在对Django模型进行平均时，为什么MYSQL DB返回损坏的值？

Question

I'm running a Django application on top of a MySQL (actually MariaDB) database.

My Django Model looks like this:

from django.db import models
from django.db.models import Avg, Max, Min, Count

class myModel(models.Model):
    my_string = models.CharField(max_length=32,)
    my_date = models.DateTimeField()

    @staticmethod
    def get_stats():            
        logger.info(myModel.objects.values('my_string').annotate(
                count=Count("my_string"), 
                min=Min('my_date'), 
                max=Max('my_date'), 
                avg=Avg('my_date'),
            )
        )

When I run get_stats() , I get the following log line:

[2015-06-21 09:45:40] INFO [all_logs:96] [{'my_string': u'A', 'count': 2, 'avg': 20080507582679.5, 'min': datetime.datetime(2007, 8, 2, 11, 33, 53, tzinfo=<UTC>), 'max': datetime.datetime(2009, 2, 13, 5, 20, 6, tzinfo=<UTC>)}]

The problem I have with this is that the average of the my_date field returned by the database is: 20080507582679.5 . Look carefully at that number. It is an invalid date format.

Why doesn't the database return a valid value for the average of these two dates? How do I get the actual average of this field if the way described fails? Is Django DateTimeField not setup to do handle averaging?

Answer 1

Q1: Why doesn't the database return a valid value for the average of these two dates?

A: The value returned is expected, it's well defined MySQL behavior.

MySQL automatically converts a date or time value to a number if the value is used in a numeric context and vice versa.

MySQL Reference Manual: https://dev.mysql.com/doc/refman/5.5/en/date-and-time-types.html

---

In MySQL, the AVG aggregate function operates on numeric values.

In MySQL, a DATE or DATETIME expression can be evaluated in a numeric context.

As a simple demonstration, performing an numeric addition operation on a DATETIME implicitly converts the datetime value into a number. This query:

  SELECT NOW(), NOW()+0

returns a result like:

  NOW()                                NOW()+0  
  -------------------  -----------------------
  2015-06-23 17:57:48    20150623175748.000000

Note that the value returned for the expression NOW()+0 is not a DATETIME , it's a number .

When you specify a SUM() or AVG() function on a DATETIME expression, that's equivalent to converting the DATETIME into a number, and then summing or averaging the number.

That is, the return from this expression AVG(mydatetimecol) is equivalent to the return from this expression: AVG(mydatetimecol+0)

What is being "averaged" is a numeric value. And you have observed, the value returned is not a valid datetime; and even in cases where it happens to look like a valid datetime, it's likely not a value you would consider a true "average".

---

Q2: How do I get the actual average of this field if the way described fails?

A2: One way to do that is to convert the datetime into a numeric value that can be "accurately" averaged, and then convert that back into a datetime.

For example, you could convert the datetime into a numeric value representing a number of seconds from some fixed point in time, eg

  TIMESTAMPDIFF(SECOND,'2015-01-01',t.my_date)

You could then "average" those values, to get an average number of seconds from a fixed point in time. (NOTE: beware of adding up an extremely large number of rows, with extremely large values, and exceeding the limit (maximum numeric value), numeric overflow issues.)

  AVG(TIMESTAMPDIFF(SECOND,'2015-01-01',t.my_date))

To convert that back to a datetime, add that value as a number of seconds back to a the fixed point in time:

  '2015-01-01' + INTERVAL AVG(TIMESTAMPDIFF(SECOND,'2015-01-01',t.my_date)) SECOND

(Note that the DATEIME values are evaluated in the timezone of the MySQL session; so there are edge cases where the setting of the time_zone variable in the MySQL session will have some influence on the value returned.)

MySQL also provides a UNIX_TIMESTAMP() function which returns a unix-style integer value, number of seconds from the beginning of the era (midnight Jan. 1, 1970 UTC). You can use that to accomplish the same operation more concisely:

  FROM_UNIXTIME(AVG(UNIX_TIMESTAMP(t.my_date)))

Note that this final expression is really doing the same thing... converting the datetime value into a number of seconds since '1970-01-01 00:00:00' UTC, taking a numeric average of that, and then adding that average number of seconds back to '1970-01-01' UTC, and finally converting that back to a DATETIME value, represented in the current session time_zone .

---

Q3: Is Django DateTimeField not setup to do handle averaging?

A: Apparently, the authors of Django are satisfied with the value returned from the database for a SQL expression AVG(datetime) .

Answer 2

Plan A: Use a TIMESTAMP field instead of a DATETIME field

Plan B: Convert DATETIME to TIMESTAMP during the computation:

FROM_UNIXTIME(ROUND(AVG(UNIX_TIMESTAMP(`my_date`))))

(Sorry, I don't know the Django syntax needed.)

Answer 3

When you use values() , Django will not convert the value it got from the database-python connector. It's up to the connector to determine how the value is returned.

In this case, it seems that the MySQL connector returns a string-representation with the separators removed. You can try to use datetime.strptime() with a matching format to parse it into a datetime object.