Elasticsearch7.6学习笔记1 Getting start with Elasticsearch
- 2020 年 4 月 10 日
- 筆記
Elasticsearch7.6学习笔记1 Getting start with Elasticsearch
前言
权威指南中文只有2.x, 但现在es已经到7.6. 就安装最新的来学下.
安装
这里是学习安装, 生产安装是另一套逻辑.
win
es下载地址:
https://artifacts.elastic.co/downloads/elasticsearch/elasticsearch-7.6.0-windows-x86_64.zip kibana下载地址:
https://artifacts.elastic.co/downloads/kibana/kibana-7.6.0-windows-x86_64.zip 官方目前最新是7.6.0, 但下载速度惨不忍睹. 使用迅雷下载速度可以到xM.
binelasticsearch.bat binkibana.bat 双击bat启动.
docker安装
对于测试学习，直接使用官方提供的docker镜像更快更方便。
安装方法见： https://www.cnblogs.com/woshimrf/p/docker-es7.html
以下内容来自:
https://www.elastic.co/guide/en/elasticsearch/reference/7.6/getting-started.html
Index some documents 索引一些文档
本次测试直接使用kibana, 当然也可以通过curl或者postman访问localhost:9200.
访问localhost:5601, 然后点击Dev Tools.
新建一个客户索引(index)
PUT /{index-name}/_doc/{id}
PUT /customer/_doc/1 { "name": "John Doe" } put 是http method, 如果es中不存在索引(index) customer, 则创建一个, 并插入一个数据, id为, name=John`.
如果存在则更新. 注意, 更新是覆盖更新, 即body json是什么, 最终结果就是什么.
返回如下:
{ "_index" : "customer", "_type" : "_doc", "_id" : "1", "_version" : 7, "result" : "updated", "_shards" : { "total" : 2, "successful" : 2, "failed" : 0 }, "_seq_no" : 6, "_primary_term" : 1 } _index是索引名称_type唯一为_doc_id是文档(document)的主键, 也就是一条记录的pk_version是该_id的更新次数, 我这里已经更新了7次_shards表示分片的结果. 我们这里一共部署了两个节点, 都写入成功了.
在kibana上设置-index manangement里可以查看index的状态. 比如我们这条记录有主副两个分片.
保存记录成功后可以立马读取出来:
GET /customer/_doc/1 返回
{ "_index" : "customer", "_type" : "_doc", "_id" : "1", "_version" : 15, "_seq_no" : 14, "_primary_term" : 1, "found" : true, "_source" : { "name" : "John Doe" } } _source就是我们记录的内容
批量插入
当有多条数据需要插入的时候, 我们可以批量插入. 下载准备好的文档, 然后通过http请求导入es.
创建一个索引bank: 由于shards(分片)和replicas(副本)创建后就不能修改了，所以要先创建的时候配置shards. 这里配置了3个shards和2个replicas.
PUT /bank { "settings": { "index": { "number_of_shards": "3", "number_of_replicas": "2" } } } 文档地址: https://gitee.com/mirrors/elasticsearch/raw/master/docs/src/test/resources/accounts.json
下载下来之后, curl命令或者postman 发送文件请求过去
curl -H "Content-Type: application/json" -XPOST "localhost:9200/bank/_bulk?pretty&refresh" --data-binary "@accounts.json" curl "localhost:9200/_cat/indices?v" 每条记录格式如下:
{ "_index": "bank", "_type": "_doc", "_id": "1", "_version": 1, "_score": 0, "_source": { "account_number": 1, "balance": 39225, "firstname": "Amber", "lastname": "Duke", "age": 32, "gender": "M", "address": "880 Holmes Lane", "employer": "Pyrami", "email": "[email protected]", "city": "Brogan", "state": "IL" } } 在kibana monitor中选择self monitor. 然后再indices中找到索引bank。可以看到我们导入的数据分布情况。
可以看到, 有3个shards分在不同的node上, 并且都有2个replicas.
开始查询
批量插入了一些数据后, 我们就可以开始学习查询了. 上文知道, 数据是银行职员表, 我们查询所有用户,并根据账号排序.
类似 sql
select * from bank order by account_number asc limit 3 Query DSL
GET /bank/_search { "query": { "match_all": {} }, "sort": [ { "account_number": "asc" } ], "size": 3, "from": 2 } _search表示查询query是查询条件, 这里是所有size表示每次查询的条数, 分页的条数. 如果不传, 默认是10条. 在返回结果的hits中显示.from表示从第几个开始
返回:
{ "took" : 1, "timed_out" : false, "_shards" : { "total" : 1, "successful" : 1, "skipped" : 0, "failed" : 0 }, "hits" : { "total" : { "value" : 1000, "relation" : "eq" }, "max_score" : null, "hits" : [ { "_index" : "bank", "_type" : "_doc", "_id" : "2", "_score" : null, "_source" : { "account_number" : 2, "balance" : 28838, "firstname" : "Roberta", "lastname" : "Bender", "age" : 22, "gender" : "F", "address" : "560 Kingsway Place", "employer" : "Chillium", "email" : "[email protected]", "city" : "Bennett", "state" : "LA" }, "sort" : [ 2 ] }, { "_index" : "bank", "_type" : "_doc", "_id" : "3", "_score" : null, "_source" : { "account_number" : 3, "balance" : 44947, "firstname" : "Levine", "lastname" : "Burks", "age" : 26, "gender" : "F", "address" : "328 Wilson Avenue", "employer" : "Amtap", "email" : "[email protected]", "city" : "Cochranville", "state" : "HI" }, "sort" : [ 3 ] }, { "_index" : "bank", "_type" : "_doc", "_id" : "4", "_score" : null, "_source" : { "account_number" : 4, "balance" : 27658, "firstname" : "Rodriquez", "lastname" : "Flores", "age" : 31, "gender" : "F", "address" : "986 Wyckoff Avenue", "employer" : "Tourmania", "email" : "[email protected]", "city" : "Eastvale", "state" : "HI" }, "sort" : [ 4 ] } ] } } 返回结果提供了如下信息
tookes查询时间, 单位是毫秒(milliseconds)timed_outsearch是否超时了_shards我们搜索了多少shards, 成功了多少, 失败了多少, 跳过了多少. 关于shard, 简单理解为数据分片, 即一个index里的数据分成了几片，可以理解为按id进行分表。max_score最相关的记录(document)的分数
接下来可可以尝试带条件的查询。
分词查询
查询address中带mill和lane的地址。
GET /bank/_search { "query": { "match": { "address": "mill lane" } }, "size": 2 } 返回
{ "took" : 3, "timed_out" : false, "_shards" : { "total" : 1, "successful" : 1, "skipped" : 0, "failed" : 0 }, "hits" : { "total" : { "value" : 19, "relation" : "eq" }, "max_score" : 9.507477, "hits" : [ { "_index" : "bank", "_type" : "_doc", "_id" : "136", "_score" : 9.507477, "_source" : { "account_number" : 136, "balance" : 45801, "firstname" : "Winnie", "lastname" : "Holland", "age" : 38, "gender" : "M", "address" : "198 Mill Lane", "employer" : "Neteria", "email" : "[email protected]", "city" : "Urie", "state" : "IL" } }, { "_index" : "bank", "_type" : "_doc", "_id" : "970", "_score" : 5.4032025, "_source" : { "account_number" : 970, "balance" : 19648, "firstname" : "Forbes", "lastname" : "Wallace", "age" : 28, "gender" : "M", "address" : "990 Mill Road", "employer" : "Pheast", "email" : "[email protected]", "city" : "Lopezo", "state" : "AK" } } ] } } - 我设置了返回2个，但实际上命中的有19个
完全匹配查询
GET /bank/_search { "query": { "match_phrase": { "address": "mill lane" } } } 这时候查的完全符合的就一个了
{ "took" : 1, "timed_out" : false, "_shards" : { "total" : 1, "successful" : 1, "skipped" : 0, "failed" : 0 }, "hits" : { "total" : { "value" : 1, "relation" : "eq" }, "max_score" : 9.507477, "hits" : [ { "_index" : "bank", "_type" : "_doc", "_id" : "136", "_score" : 9.507477, "_source" : { "account_number" : 136, "balance" : 45801, "firstname" : "Winnie", "lastname" : "Holland", "age" : 38, "gender" : "M", "address" : "198 Mill Lane", "employer" : "Neteria", "email" : "[email protected]", "city" : "Urie", "state" : "IL" } } ] } } 多条件查询
实际查询中通常是多个条件一起查询的
GET /bank/_search { "query": { "bool": { "must": [ { "match": { "age": "40" } } ], "must_not": [ { "match": { "state": "ID" } } ] } } } bool用来合并多个查询条件must,should,must_not是boolean查询的子语句，must,should决定相关性的score，结果默认按照score排序must not是作为一个filter，影响查询的结果，但不影响score，只是从结果中过滤。
还可以显式地指定任意过滤器，以包括或排除基于结构化数据的文档。
比如，查询balance在20000和30000之间的。
GET /bank/_search { "query": { "bool": { "must": { "match_all": {} }, "filter": { "range": { "balance": { "gte": 20000, "lte": 30000 } } } } } } 聚合运算group by
按照省份统计人数
按sql的写法可能是
select state AS group_by_state, count(*) from tbl_bank limit 3; 对应es的请求是
GET /bank/_search { "size": 0, "aggs": { "group_by_state": { "terms": { "field": "state.keyword", "size": 3 } } } } size=0是限制返回内容， 因为es会返回查询的记录， 我们只想要聚合值aggs是聚合的语法词group_by_state是一个聚合结果， 名称自定义terms查询的字段精确匹配, 这里是需要分组的字段state.keywordstate是text类型, 字符类型需要统计和分组的，类型必须是keywordsize=3限制group by返回的数量，这里是top3, 默认top10, 系统最大10000，可以通过修改search.max_buckets实现， 注意多个shards会产生精度问题， 后面再深入学习
返回值：
{ "took" : 5, "timed_out" : false, "_shards" : { "total" : 3, "successful" : 3, "skipped" : 0, "failed" : 0 }, "hits" : { "total" : { "value" : 1000, "relation" : "eq" }, "max_score" : null, "hits" : [ ] }, "aggregations" : { "group_by_state" : { "doc_count_error_upper_bound" : 26, "sum_other_doc_count" : 928, "buckets" : [ { "key" : "MD", "doc_count" : 28 }, { "key" : "ID", "doc_count" : 23 }, { "key" : "TX", "doc_count" : 21 } ] } } } hits命中查询条件的记录，因为设置了size=0， 返回[].total是本次查询命中了1000条记录aggregations是聚合指标结果group_by_state是我们查询中命名的变量名doc_count_error_upper_bound没有在这次聚合中返回、但是可能存在的潜在聚合结果.键名有「上界」的意思，也就是表示在预估的最坏情况下沒有被算进最终结果的值，当然doc_count_error_upper_bound的值越大，最终数据不准确的可能性越大，能确定的是，它的值为 0 表示数据完全正确，但是它不为 0，不代表这次聚合的数据是错误的.sum_other_doc_count聚合中没有统计到的文档数
值得注意的是, top3是否是准确的呢. 我们看到doc_count_error_upper_bound是有错误数量的, 即统计结果很可能不准确, 并且得到的top3分别是28,23,21. 我们再来添加另个查询参数来比较结果:
GET /bank/_search { "size": 0, "aggs": { "group_by_state": { "terms": { "field": "state.keyword", "size": 3, "shard_size": 60 } } } } ----------------------------------------- "aggregations" : { "group_by_state" : { "doc_count_error_upper_bound" : 0, "sum_other_doc_count" : 915, "buckets" : [ { "key" : "TX", "doc_count" : 30 }, { "key" : "MD", "doc_count" : 28 }, { "key" : "ID", "doc_count" : 27 } ] } } shard_size表示每个分片计算的数量. 因为agg聚合运算是每个分片计算出一个结果,然后最后聚合计算最终结果. 数据在分片分布不均衡, 每个分片的topN并不是一样的, 就有可能最终聚合结果少算了一部分. 从而导致doc_count_error_upper_bound不为0. es默认shard_size的值是size*1.5+10, size=3对应就是14.5, 验证shar_size=14.5时返回值确实和不传一样. 而设置为60时, error终于为0了, 即, 可以保证这个3个绝对是最多的top3. 也就是说, 聚合运算要设置shard_size尽可能大, 比如size的20倍.
按省份统计人数并计算平均薪酬
我们想要查看每个省的平均薪酬, sql可能是
select state, avg(balance) AS average_balance, count(*) AS group_by_state from tbl_bank group by state limit 3 在es可以这样查询:
GET /bank/_search { "size": 0, "aggs": { "group_by_state": { "terms": { "field": "state.keyword", "size": 3, "shard_size": 60 }, "aggs": { "average_balance": { "avg": { "field": "balance" } }, "sum_balance": { "sum": { "field": "balance" } } } } } } - 第二个
aggs是计算每个state的聚合指标 average_balance自定义的变量名称, 值为相同state的balanceavg运算sum_balance自定义的变量名称, 值为相同state的balancesum运算
结果如下:
{ "took" : 12, "timed_out" : false, "_shards" : { "total" : 3, "successful" : 3, "skipped" : 0, "failed" : 0 }, "hits" : { "total" : { "value" : 1000, "relation" : "eq" }, "max_score" : null, "hits" : [ ] }, "aggregations" : { "group_by_state" : { "doc_count_error_upper_bound" : 0, "sum_other_doc_count" : 915, "buckets" : [ { "key" : "TX", "doc_count" : 30, "sum_balance" : { "value" : 782199.0 }, "average_balance" : { "value" : 26073.3 } }, { "key" : "MD", "doc_count" : 28, "sum_balance" : { "value" : 732523.0 }, "average_balance" : { "value" : 26161.535714285714 } }, { "key" : "ID", "doc_count" : 27, "sum_balance" : { "value" : 657957.0 }, "average_balance" : { "value" : 24368.777777777777 } } ] } } } 按省份统计人数并按照平均薪酬排序
agg terms默认排序是count降序, 如果我们想用其他方式, sql可能是这样:
select state, avg(balance) AS average_balance, count(*) AS group_by_state from tbl_bank group by state order by average_balance limit 3 对应es可以这样查询:
GET /bank/_search { "size": 0, "aggs": { "group_by_state": { "terms": { "field": "state.keyword", "order": { "average_balance": "desc" }, "size": 3 }, "aggs": { "average_balance": { "avg": { "field": "balance" } } } } } } 返回结果的top3就不是之前的啦:
"aggregations" : { "group_by_state" : { "doc_count_error_upper_bound" : -1, "sum_other_doc_count" : 983, "buckets" : [ { "key" : "DE", "doc_count" : 2, "average_balance" : { "value" : 39040.5 } }, { "key" : "RI", "doc_count" : 5, "average_balance" : { "value" : 36035.4 } }, { "key" : "NE", "doc_count" : 10, "average_balance" : { "value" : 35648.8 } } ] } } 参考
- 中文社区:https://elasticsearch.cn/
- es官方文档: https://www.elastic.co/guide/en/elasticsearch/reference/current/documents-indices.html
- es官方文档: https://www.elastic.co/guide/en/elasticsearch/reference/current/getting-started-index.html
- terms 聚合计算不准确: https://www.dongwm.com/post/elasticsearch-terms-agg-is-not-accurate/
