使用django从数据库中随机取N条记录的不同方法及其性能实测

这里（stackoverflow）有一篇关于使用Django随机获取记录的讨论。

主要意思是说

Record.objects.order_by('?')[:2]

这样获取2个记录会导致性能问题，原因如下：

“

对于有着相当多数量记录的表来说，这种方法异常糟糕。这会导致一个 ORDER BY RAND() 的SQL查询。举个栗子，这里是MYSQL是如何处理这个查询的（其他数据库的情况也差不多），想象一下当一个表有十亿行的时候会怎样：

1. 为了完成ORDER BY RAND() ，需要一个RAND()列来排序
2. 为了有RAND()列，需要一个新表，因为现有的表没有这个列。
3. 为了这个新表，mysql建立了一个带有新列的，新的临时表，并且将已有的一百万行数据复制进去。
4. 当其新建完了，他如你所要求的，为每一行运行RAND()函数来填上这个值。是的，你派mysql创建一百万个随机数，这要点时间：）
5. 几个小时或几天后，当他干完这活，他要排序。是的，你排mysql去排序一个一百万行的，最糟糕的表（说他最糟糕是因为排序的键是随机的）。
6. 几天或者几星期后，当排序完了，他忠诚地将你实际需要的可怜的两行抓出来返回给你。做的好。；）

注意：只是稍微说一句，得注意到mysql一开始会试着在内存中创建临时表。当内存不够了，他将会把所有东西放在硬盘上，所以你会因为近乎于整个过程中的I/O瓶颈而雪上加霜。

怀疑者可以去看看python代码引起的查询语句，确认是ORDER BY RAND()， 然后去Google下"order by rand()"（带上引号）。

一个更好的方式是将这个耗费严重的查询换成3个耗费更轻的：

last = MyModel.objects.count() - 1# 这是一个获取两个不重复随机数的简单方法index1 = randint(0, last)index2 = randint(0, last - 1)if index2 == index1:     index2 = lastMyObj1 = MyModel.objects.all()[index1]MyObj2 = MyModel.objects.all()[index2]

”

如上Manganeez所说的方法，相应的获取n条记录的代码应该如下：

sample = random.sample(xrange(Record.objects.count()),n)result = [Record.objects.all()[i]) for i in sample]

基于Python代码应该简洁优雅的想法，如上的代码似乎可以写成：

result = random.sample(Record.objects.all(),n)

就性能问题，请教了stackoverflow上的大神 （虽然被踩和被教育了=。=）

“
Record.objects.count() 将被转换成一个相当轻量级的SQL请求：

SELECT COUNT(*) FROM TABLE

Record.objects.all()[0]也会被转换成一个十分轻量级的SQL请求：

SELECT * FROM TABLE LIMIT 1

Querying all 是一个耗费十分严重的请求

SELECT * FROM TABLE

通常情况下Django会不显示其他的结果，这样你不会真正的获取到所有的记录。

SELECT * FROM table LIMIT 20;  // or something similar

任何时候你将一个Queryset转换成list的时候，将是资源消耗严重的时候。

如果我没错的话，在这个例子里，sample方法将把Queryset转换成list。

这样如果你result = random.sample(Record.objects.all(),n) 这样做的话，全部的Queryset将会转换成list，然后从中随机选择。

想象一下如果你有十亿行的数据。你是打算把它存储在一个有百万元素的list中，还是愿意一个一个的query？

”

在上边Yeo的回答中，freakish回复道:

“

.count的性能是基于数据库的。而Postgres的.count为人所熟知的相当之慢。

”

某人说过，要知道梨子的滋味，就得变革梨子，亲口尝一尝。

在一个已有的测试project中新建一个app，数据库是MYSQL：

D:\PyWorkspace\DjangoTest>python manage.py startapp randomrecords

在models.py中添加模型：

class Record(models.Model):    """docstring for Record"""        id = models.AutoField(primary_key = True)    content = models.CharField(max_length = 16)    def __str__(self):        return "id:%s content:%s" % (self.id, self.content)    def __unicode__(self):        return u"id:%s content:%s" % (self.id, self.content)

添加一万行数据：

D:\PyWorkspace\DjangoTest>python manage.py syncdbCreating tables ...Creating table randomrecords_recordInstalling custom SQL ...Installing indexes ...Installed 0 object(s) from 0 fixture(s)D:\PyWorkspace\DjangoTest>python manage.py shellPython 2.7.5 (default, May 15 2013, 22:44:16) [MSC v.1500 64 bit (AMD64)] on win32Type "help", "copyright", "credits" or "license" for more information.(InteractiveConsole)>>> from randomrecords.models import Record>>> for i in xrange(10000):...   Record.objects.create(content = 'c of %s' % i).save()...

15分钟以后我得到了这个MYSQL表。真的，不骗你，真的是15分钟。看了记录才知道 每次save都要调用一次insert和一次update。。。。下次一定用SQL语句初始化。。。。

先写了个脚本 在manage.py shell中调用了下 结果让我震惊了。

我表示不敢相信 又写了view 并添加了显示Query的log 

这里是写的view：

def test1(request):    start = datetime.datetime.now()    result = Record.objects.order_by('?')[:20]    l = list(result) # Queryset是惰性的，强制将Queryset转为list    end = datetime.datetime.now()    return HttpResponse("time: <br/> %s" % (end-start).microseconds/1000))def test2(request):    start = datetime.datetime.now()    sample = random.sample(xrange(Record.objects.count()),20)    result = [Record.objects.all()[i] for i in sample]    l = list(result)    end = datetime.datetime.now()    return HttpResponse("time: <br/> %s" % (end-start)def test3(request):    start = datetime.datetime.now()    result = random.sample(Record.objects.all(),20)    l = list(result)    end = datetime.datetime.now()    return HttpResponse("time: <br/> %s" % (end-start)

运行结果如下，第一行是页面显示的时间，后边是Queryset实际调用的SQL语句：

test1：time: 0:00:00.012000(0.009) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` ORDER BY RAND() LIMIT 20; args=()[05/Dec/2013 17:48:19] "GET /dbtest/test1 HTTP/1.1" 200 775test2:time: 0:00:00.055000(0.002) SELECT COUNT(*) FROM `randomrecords_record`; args=()(0.002) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 6593; args=()(0.001) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 2570; args=()(0.001) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 620; args=()(0.001) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 5814; args=()(0.003) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 6510; args=()(0.002) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 3536; args=()(0.001) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 3362; args=()(0.003) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 8948; args=()(0.002) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 7723; args=()(0.001) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 2374; args=()(0.002) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 8269; args=()(0.002) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 4370; args=()(0.002) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 6953; args=()(0.001) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 1441; args=()(0.000) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 772; args=()(0.002) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 4323; args=()(0.002) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 8139; args=()(0.002) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 7441; args=()(0.001) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 1306; args=()(0.001) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 5462; args=()[05/Dec/2013 17:50:34] "GET /dbtest/test2 HTTP/1.1" 200 777test3:time: 0:00:00.156000(0.032) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record`; args=()[05/Dec/2013 17:51:29] "GET /dbtest/test3 HTTP/1.1" 200 774

令人难以置信的，在10000行的MYSQL表中 方法1的效率是最高的

无论是结果上看（12ms）还是SQL语句的运行时间上看（9ms）方法1甩了其他方法一大截

即便数据量增加到21万：

time: 98(0.094) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` ORDER BY RAND() LIMIT 20; args=()[05/Dec/2013 19:18:59] "GET /dbtest/test1 HTTP/1.1" 200 14time: 0:00:00.668000//这里没有注意到 掉了一行count语句(0.045) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 176449; args=()(0.016) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 68082; args=()(0.036) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 145571; args=()(0.033) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 111029; args=()(0.043) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 169675; args=()(0.046) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 186234; args=()(0.043) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 167233; args=()(0.015) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 54404; args=()(0.036) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 140395; args=()(0.004) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 13128; args=()(0.039) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 153695; args=()(0.034) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 131863; args=()(0.021) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 82785; args=()(0.015) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 57253; args=()(0.021) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 77836; args=()(0.049) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 199567; args=()(0.002) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 3867; args=()(0.027) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 104470; args=()(0.026) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 107058; args=()(0.043) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 150979; args=()[05/Dec/2013 19:21:33] "GET /dbtest/test2 HTTP/1.1" 200 15time 0:00:00.781000[05/Dec/2013 19:23:01] "GET /dbtest/test3 HTTP/1.1" 200 15(0.703) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record`; args=()[05/Dec/2013 19:23:06] "GET /dbtest/test3 HTTP/1.1" 200 15

方法1也会比其他两种方法快

数据量再次提升至百万级别 1066768条数据

time: 0:00:02.197000(2.193) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` ORDER BY RAND() LIMIT 20; args=()[05/Dec/2013 20:00:55] "GET /dbtest/test1 HTTP/1.1" 200 26time: 0:00:02.659000(0.204) SELECT COUNT(*) FROM `randomrecords_record`; args=()(0.180) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 703891; args=()(0.038) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 156668; args=()(0.013) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 50742; args=()(0.031) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 121107; args=()(0.033) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 130565; args=()(0.017) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 66225; args=()(0.234) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 922479; args=()(0.267) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 1027166; args=()(0.189) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 765499; args=()(0.009) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 31569; args=()(0.233) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 934055; args=()(0.264) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 1052741; args=()(0.155) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 621692; args=()(0.014) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 52388; args=()(0.199) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 759669; args=()(0.170) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 655598; args=()(0.035) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 139709; args=()(0.228) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 919480; args=()(0.104) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 422051; args=()(0.017) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 67549; args=()[05/Dec/2013 20:00:45] "GET /dbtest/test2 HTTP/1.1" 200 26time: 0:00:19.651000(3.645) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record`; args=()[05/Dec/2013 20:02:50] "GET /dbtest/test3 HTTP/1.1" 200 26

第三种方法所用时间长到令人无法接受 但有意思的是 SQL语句所花费的时间“只有”3.6秒。而大部分的时间都用在python上了。

既然第二种方法和第三种方法都需要random.sample 一个百万个数据的list，那就是说，有大量的时间花费在将SELECT到的结果转化为django对象的过程中了。

此后将不再测试第三种方法

最后，数据量增加到5,195,536个

time: 0:00:22.278000(22.275) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` ORDER BY RAND() LIMIT 20; args=()[05/Dec/2013 21:46:33] "GET /dbtest/test1 HTTP/1.1" 200 26time: 0:00:33.319000(1.393) SELECT COUNT(*) FROM `randomrecords_record`; args=()(3.201) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 4997880; args=()(1.229) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 2169311; args=()(0.445) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 1745307; args=()(1.306) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 3233861; args=()(1.881) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 3946647; args=()(1.624) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 3534377; args=()(1.068) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 1684337; args=()(0.902) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 2607361; args=()(2.938) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 4872494; args=()(0.493) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 851494; args=()(3.275) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 5182414; args=()(0.946) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 1684670; args=()(0.701) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 1819730; args=()(0.915) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 1626221; args=()(1.809) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 3638682; args=()(3.237) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 4801027; args=()(1.187) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 1955843; args=()(2.736) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 4835733; args=()(1.705) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 2756641; args=()(0.286) SELECT `randomrecords_record`.`id`, `randomrecords_record`.`content` FROM `randomrecords_record` LIMIT 1 OFFSET 1117426; args=()

随着表中数据行数的增加，两个方法的所用的时间都到了一个完全不能接受的程度。两种方法所用的时间也几乎相同。

值得注意的是，Mysql数据库有一个特点是，对于一个大表，OFFSET越大，查询时间越长。或许有其他方法可以在offset较大的时候加快select的速度，然而django明显没有做到。如果能够减少这种消耗，方法2明显会优于方法1。

附上三种方法数据量和SQL时间/总时间的图

最后总结，Django下，使用mysql数据库，数据量在百万级以下时，使用

Record.objects.order_by('?')[:2]

来获取随机记录系列，性能不会比

sample = random.sample(xrange(Record.objects.count()),n)result = [Record.objects.all()[i]) for i in sample]

差。