Elasticsearch 深入4

it2026-10-08  3

将一个field索引两次来解决字符串排序

 

如果对一个string field进行排序,结果往往不准确,因为分词后是多个单词,再排序就不是我们想要的结果了

通常解决方案是,将一个string field建立两次索引,一个分词,用来进行搜索;一个不分词,用来进行排序

PUT /website {    "mappings":{        "article":{            "properties":{                "title":{                    "type":"text", 第一次索引进行分词                    "fields":{ 第二次索引不进行分词                        "raw":{                            "type":"string",                            "index":"not_analyzed"                        }                    },                    "fielddata":true 正排索引                },                "content":{                    "type":"text"                },                "post_date":{                    "type":"date"                },                "author_id":{                    "type":"long"                }            }        }    }}

 

GET /website/article/_search{    "query":{        "match_all":{        }    },    "sort":[        {            "title.raw":{   //如果直接使用title的话是对分词之后的结果排序可能存在问题 title.raw 使用不分词的索引进行排序                "order":"desc"            }        }    ]}

 相关度评分TF&IDF算法独家解密

 

1、算法介绍

relevance score算法,简单来说,就是计算出,一个索引中的文本,与搜索文本,他们之间的关联匹配程度

Elasticsearch使用的是 term frequency/inverse document frequency算法,简称为TF/IDF算法

Term frequency:搜索文本中的各个词条在field文本中出现了多少次,出现次数越多,就越相关

搜索请求:hello world

doc1:hello you, and world is very gooddoc2:hello, how are you

Inverse document frequency:搜索文本中的各个词条在整个索引的所有文档中出现了多少次,出现的次数越多,就越不相关

搜索请求:hello world

doc1:hello, today is very gooddoc2:hi world, how are you

比如说,在index中有1万条document,hello这个单词在所有的document中,一共出现了1000次;world这个单词在所有的document中,一共出现了100次

doc2更相关

Field-length norm:field长度,field越长,相关度越弱

搜索请求:hello world

doc1:{ "title": "hello article", "content": "babaaba 1万个单词" }doc2:{ "title": "my article", "content": "blablabala 1万个单词,hi world" }

hello world在整个index中出现的次数是一样多的

doc1更相关,title field更短

GET /people/man/111/_explain

GET /people/man/_search?explain

{    "query":{        "match":{            "name":"ajax"        }    }}

内核级知识点之doc value初步探秘

搜索的时候,要依靠倒排索引;排序的时候,需要依靠正排索引,看到每个document的每个field,然后进行排序,所谓的正排索引,其实就是doc values

在建立索引的时候,一方面会建立倒排索引,以供搜索用;一方面会建立正排索引,也就是doc values,以供排序,聚合,过滤等操作使用

doc values是被保存在磁盘上的,此时如果内存足够,os会自动将其缓存在内存中,性能还是会很高;如果内存不足够,os会将其写入磁盘上

 doc1: hello world you and me

 doc2: hi, world, how are you

 word                     doc1                      doc2

 hello                      *

 world                    *                             *

 you                        *                             *

 and                        *

 me                         *

 hi                                                            *

 how                                                       *

 are                                                         *

 hello you --> hello, you

hello 匹配--> doc1

 you 匹配--> doc1,doc2

 doc1: hello world you and me

doc2: hi, world, how are you

 sort by age

 doc1: { "name": "jack", "age": 27 }

 doc2: { "name": "tom", "age": 30 }

 document           name                    age

 doc1                      jack                        27

 doc2                      tom                        30          

 

 分布式搜索引擎内核解密之query phase

 

1、query phase

(1)搜索请求发送到某一个coordinate node,构构建一个priority queue,长度以paging操作from和size为准,默认为10(2)coordinate node将请求转发到所有shard,每个shard本地搜索,并构建一个本地的priority queue(3)各个shard将自己的priority queue返回给coordinate node,并构建一个全局的priority queue

2、replica shard如何提升搜索吞吐量

一次请求要打到所有shard的一个replica/primary上去,如果每个shard都有多个replica,那么同时并发过来的搜索请求可以同时打到其他的replica上去

 

1、fetch phbase工作流程

 

(1)coordinate node构建完priority queue之后,就发送mget请求去所有shard上获取对应的document(2)各个shard将document返回给coordinate node(3)coordinate node将合并后的document结果返回给client客户端

 

2、一般搜索,如果不加from和size,就默认搜索前10条,按照_score排序

 

 

实战基于scoll技术滚动搜索大量数据

 

如果一次性要查出来比如10万条数据,那么性能会很差,此时一般会采取用scoll滚动查询,一批一批的查,直到所有数据都查询完处理完

使用scoll滚动搜索,可以先搜索一批数据,然后下次再搜索一批数据,以此类推,直到搜索出全部的数据来scoll搜索会在第一次搜索的时候,保存一个当时的视图快照,之后只会基于该旧的视图快照提供数据搜索,如果这个期间数据变更,是不会让用户看到的采用基于_doc进行排序的方式,性能较高每次发送scroll请求,我们还需要指定一个scoll参数,指定一个时间窗口,每次搜索请求只要在这个时间窗口内能完成就可以了

GET /people/man/_search?scroll=1m{ "query": { "match_all": {}}, "sort" : ["_doc"], "size": 3}

GET /_search/scroll{ "scroll": "1m", "scroll_id": "DnF1ZXJ5VGhlbkZldGNoBQAAAAAAACh0FmNIYTh2a2k1UV82RzlYajVaWlFrWncAAAAAAAAocBZOZ21zeV9DTVR2cXdmSnltTF9tRkxRAAAAAAAAKHEWTmdtc3lfQ01UdnF3Zkp5bUxfbUZMUQAAAAAAAChyFk5nbXN5X0NNVHZxd2ZKeW1MX21GTFEAAAAAAAAocxZjSGE4dmtpNVFfNkc5WGo1WlpRa1p3"}

 

获得的结果会有一个scoll_id,下一次再发送scoll请求的时候,必须带上这个scoll_id

scoll,看起来挺像分页的,但是其实使用场景不一样。分页主要是用来一页一页搜索,给用户看的;scoll主要是用来一批一批检索数据,让系统进行处理的

定制化自己的dynamic mapping策略

1、定制dynamic策略

true:遇到陌生字段,就进行dynamic mappingfalse:遇到陌生字段,就忽略strict:遇到陌生字段,就报错

PUT /my_index{    "mappings":{        "my_type":{            "dynamic":"strict",            "properties":{                "title":{                    "type":"text"                },                "address":{                    "type":"object",                    "dynamic":"true"                }            }        }    }}

PUT /my_index/my_type/1{    "title":"my article",    "content":"this is my article",    "address":{        "province":"guangdong",        "city":"guangzhou"    }}

{ "error": { "root_cause": [ { "type": "strict_dynamic_mapping_exception", "reason": "mapping set to strict, dynamic introduction of [content] within [my_type] is not allowed" } ], "type": "strict_dynamic_mapping_exception", "reason": "mapping set to strict, dynamic introduction of [content] within [my_type] is not allowed" }, "status": 400}报错意思就是这个index第一层全局不能再添加新的field

PUT /my_index/my_type/1{    "title":"my article",    "address":{        "province":"guangdong",        "city":"guangzhou"    }}

GET /my_index/_mapping/my_type

{    "my_index":{        "mappings":{            "my_type":{                "dynamic":"strict",                "properties":{                    "address":{                        "dynamic":"true",                        "properties":{                            "city":{                                "type":"text",                                "fields":{                                    "keyword":{                                        "type":"keyword",                                        "ignore_above":256                                    }                                }                            },                            "province":{                                "type":"text",                                "fields":{                                    "keyword":{                                        "type":"keyword",                                        "ignore_above":256                                    }                                }                            }                        }                    },                    "title":{                        "type":"text"                    }                }            }        }    }}但是address字段内部"dynamic":"true"是可以添加字段的

 

2、定制dynamic mapping策略

(1)date_detection

默认会按照一定格式识别date,比如yyyy-MM-dd。但是如果某个field先过来一个2017-01-01的值,就会被自动dynamic mapping成date,后面如果再来一个"hello world"之类的值,就会报错。可以手动关闭某个type的date_detection,如果有需要,自己手动指定某个field为date类型。

PUT /my_index/_mapping/my_type{ "date_detection": false}

(2)定制自己的dynamic mapping template(type level)

PUT /my_index{    "mappings":{        "my_type":{            "dynamic_templates":[                {                    "en":{                        "match":"*_en",                        "match_mapping_type":"string",                        "mapping":{                            "type":"string",                            "analyzer":"english"                        }                    }                }            ]        }    }}

PUT /my_index/my_type/1{ "title": "this is my first article"}

PUT /my_index/my_type/2{ "title_en": "this is my first article"}

title没有匹配到任何的dynamic模板,默认就是standard分词器,不会过滤停用词,is会进入倒排索引,用is来搜索是可以搜索到的title_en匹配到了dynamic模板,就是english分词器,会过滤停用词,is这种停用词就会被过滤掉,用is来搜索就搜索不到了

 

转载于:https://www.cnblogs.com/jiahaoJAVA/p/11015026.html

相关资源:深入理解ElasticSearch PDF
最新回复(0)