Python 用于快速多维数据查找的数据模型和数据存储技术

我有一个

parent

hashmap数据结构，其中字符串作为键，hashmap数据结构作为子结构（guess

child1

，

child2

，…，

childN

）。每个子元素都是一个简单的键值映射，以数字作为键，以字符串作为值。 在伪代码中：

parent['key1'] = child1;    // child1 is a hash map data structure
child1[0] = 'foo';
child[1] = 'bar';
...

我需要将此数据结构实现为数据库系统中的快速查找表。 让我们以Python作为参考语言

解决方案的要求：

尽可能快地检索儿童hasmaps

父散列的估计总重量最多为500 MB

用例如下所示：

客户端Python程序查询数据存储中的特定子哈希

数据存储返回子散列

Python程序将整个散列传递给特定函数，从散列中提取特定值（它已经知道要使用哪个键），然后将其传递给第二个函数

您会推荐内存中的键值数据存储（如Redis）还是更经典的“关系”数据库解决方案？你建议我使用哪种数据模型？

绝对可以。它不仅速度非常快，而且可以精确地处理您需要的结构：

在您的情况下，您可以避免读取整个“子哈希”，因为客户端“从哈希中提取特定值（它已经知道要使用哪个键）”

或者，如果您确实想要整个散列：

redis> HGETALL myhash
1) "field1"
2) "Hello"
3) "field2"
4) "World"
redis>

当然，在一个可行的对象（在您的例子中是一个Python字典）中使用a会给出正确的结果。

绝对可以。它不仅速度非常快，而且可以精确地处理您需要的结构：

在您的情况下，您可以避免读取整个“子哈希”，因为客户端“从哈希中提取特定值（它已经知道要使用哪个键）”

或者，如果您确实想要整个散列：

redis> HGETALL myhash
1) "field1"
2) "Hello"
3) "field2"
4) "World"
redis>

---

当然，使用a可以在一个可行的对象（在您的例子中是Python字典）中给出正确的结果。

在基于提示的快速搜索之后，我想出了这个解决方案：我可以在Redis中实现一个单一的

父项

哈希，其中值字段将是子哈希的字符串表示形式。通过这种方式，我可以从Python程序中快速读取并评估它们

举个例子，我的Redis数据结构类似于：

//write a hash with N key-value pairs: each value is an M key-value pairs hash
redis> HMSET parent_key1 child_hash "c1k1:c1v1, c1k2:c1v2, [...], c1kM:c1vM"
  OK
redis> HMSET parent_key2 child_hash "c2k1:c2v1, c2k2:c2v2, [...], c2kM:c2vM"
  OK
[...]
redis> HMSET parent_keyN child_hash "cNk1:cNv1, cNk2:cNv2, [...], cNkM:cNvM"
  OK

//read data
redis> HGET parent_key1 child_hash
  "c1k1:c1v1, c1k2:c1v2, [...], c1kM:c1vM"

然后，我的Python代码只需要使用Redis绑定来查询所需的子哈希，并返回它们的实际字符串表示形式；剩下要做的就是将字符串表示形式转换为相应的字典，因此可以方便地查找字典

示例代码（如中所建议）：

---

希望我没有错过任何东西

在基于提示的快速搜索之后，我想到了这个解决方案：我可以在Redis中实现一个

父类
散列，其中值字段将是子散列的字符串表示形式。通过这种方式，我可以从Python程序中快速读取并评估它们

举个例子，我的Redis数据结构类似于：

//write a hash with N key-value pairs: each value is an M key-value pairs hash
redis> HMSET parent_key1 child_hash "c1k1:c1v1, c1k2:c1v2, [...], c1kM:c1vM"
  OK
redis> HMSET parent_key2 child_hash "c2k1:c2v1, c2k2:c2v2, [...], c2kM:c2vM"
  OK
[...]
redis> HMSET parent_keyN child_hash "cNk1:cNv1, cNk2:cNv2, [...], cNkM:cNvM"
  OK

//read data
redis> HGET parent_key1 child_hash
  "c1k1:c1v1, c1k2:c1v2, [...], c1kM:c1vM"

然后，我的Python代码只需要使用Redis绑定来查询所需的子哈希，并返回它们的实际字符串表示形式；剩下要做的就是将字符串表示形式转换为相应的字典，因此可以方便地查找字典

示例代码（如中所建议）：

希望我没有错过任何东西
 使用示例代码，假设已经安装了Redis（理想情况下），将每个父项保存为哈希字段，子项保存为序列化字符串，并在客户端处理序列化和反序列化：

JSON版本：

## JSON version
import json 
# you could use pickle instead, 
# just replace json.dumps/json.loads with pickle/unpickle

import redis

# set up the redis client
r = redis.StrictRedis(host = '', port = 6379, db = 0)

# sample parent dicts
parent0 = {'child0': {0:'a', 1:'b', 2:'c',}, 'child1':{5:'e', 6:'f', 7:'g'}}
parent1 = {'child0': {0:'h', 1:'i', 2:'j',}, 'child1':{5:'k', 6:'l', 7:'m'}}

# save the parents as hashfields, with the children as serialized strings
# bear in mind that JSON will convert the int keys to strings in the dumps() process
r.hmset('parent0', {key: json.dumps(parent0[key]) for key in parent0})
r.hmset('parent1', {key: json.dumps(parent0[key]) for key in parent1})


# Get a child dict from a parent
# say child1 of parent0
childstring = r.hget('parent0', 'child1') 
childdict = json.loads(childstring) 
# this could have been done in a single line... 

# if you want to convert the keys back to ints:
for key in childdict.keys():
    childdict[int(key)] = childdict[key]
    del childdict[key]

print childdict

## pickle version
# For pickle, you need a file-like object. 
# StringIO is the native python one, whie cStringIO 
# is the c implementation of the same.
# cStringIO is faster
# see http://docs.python.org/library/stringio.html and
# http://www.doughellmann.com/PyMOTW/StringIO/ for more information
import pickle
# Find the best implementation available on this platform
try:
    from cStringIO import StringIO
except:
    from StringIO import StringIO

import redis

# set up the redis client
r = redis.StrictRedis(host = '', port = 6379, db = 0)

# sample parent dicts
parent0 = {'child0': {0:'a', 1:'b', 2:'c',}, 'child1':{5:'e', 6:'f', 7:'g'}}
parent1 = {'child0': {0:'h', 1:'i', 2:'j',}, 'child1':{5:'k', 6:'l', 7:'m'}}

# define a class with a reusable StringIO object
class Pickler(object):
    """Simple helper class to use pickle with a reusable string buffer object"""
    def __init__(self):
        self.tmpstr = StringIO()

    def __del__(self):
        # close the StringIO buffer and delete it
        self.tmpstr.close()
        del self.tmpstr

    def dump(self, obj):
        """Pickle an object and return the pickled string"""
        # empty current buffer
        self.tmpstr.seek(0,0)
        self.tmpstr.truncate(0)
        # pickle obj into the buffer
        pickle.dump(obj, self.tmpstr)
        # move the buffer pointer to the start
        self.tmpstr.seek(0,0)
        # return the pickled buffer as a string
        return self.tmpstr.read()

    def load(self, obj):
        """load a pickled object string and return the object"""
        # empty the current buffer
        self.tmpstr.seek(0,0)
        self.tmpstr.truncate(0)
        # load the pickled obj string into the buffer
        self.tmpstr.write(obj)
        # move the buffer pointer to start
        self.tmpstr.seek(0,0)
        # load the pickled buffer into an object
        return pickle.load(self.tmpstr)


pickler = Pickler()

# save the parents as hashfields, with the children as pickled strings, 
# pickled using our helper class
r.hmset('parent0', {key: pickler.dump(parent0[key]) for key in parent0})
r.hmset('parent1', {key: pickler.dump(parent1[key]) for key in parent1})


# Get a child dict from a parent
# say child1 of parent0
childstring = r.hget('parent0', 'child1') 
# this could be done in a single line... 
childdict = pickler.load(childstring) 

# we don't need to do any str to int conversion on the keys.

print childdict

pickle版本：

## JSON version
import json 
# you could use pickle instead, 
# just replace json.dumps/json.loads with pickle/unpickle

import redis

# set up the redis client
r = redis.StrictRedis(host = '', port = 6379, db = 0)

# sample parent dicts
parent0 = {'child0': {0:'a', 1:'b', 2:'c',}, 'child1':{5:'e', 6:'f', 7:'g'}}
parent1 = {'child0': {0:'h', 1:'i', 2:'j',}, 'child1':{5:'k', 6:'l', 7:'m'}}

# save the parents as hashfields, with the children as serialized strings
# bear in mind that JSON will convert the int keys to strings in the dumps() process
r.hmset('parent0', {key: json.dumps(parent0[key]) for key in parent0})
r.hmset('parent1', {key: json.dumps(parent0[key]) for key in parent1})


# Get a child dict from a parent
# say child1 of parent0
childstring = r.hget('parent0', 'child1') 
childdict = json.loads(childstring) 
# this could have been done in a single line... 

# if you want to convert the keys back to ints:
for key in childdict.keys():
    childdict[int(key)] = childdict[key]
    del childdict[key]

print childdict

## pickle version
# For pickle, you need a file-like object. 
# StringIO is the native python one, whie cStringIO 
# is the c implementation of the same.
# cStringIO is faster
# see http://docs.python.org/library/stringio.html and
# http://www.doughellmann.com/PyMOTW/StringIO/ for more information
import pickle
# Find the best implementation available on this platform
try:
    from cStringIO import StringIO
except:
    from StringIO import StringIO

import redis

# set up the redis client
r = redis.StrictRedis(host = '', port = 6379, db = 0)

# sample parent dicts
parent0 = {'child0': {0:'a', 1:'b', 2:'c',}, 'child1':{5:'e', 6:'f', 7:'g'}}
parent1 = {'child0': {0:'h', 1:'i', 2:'j',}, 'child1':{5:'k', 6:'l', 7:'m'}}

# define a class with a reusable StringIO object
class Pickler(object):
    """Simple helper class to use pickle with a reusable string buffer object"""
    def __init__(self):
        self.tmpstr = StringIO()

    def __del__(self):
        # close the StringIO buffer and delete it
        self.tmpstr.close()
        del self.tmpstr

    def dump(self, obj):
        """Pickle an object and return the pickled string"""
        # empty current buffer
        self.tmpstr.seek(0,0)
        self.tmpstr.truncate(0)
        # pickle obj into the buffer
        pickle.dump(obj, self.tmpstr)
        # move the buffer pointer to the start
        self.tmpstr.seek(0,0)
        # return the pickled buffer as a string
        return self.tmpstr.read()

    def load(self, obj):
        """load a pickled object string and return the object"""
        # empty the current buffer
        self.tmpstr.seek(0,0)
        self.tmpstr.truncate(0)
        # load the pickled obj string into the buffer
        self.tmpstr.write(obj)
        # move the buffer pointer to start
        self.tmpstr.seek(0,0)
        # load the pickled buffer into an object
        return pickle.load(self.tmpstr)


pickler = Pickler()

# save the parents as hashfields, with the children as pickled strings, 
# pickled using our helper class
r.hmset('parent0', {key: pickler.dump(parent0[key]) for key in parent0})
r.hmset('parent1', {key: pickler.dump(parent1[key]) for key in parent1})


# Get a child dict from a parent
# say child1 of parent0
childstring = r.hget('parent0', 'child1') 
# this could be done in a single line... 
childdict = pickler.load(childstring) 

# we don't need to do any str to int conversion on the keys.

print childdict

示例代码使用，假设已经安装了Redis（理想情况下），将每个父项保存为哈希字段，子项保存为序列化字符串，并在客户端处理序列化和反序列化：

JSON版本：

## JSON version
import json 
# you could use pickle instead, 
# just replace json.dumps/json.loads with pickle/unpickle

import redis

# set up the redis client
r = redis.StrictRedis(host = '', port = 6379, db = 0)

# sample parent dicts
parent0 = {'child0': {0:'a', 1:'b', 2:'c',}, 'child1':{5:'e', 6:'f', 7:'g'}}
parent1 = {'child0': {0:'h', 1:'i', 2:'j',}, 'child1':{5:'k', 6:'l', 7:'m'}}

# save the parents as hashfields, with the children as serialized strings
# bear in mind that JSON will convert the int keys to strings in the dumps() process
r.hmset('parent0', {key: json.dumps(parent0[key]) for key in parent0})
r.hmset('parent1', {key: json.dumps(parent0[key]) for key in parent1})


# Get a child dict from a parent
# say child1 of parent0
childstring = r.hget('parent0', 'child1') 
childdict = json.loads(childstring) 
# this could have been done in a single line... 

# if you want to convert the keys back to ints:
for key in childdict.keys():
    childdict[int(key)] = childdict[key]
    del childdict[key]

print childdict

## pickle version
# For pickle, you need a file-like object. 
# StringIO is the native python one, whie cStringIO 
# is the c implementation of the same.
# cStringIO is faster
# see http://docs.python.org/library/stringio.html and
# http://www.doughellmann.com/PyMOTW/StringIO/ for more information
import pickle
# Find the best implementation available on this platform
try:
    from cStringIO import StringIO
except:
    from StringIO import StringIO

import redis

# set up the redis client
r = redis.StrictRedis(host = '', port = 6379, db = 0)

# sample parent dicts
parent0 = {'child0': {0:'a', 1:'b', 2:'c',}, 'child1':{5:'e', 6:'f', 7:'g'}}
parent1 = {'child0': {0:'h', 1:'i', 2:'j',}, 'child1':{5:'k', 6:'l', 7:'m'}}

# define a class with a reusable StringIO object
class Pickler(object):
    """Simple helper class to use pickle with a reusable string buffer object"""
    def __init__(self):
        self.tmpstr = StringIO()

    def __del__(self):
        # close the StringIO buffer and delete it
        self.tmpstr.close()
        del self.tmpstr

    def dump(self, obj):
        """Pickle an object and return the pickled string"""
        # empty current buffer
        self.tmpstr.seek(0,0)
        self.tmpstr.truncate(0)
        # pickle obj into the buffer
        pickle.dump(obj, self.tmpstr)
        # move the buffer pointer to the start
        self.tmpstr.seek(0,0)
        # return the pickled buffer as a string
        return self.tmpstr.read()

    def load(self, obj):
        """load a pickled object string and return the object"""
        # empty the current buffer
        self.tmpstr.seek(0,0)
        self.tmpstr.truncate(0)
        # load the pickled obj string into the buffer
        self.tmpstr.write(obj)
        # move the buffer pointer to start
        self.tmpstr.seek(0,0)
        # load the pickled buffer into an object
        return pickle.load(self.tmpstr)


pickler = Pickler()

# save the parents as hashfields, with the children as pickled strings, 
# pickled using our helper class
r.hmset('parent0', {key: pickler.dump(parent0[key]) for key in parent0})
r.hmset('parent1', {key: pickler.dump(parent1[key]) for key in parent1})


# Get a child dict from a parent
# say child1 of parent0
childstring = r.hget('parent0', 'child1') 
# this could be done in a single line... 
childdict = pickler.load(childstring) 

# we don't need to do any str to int conversion on the keys.

print childdict

pickle版本：

## JSON version
import json 
# you could use pickle instead, 
# just replace json.dumps/json.loads with pickle/unpickle

import redis

# set up the redis client
r = redis.StrictRedis(host = '', port = 6379, db = 0)

# sample parent dicts
parent0 = {'child0': {0:'a', 1:'b', 2:'c',}, 'child1':{5:'e', 6:'f', 7:'g'}}
parent1 = {'child0': {0:'h', 1:'i', 2:'j',}, 'child1':{5:'k', 6:'l', 7:'m'}}

# save the parents as hashfields, with the children as serialized strings
# bear in mind that JSON will convert the int keys to strings in the dumps() process
r.hmset('parent0', {key: json.dumps(parent0[key]) for key in parent0})
r.hmset('parent1', {key: json.dumps(parent0[key]) for key in parent1})


# Get a child dict from a parent
# say child1 of parent0
childstring = r.hget('parent0', 'child1') 
childdict = json.loads(childstring) 
# this could have been done in a single line... 

# if you want to convert the keys back to ints:
for key in childdict.keys():
    childdict[int(key)] = childdict[key]
    del childdict[key]

print childdict

## pickle version
# For pickle, you need a file-like object. 
# StringIO is the native python one, whie cStringIO 
# is the c implementation of the same.
# cStringIO is faster
# see http://docs.python.org/library/stringio.html and
# http://www.doughellmann.com/PyMOTW/StringIO/ for more information
import pickle
# Find the best implementation available on this platform
try:
    from cStringIO import StringIO
except:
    from StringIO import StringIO

import redis

# set up the redis client
r = redis.StrictRedis(host = '', port = 6379, db = 0)

# sample parent dicts
parent0 = {'child0': {0:'a', 1:'b', 2:'c',}, 'child1':{5:'e', 6:'f', 7:'g'}}
parent1 = {'child0': {0:'h', 1:'i', 2:'j',}, 'child1':{5:'k', 6:'l', 7:'m'}}

# define a class with a reusable StringIO object
class Pickler(object):
    """Simple helper class to use pickle with a reusable string buffer object"""
    def __init__(self):
        self.tmpstr = StringIO()

    def __del__(self):
        # close the StringIO buffer and delete it
        self.tmpstr.close()
        del self.tmpstr

    def dump(self, obj):
        """Pickle an object and return the pickled string"""
        # empty current buffer
        self.tmpstr.seek(0,0)
        self.tmpstr.truncate(0)
        # pickle obj into the buffer
        pickle.dump(obj, self.tmpstr)
        # move the buffer pointer to the start
        self.tmpstr.seek(0,0)
        # return the pickled buffer as a string
        return self.tmpstr.read()

    def load(self, obj):
        """load a pickled object string and return the object"""
        # empty the current buffer
        self.tmpstr.seek(0,0)
        self.tmpstr.truncate(0)
        # load the pickled obj string into the buffer
        self.tmpstr.write(obj)
        # move the buffer pointer to start
        self.tmpstr.seek(0,0)
        # load the pickled buffer into an object
        return pickle.load(self.tmpstr)


pickler = Pickler()

# save the parents as hashfields, with the children as pickled strings, 
# pickled using our helper class
r.hmset('parent0', {key: pickler.dump(parent0[key]) for key in parent0})
r.hmset('parent1', {key: pickler.dump(parent1[key]) for key in parent1})


# Get a child dict from a parent
# say child1 of parent0
childstring = r.hget('parent0', 'child1') 
# this could be done in a single line... 
childdict = pickler.load(childstring) 

# we don't need to do any str to int conversion on the keys.

print childdict

谢谢你的回答。我对用例的第3步描述得很糟糕（现在已经修复了），事实上，我希望整个子散列由Python代码读取和处理。Redis仍然适合吗？Redis仍然很好，但将子哈希保存为序列化/pickle字符串，并在读取后在客户端上反序列化/取消pickle会更快。另外，请确保为您各自的客户端（即）安装了
hiredis
，这样可以更快地将数据从redis转换为python，反之亦然。谢谢您的回答。我对用例的第3步描述得很糟糕（现在已经修复了），事实上，我希望整个子散列由Python代码读取和处理。Redis仍然适合吗？Redis仍然很好，但将子哈希保存为序列化/pickle字符串，并在读取后在客户端上反序列化/取消pickle会更快。另外，请确保为您各自的客户端（即）安装
hiredis
，这样可以更快地将数据从redis转换为python，反之亦然。好的，您可以使用此对象编组来存储字符串，但如果您希望加快速度，则应将每个python哈希存储在redis哈希上。当需要多个键空间时，通常的习惯用法是连接键：
HMSET“parentkey1:childkeyX”f1 v1 f2 v2 f3 v3
。这允许Redis使用ziplists优化小散列（对于不到100个左右的字段来说更紧凑、更快），或者，如果您真的想将对象存储为字符串，可以使用pickle，因为Redis字符串是8位干净的。但是如果你想做一些服务器端处理，那么考虑JSON或MeasAgPACK，因为两者都可以被嵌入式Lua引擎解码。当然，如果你不把你的对象串起来（如前一条评论所述），那就更容易了。最快的检索