生成全局唯一Id

参考了: http://www.cnblogs.com/heyuquan/p/global-guid-identity-maxId.html

1. GUID

import uuiduuid.uuid1()

优点： 确保唯一， 速度快

缺点： 太长， 不友好, 不好索引

2. 数据库唯一索引

时间戳加上随机数，然后通过数据库做唯一性校验

import timeimport randomimport stringm = time.strftime('%y%m%d%H%M%S') + ''.join([random.choice(string.lowercase + string.digits) for _ in range(5)])#检查m在数据库中是否存在，存在则重复上述过程，不存在则存入数据库并返回

优点：适合简单应用，id较短，有一定亲和力

缺点：每秒id总数有限制，并发越大性能越低, 加大数据库访问压力，需要锁表

优化：将时间戳转成62进制数

digit62 = '0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz'#整数转化为62进制字符串#入口：#   x : 整数#返回： 字符串def int_to_str62( x ):    try:        x=int(x)    except:        x=0    if x<0:        x=-x    if x==0:        return "0"    s=""    while x>=62:        x1= x % 62        s = digit62[x1]+s        x = x // 62    if x>0:       s = digit62[x]+s    return s#62进制字符串转化为整数#入口：#   s : 62进制字符串#返回： 整数def str62_to_int( s ):    x = 0    s = str(s).strip()    if s=="":        return x    for y in s:        k = digit62.find(y)        if k>=0:           x=x*62+k    return ximport timeimport randomimport stringt = time.strftime('%y%m%d%H%M%S')cut = [ t[i:i+2] for i in range(0, len(t), 2) ]62t = ''.join([ int_to_str62(int(x)) for x in cut])m = 62t + ''.join([random.choice(string.lowercase + string.digits) for _ in range(6)])

再ps. 有人说random.choice慢而且随机不均匀，我就写了两个小程序测试一下

import randomimport timeimport stringimport timeitimport hashlibimport uuidimport threadingdef randomchoice():    return ''.join([ random.choice(string.lowercase + string.digits) for _ in range(6)])def _time(f, n=1000000):    print 'start timeit function ', f    t = timeit.timeit(f, number=n)    print 'repeat %s times and used %ss' % (n, t)    print 'end timeit function ', f    print_time(randomchoice)

result

start timeit function  <function randomchoice at 0x2a7d6e0>repeat 1000000 times and used 3.97338795662send timeit function  <function randomchoice at 0x2a7d6e0>

随机分布

from random import choiceimport stringimport collectionsfrom matplotlib.pyplot import plot, show, barh, yticks, xlabel, title, figureimport numpy as nptables = string.ascii_letters + string.digitscounter = collections.Counter()for _ in range(1000000):    counter[choice(tables)] += 1alphats = counter.keys()y_pos = np.arange(len(alphats))freq = counter.values()figure(figsize=(100,100))barh(y_pos, freq, align='edge', alpha=1, height=0.05)yticks(y_pos, alphats)xlabel('frequence')title('random choice')show()

结果图:

可见分布还是比较平均的

3. like mongo objectid

      时间 ＋ md5(hostname) + pid + 递增id

import structimport socketimport osimport timefrom hashlib import md5import threadingimport randomimport binascii_inc = random.randint(0, 0xFFFFFF)_inc_lock = threading.Lock()oid = ""oid += struct.pack(">i", int(time.time()))m = md5()m.update(socket.gethostname())oid += m.digest()[0:3]oid += struct.pack(">H", os.getpid() % 0xFFFF)_inc_lock.acquire()oid += struct.pack(">i", _inc)[1:4]_inc = (_inc + 1) % 0xFFFFFF_inc_lock.release()print len(oid)print binascii.hexlify(oid)