搜索是大数据领域里常见的需求。Splunk和ELK分别是该领域在非开源和开源领域里的领导者。本文利用很少的Python代码实现了一个基本的数据搜索功能,试图让大家理解大数据搜索的基本原理。

布隆过滤器 (Bloom Filter)

第一步我们先要实现一个布隆过滤器

布隆过滤器是大数据领域的一个常见算法,它的目的是过滤掉那些不是目标的元素。也就是说如果一个要搜索的词并不存在与我的数据中,那么它可以以很快的速度返回目标不存在。

让我们看看以下布隆过滤器的代码:

 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
class Bloomfilter(object):
    """
    A Bloom filter is a probabilistic data-structure that trades space for accuracy
    when determining if a value is in a set.  It can tell you if a value was possibly
    added, or if it was definitely not added, but it can't tell you for certain that
    it was added.
    """
    def __init__(self, size):
        """Setup the BF with the appropriate size"""
        self.values = [False] * size
        self.size = size
 
    def hash_value(self, value):
        """Hash the value provided and scale it to fit the BF size"""
        return hash(value) % self.size
 
    def add_value(self, value):
        """Add a value to the BF"""
        h = self.hash_value(value)
        self.values[h] = True
 
    def might_contain(self, value):
        """Check if the value might be in the BF"""
        h = self.hash_value(value)
        return self.values[h]
 
    def print_contents(self):
        """Dump the contents of the BF for debugging purposes"""
        print self.values
  • 基本的数据结构是个数组(实际上是个位图,用1/0来记录数据是否存在),初始化是没有任何内容,所以全部置False。实际的使用当中,该数组的长度是非常大的,以保证效率。
  • 利用哈希算法来决定数据应该存在哪一位,也就是数组的索引
  • 当一个数据被加入到布隆过滤器的时候,计算它的哈希值然后把相应的位置为True
  • 当检查一个数据是否已经存在或者说被索引过的时候,只要检查对应的哈希值所在的位的True/Fasle

看到这里,大家应该可以看出,如果布隆过滤器返回False,那么数据一定是没有索引过的,然而如果返回True,那也不能说数据一定就已经被索引过。在搜索过程中使用布隆过滤器可以使得很多没有命中的搜索提前返回来提高效率。

我们看看这段 code是如何运行的:

 
 
1
2
3
4
5
6
7
8
9
10
bf = Bloomfilter(10)
bf.add_value('dog')
bf.add_value('fish')
bf.add_value('cat')
bf.print_contents()
bf.add_value('bird')
bf.print_contents()
# Note: contents are unchanged after adding bird - it collides
for term in ['dog', 'fish', 'cat', 'bird', 'duck', 'emu']:
    print '{}: {} {}'.format(term, bf.hash_value(term), bf.might_contain(term))

结果:

 
 
1
2
3
4
5
6
7
8
[False, False, False, False, True, True, False, False, False, True]
[False, False, False, False, True, True, False, False, False, True]
dog: 5 True
fish: 4 True
cat: 9 True
bird: 9 True
duck: 5 True
emu: 8 False

首先创建了一个容量为10的的布隆过滤器

然后分别加入 ‘dog’,‘fish’,‘cat’三个对象,这时的布隆过滤器的内容如下:

然后加入‘bird’对象,布隆过滤器的内容并没有改变,因为‘bird’和‘fish’恰好拥有相同的哈希。

最后我们检查一堆对象(’dog’, ‘fish’, ‘cat’, ‘bird’, ‘duck’, ’emu’)是不是已经被索引了。结果发现‘duck’返回True,2而‘emu’返回False。因为‘duck’的哈希恰好和‘dog’是一样的。

分词

下面一步我们要实现分词。 分词的目的是要把我们的文本数据分割成可搜索的最小单元,也就是词。这里我们主要针对英语,因为中文的分词涉及到自然语言处理,比较复杂,而英文基本只要用标点符号就好了。

下面我们看看分词的代码:

 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
def major_segments(s):
    """
    Perform major segmenting on a string.  Split the string by all of the major
    breaks, and return the set of everything found.  The breaks in this implementation
    are single characters, but in Splunk proper they can be multiple characters.
    A set is used because ordering doesn't matter, and duplicates are bad.
    """
    major_breaks = ' '
    last = -1
    results = set()
 
    # enumerate() will give us (0, s[0]), (1, s[1]), ...
    for idx, ch in enumerate(s):
        if ch in major_breaks:
            segment = s[last+1:idx]
            results.add(segment)
 
            last = idx
 
    # The last character may not be a break so always capture
    # the last segment (which may end up being "", but yolo)    
    segment = s[last+1:]
    results.add(segment)
 
    return results

主要分割

主要分割使用空格来分词,实际的分词逻辑中,还会有其它的分隔符。例如Splunk的缺省分割符包括以下这些,用户也可以定义自己的分割符。

] < >( ) { } | ! ; , ‘ ” * \n \r \s \t & ? + %21 %26 %2526 %3B %7C %20 %2B %3D — %2520 %5D %5B %3A %0A %2C %28 %29

 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
def minor_segments(s):
    """
    Perform minor segmenting on a string.  This is like major
    segmenting, except it also captures from the start of the
    input to each break.
    """
    minor_breaks = '_.'
    last = -1
    results = set()
 
    for idx, ch in enumerate(s):
        if ch in minor_breaks:
            segment = s[last+1:idx]
            results.add(segment)
 
            segment = s[:idx]
            results.add(segment)
 
            last = idx
 
    segment = s[last+1:]
    results.add(segment)
    results.add(s)
 
    return results

次要分割

次要分割和主要分割的逻辑类似,只是还会把从开始部分到当前分割的结果加入。例如“1.2.3.4”的次要分割会有1,2,3,4,1.2,1.2.3

 
 
1
2
3
4
5
6
7
def segments(event):
    """Simple wrapper around major_segments / minor_segments"""
    results = set()
    for major in major_segments(event):
        for minor in minor_segments(major):
            results.add(minor)
    return results

分词的逻辑就是对文本先进行主要分割,对每一个主要分割在进行次要分割。然后把所有分出来的词返回。

我们看看这段 code是如何运行的:

 
 
1
2
for term in segments('src_ip = 1.2.3.4'):
        print term
 
 
1
2
3
4
5
6
7
8
9
10
11
src
1.2
1.2.3.4
src_ip
3
1
1.2.3
ip
2
=
4

搜索

好了,有个分词和布隆过滤器这两个利器的支撑后,我们就可以来实现搜索的功能了。

上代码:

 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
class Splunk(object):
    def __init__(self):
        self.bf = Bloomfilter(64)
        self.terms = {}  # Dictionary of term to set of events
        self.events = []
    
    def add_event(self, event):
        """Adds an event to this object"""
 
        # Generate a unique ID for the event, and save it
        event_id = len(self.events)
        self.events.append(event)
 
        # Add each term to the bloomfilter, and track the event by each term
        for term in segments(event):
            self.bf.add_value(term)
 
            if term not in self.terms:
                self.terms[term] = set()
            self.terms[term].add(event_id)
 
    def search(self, term):
        """Search for a single term, and yield all the events that contain it"""
        
        # In Splunk this runs in O(1), and is likely to be in filesystem cache (memory)
        if not self.bf.might_contain(term):
            return
 
        # In Splunk this probably runs in O(log N) where N is the number of terms in the tsidx
        if term not in self.terms:
            return
 
        for event_id in sorted(self.terms[term]):
            yield self.events[event_id]
  • Splunk代表一个拥有搜索功能的索引集合
  • 每一个集合中包含一个布隆过滤器,一个倒排词表(字典),和一个存储所有事件的数组
  • 当一个事件被加入到索引的时候,会做以下的逻辑
    • 为每一个事件生成一个unqie id,这里就是序号
    • 对事件进行分词,把每一个词加入到倒排词表,也就是每一个词对应的事件的id的映射结构,注意,一个词可能对应多个事件,所以倒排表的的值是一个Set。倒排表是绝大部分搜索引擎的核心功能。
  • 当一个词被搜索的时候,会做以下的逻辑
    • 检查布隆过滤器,如果为假,直接返回
    • 检查词表,如果被搜索单词不在词表中,直接返回
    • 在倒排表中找到所有对应的事件id,然后返回事件的内容

我们运行下看看把:

 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
s = Splunk()
s.add_event('src_ip = 1.2.3.4')
s.add_event('src_ip = 5.6.7.8')
s.add_event('dst_ip = 1.2.3.4')
 
for event in s.search('1.2.3.4'):
    print event
print '-'
for event in s.search('src_ip'):
    print event
print '-'
for event in s.search('ip'):
    print event
 
 
1
2
3
4
5
6
7
8
9
src_ip = 1.2.3.4
dst_ip = 1.2.3.4
-
src_ip = 1.2.3.4
src_ip = 5.6.7.8
-
src_ip = 1.2.3.4
src_ip = 5.6.7.8
dst_ip = 1.2.3.4

是不是很赞!

更复杂的搜索

更进一步,在搜索过程中,我们想用And和Or来实现更复杂的搜索逻辑。

上代码:

 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
class SplunkM(object):
    def __init__(self):
        self.bf = Bloomfilter(64)
        self.terms = {}  # Dictionary of term to set of events
        self.events = []
    
    def add_event(self, event):
        """Adds an event to this object"""
 
        # Generate a unique ID for the event, and save it
        event_id = len(self.events)
        self.events.append(event)
 
        # Add each term to the bloomfilter, and track the event by each term
        for term in segments(event):
            self.bf.add_value(term)
            if term not in self.terms:
                self.terms[term] = set()
            
            self.terms[term].add(event_id)
 
    def search_all(self, terms):
        """Search for an AND of all terms"""
 
        # Start with the universe of all events...
        results = set(range(len(self.events)))
 
        for term in terms:
            # If a term isn't present at all then we can stop looking
            if not self.bf.might_contain(term):
                return
            if term not in self.terms:
                return
 
            # Drop events that don't match from our results
            results = results.intersection(self.terms[term])
 
        for event_id in sorted(results):
            yield self.events[event_id]
 
 
    def search_any(self, terms):
        """Search for an OR of all terms"""
        results = set()
 
        for term in terms:
            # If a term isn't present, we skip it, but don't stop
            if not self.bf.might_contain(term):
                continue
            if term not in self.terms:
                continue
 
            # Add these events to our results
            results = results.union(self.terms[term])
 
        for event_id in sorted(results):
            yield self.events[event_id]

利用Python集合的intersection和union操作,可以很方便的支持And(求交集)和Or(求合集)的操作。

运行结果如下:

 
 
1
2
3
4
5
6
7
8
9
10
s = SplunkM()
s.add_event('src_ip = 1.2.3.4')
s.add_event('src_ip = 5.6.7.8')
s.add_event('dst_ip = 1.2.3.4')
 
for event in s.search_all(['src_ip', '5.6']):
    print event
print '-'
for event in s.search_any(['src_ip', 'dst_ip']):
    print event
 
 
1
2
3
4
5
src_ip = 5.6.7.8
-
src_ip = 1.2.3.4
src_ip = 5.6.7.8
dst_ip = 1.2.3.4

总结

以上的代码只是为了说明大数据搜索的基本原理,包括布隆过滤器,分词和倒排表。如果大家真的想要利用这代码来实现真正的搜索功能,还差的太远

用 Python 实现一个大数据搜索引擎的更多相关文章

  1. 用Python实现一个大数据搜索引擎

    用Python实现一个大数据搜索引擎 搜索是大数据领域里常见的需求.Splunk和ELK分别是该领域在非开源和开源领域里的领导者.本文利用很少的Python代码实现了一个基本的数据搜索功能,试图让大家 ...

  2. 如何基于Go搭建一个大数据平台

    如何基于Go搭建一个大数据平台 - Go中国 - CSDN博客 https://blog.csdn.net/ra681t58cjxsgckj31/article/details/78333775 01 ...

  3. 零起点PYTHON足彩大数据与机器学习实盘分析

    零起点PYTHON足彩大数据与机器学习实盘分析 第1章 足彩与数据分析 1 1.1 “阿尔法狗”与足彩 1 1.2 案例1-1:可怕的英国足球 3 1.3 关于足彩的几个误区 7 1.4 足彩·大事件 ...

  4. 我的ElasticSearch集群部署总结--大数据搜索引擎你不得不知

    摘要:世上有三类书籍:1.介绍知识,2.阐述理论,3.工具书:世间也存在两类知识:1.技术,2.思想.以下是我在部署ElasticSearch集群时的经验总结,它们大体属于第一类知识“techknow ...

  5. 如何做一个大数据seo人员

    作为流量运营者或者SEO人员,对于所从事行业领域的认识往往建立在一种直觉之上,我们很难对一个行业有一个全面的了解,这个行业领域有多宽,流量聚焦在哪里,那些是用户最关心的问题? 有的时候很难准确的把握, ...

  6. 一个大数据平台省了20个IT人力——敦奴数据平台建设案例分享

    认识敦奴 敦奴集团创立于1987年,主营服装.酒店.地产,总部位于中国皮都-海宁.浙江敦奴联合实业股份有限公司(以下简称"敦奴")是一家集开发.设计.生产.销售于一体的大型专业服装 ...

  7. 一个大数据的demo

    package test.admin; import java.io.BufferedReader; import java.io.BufferedWriter; import java.io.Fil ...

  8. 大数据搜索引擎之elasticsearch使用篇(一)

    作者:yanzm 原文来自:https://bbs.ichunqiu.com/thread-42421-1-1.html 1.基础介绍 本期,我们将着重介绍elasticsearch的基本使用方法. ...

  9. 一个大数据方案:基于Nutch+Hadoop+Hbase+ElasticSearch的网络爬虫及搜索引擎

    网络爬虫架构在Nutch+Hadoop之上,是一个典型的分布式离线批量处理架构,有非常优异的吞吐量和抓取性能并提供了大量的配置定制选项.由于网络爬虫只负责网络资源的抓取,所以,需要一个分布式搜索引擎, ...

随机推荐

  1. CentOS 6 / RHEL 6配置bonding 4模式

    实现bond 802.3ad or 4 模式:(IEEE 802.3ad), 方式:创建一个整合的组,这个组会共享网速和网络双工(duplex)设置.模式 4 会根据 IEEE 802.3ad 标准使 ...

  2. php empty()与isset()

    empty() : 检查一个变量是否为空. 判断一个变量是否被认为是空的.当一个变量并不存在,或者它的值等同于FALSE,那么它会被认为不存在.如果变量不存在的话,empty()并不会产生警告. em ...

  3. 【译】第十篇 Integration Services:高级事件行为

    本篇文章是Integration Services系列的第十篇,详细内容请参考原文. 简介在前一篇, we introduced fault tolerance by examining method ...

  4. Chrome 清除某个特定网站下的缓存

    打开开发者工具(F12),选择 Network--Disable cache 即可.需要清除某网站缓存时 F12 打开开发者工具就会自动清除这个网站的缓存,而不必清除所有网站的缓存了.

  5. mysql中使用日期加减时无法识别年-月格式数据的问题,%Y-%m"这种格式数据

    最新做报表统计的时候处理按月统计部分时发现,虽然使用 DATE_FORMAT( time, '%Y-%m' ) 函数可以将日期格式转成年-月,但是如果是参数是年-月格式,即"2018-10& ...

  6. 读sru代码

    1. def read_corpus(path, eos="</s>"): data = [ ] with open(path) as fin: for line in ...

  7. python的新特性

    http://www.jb51.net/article/67699.htm Python中用format函数格式化字符串的用法:http://www.jb51.net/article/63672.ht ...

  8. Ubuntu使用apt-get upgrade升级时出错

    今天在按照常规的sudo apt-get update更新软件列表后,再使用sudo apt-get upgrade升级软件时,出现了以下的错误: 正在设置 linux-image-extra-4.4 ...

  9. Android页面之间进行数据回传

    要求:页面1跳转到页面2,页面2再返回页面1同时返回数据 页面1添加如下代码: Intent intent = new Intent(); intent.setClass(页面1.this, 页面2. ...

  10. vue总结05 过渡--状态过渡

    状态过渡 Vue 的过渡系统提供了非常多简单的方法设置进入.离开和列表的动效.那么对于数据元素本身的动效呢,比如: 数字和运算 颜色的显示 SVG 节点的位置 元素的大小和其他的属性 所有的原始数字都 ...