__(self):
        return f'{self.online}, {self.created}'

问题:

1. 我应该为此创建静态方法还是自定义 manger 是正确的解决方案？

2. 如果两个计算都在内存中运行，为什么执行时间会有如此巨大的差异？有没有办法改进并使其更像 Python 等效方法？

测试:

# python manage.py shell
Python 3.6.5 (default, Apr 10 2018, 17:08:37) 
Type 'copyright', 'credits' or 'license' for more information
IPython 6.5.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: from myapp.models import Ping

In [2]: Ping.objects.downtime_sql()[0]
Out[2]: 
{'start': '2018-07-13 16:32:16.009356+00:00',
 'end': '2018-07-13 16:33:15.942784+00:00'}

In [3]: Ping.objects.downtime_python()[0]
Out[3]: 
{'start': '2018-07-13 16:32:16.009356+00:00',
 'end': '2018-07-13 16:33:15.942784+00:00'}

In [4]: Ping.objects.downtime_sql() == Ping.objects.downtime_python()
Out[4]: True

In [5]: import timeit

In [6]: timeit.timeit(stmt=Ping.objects.downtime_python, number=1)
Out[6]: 5.720254830084741

In [7]: timeit.timeit(stmt=Ping.objects.downtime_sql, number=1)
Out[7]: 0.25946347787976265

最佳答案

扩展我的评论:

I'm not sure even SQL case/if statements can get you that result, since the result rows depend on previous rows. This is easy to do procedurally in Python though.

1. 最明显的方法是循环遍历 Ping.objects.all()(或 Ping.objects.iterator())并跟踪 online 变量来形成你想要的“条纹”。这样做的缺点是您确实需要遍历每个对象，这最终会很慢(和/或耗尽您的内存)。
2. 一种更复杂的方法，它使用更多的查询但更少的内存，是找到第一个离线的 Ping 对象，然后找到下一个(按时间)Ping 对象再次联机——这将形成一个连胜。然后冲洗并重复此操作，直到用完要检查的 Ping 对象。

编辑

是的，这是方法 2 的具体实现(相当优雅，如果你不介意我这么说的话)(在 https://github.com/akx/so51656477 找到完整的测试库):

class PingQuerySet(models.QuerySet):
    def streaks(self):
        queryset = self.values_list('created', 'online').order_by('created')
        entry = queryset.first()
        while entry:
            next_entry = queryset.filter(created__gt=entry[0], online=(not entry[1])).first()
            yield (entry, next_entry)
            entry = next_entry

它是元组的二元组生成器:((start_timestamp, start_online), (end_timestamp, end_online) | None)。

例如，要获取最近 10 天的上涨/下跌或下跌/上涨对，

for start, end in Ping.objects.filter(created__gt=now() - timedelta(days=10)).streaks():
    print(start, end)

会打印类似的东西

[...snip...]

(datetime.datetime(2018, 8, 8, 8, 10, 12, 943500), False) (datetime.datetime(2018, 8, 8, 10, 10, 12, 943500), True)
(datetime.datetime(2018, 8, 8, 10, 10, 12, 943500), True) (datetime.datetime(2018, 8, 8, 11, 10, 12, 943500), False)
(datetime.datetime(2018, 8, 8, 11, 10, 12, 943500), False) (datetime.datetime(2018, 8, 8, 11, 40, 12, 943500), True)
(datetime.datetime(2018, 8, 8, 11, 40, 12, 943500), True) (datetime.datetime(2018, 8, 8, 12, 40, 12, 943500), False)
(datetime.datetime(2018, 8, 8, 12, 40, 12, 943500), False) (datetime.datetime(2018, 8, 8, 16, 40, 12, 943500), True)
(datetime.datetime(2018, 8, 8, 16, 40, 12, 943500), True) (datetime.datetime(2018, 8, 8, 17, 40, 12, 943500), False)
(datetime.datetime(2018, 8, 8, 17, 40, 12, 943500), False) (datetime.datetime(2018, 8, 8, 18, 10, 12, 943500), True)
(datetime.datetime(2018, 8, 8, 18, 10, 12, 943500), True) (datetime.datetime(2018, 8, 8, 19, 40, 12, 943500), False)
(datetime.datetime(2018, 8, 8, 19, 40, 12, 943500), False) (datetime.datetime(2018, 8, 8, 23, 10, 12, 943500), True)
(datetime.datetime(2018, 8, 8, 23, 10, 12, 943500), True) (datetime.datetime(2018, 8, 9, 0, 10, 12, 943500), False)
(datetime.datetime(2018, 8, 9, 0, 10, 12, 943500), False) (datetime.datetime(2018, 8, 9, 3, 10, 12, 943500), True)
(datetime.datetime(2018, 8, 9, 3, 10, 12, 943500), True) (datetime.datetime(2018, 8, 9, 3, 40, 12, 943500), False)
(datetime.datetime(2018, 8, 9, 3, 40, 12, 943500), False) (datetime.datetime(2018, 8, 9, 5, 10, 12, 943500), True)
(datetime.datetime(2018, 8, 9, 5, 10, 12, 943500), True) (datetime.datetime(2018, 8, 9, 5, 40, 12, 943500), False)
(datetime.datetime(2018, 8, 9, 5, 40, 12, 943500), False) (datetime.datetime(2018, 8, 9, 7, 10, 12, 943500), True)
(datetime.datetime(2018, 8, 9, 7, 10, 12, 943500), True) None

一些注意事项:

• 最后的end值可能是None，这意味着机器仍然是up或down(取决于start的status值> 元组)。
• 如果您只关心机器停机的时间，只需忽略 start 元组的状态值为 True 的对。
• 由于这是一个生成器，当您有足够的数据时，您可以停止迭代它，它不会再进行任何查询。
• 由于这是一个 QuerySet 扩展方法，您可以根据需要添加其他过滤器(只要它们不在 online 上过滤)。例如，如果您有一个 host 字段，Ping.objects.filter(host='example.com').streaks()。

关于python - 带有 MySQL 后端的 Django - 按时间范围分组，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/51656477/