转自:http://www.aboutyun.com/thread-7742-1-1.html
Cloudera公司已经推出了基于Hadoop平台的查询统计分析工具Impala,只要熟悉SQL,就可以熟练地使用Impala来执行查询与分析的功能。不过Impala的SQL和关系数据库的SQL还是有一点微妙地不同的。下面,我们设计一个表,通过该表中的数据,来将SQL查询与统计的语句,使用Solr查询的方式来与SQL查询对应。这个翻译的过程,是非常有趣的,你可以看到Solr一些很不错的功能。用来示例的表结构设计,如图所示:
下面,我们通过给出一些SQL查询统计语句,然后对应翻译成Solr查询语句,然后对比结果查询对比
条件组合查询
SQL查询语句:
SELECT log_id,start_time,end_time,prov_id,city_id,area_id,idt_id,cnt,net_type
FROM v_i_event
WHERE prov_id = 1 AND net_type = 1 AND area_id = 10304 AND time_type = 1 AND time_id >= 20130801 AND time_id <= 20130815
ORDER BY log_id LIMIT 10;
查询结果如下图:
Solr查询URL
http://slave1:8888/solr-cloud/i_event/select?q=*:*&fl=log_id,start_time,end_time,prov_id,city_id,area_id,idt_id,cnt,net_type&fq=prov_id:1 AND net_type:1 AND area_id:10304 AND time_type:1 AND time_id:[20130801 TO 20130815]&sort=log_id asc&start=0&rows=10
查询结果,如下所示:
忽略。
对比上面结果,除了根据idt_id排序方式不同以外(Impala是升序,Solr是降序),其他是相同的。
2.分组查询
SELECT prov_id, SUM(cnt) AS sum_cnt, AVG(cnt) AS avg_cnt, MAX(cnt) AS max_cnt, MIN(cnt) AS min_cnt, COUNT(cnt) AS count_cnt
FROM v_i_event
GROUP BY prov_id;
查询结果如下图:
Solr 分组查询
http://slave1:8888/solr-cloud/i_event/select?q=*:*&stats=true&stats.field=cnt&rows=0&indent=true
查询结果,如下所示:
忽略。
对比查询结果,Solr提供了更多的统计项,如标准差(stddev)等,与SQL查询结果是一致的。
3.IN条件查询
SELECT log_id,start_time,end_time,prov_id,city_id,area_id,idt_id,cnt,net_typ
FROM v_i_event
WHERE prov_id = 1 AND net_type = 1 ANDcity_id IN(106,103) AND idt_id IN(12011,5004,6051,6056,8002) AND time_type = 1AND time_id >= 20130801 AND time_id <= 20130815
ORDER BY log_id, start_time DESC LIMIT 10;
Solr查询URL:
http://slave1:8888/solr-cloud/i_event/select?q=*:*&fl=log_id,start_time,end_time,prov_id,city_id,area_id,idt_id, cnt,net_type&fq=prov_id:1 AND net_type:1 AND (city_id:106 OR city_id:103) AND (idt_id:12011 OR idt_id:5004 OR idt_id:6051 OR idt_id:6056 OR idt_id:8002) AND time_type:1 AND time_id:[20130801 TO 20130815]&sort=log_id asc ,start_time desc&start=0&rows=10
或
http://slave1:8888/solr-cloud/i_event/select?q=*:*&fl=log_id,start_time,end_time,prov_id,city_id,area_id,idt_id, cnt ,net_type&fq=prov_id:1&fq=net_type:1&fq=(city_id:106 OR city_id:103)&fq=(idt_id:12011 OR idt_id:5004 OR idt_id:6051 OR idt_id:6056 OR idt_id:8002)&fq=time_type:1&fq=time_id:[20130801 TO 20130815]&sort=log_id asc,start_time desc&start=0&rows=10
查询结果,如下所示:
忽略。
4.开区间范围条件查询
SQL查询
SELECTlog_id,start_time,end_time,prov_id,city_id,area_id,idt_id,cnt,net_type
FROM v_i_event
WHERE net_type = 1 AND idt_id IN(12011,5004,6051,6056,8002) AND time_type = 1 AND start_time >= 1373598465 AND start_time < 1374055254
ORDER BY log_id, start_time, idt_id DESCLIMIT 30;
查询结果:
SOLR 查询
http://slave1:8888/solr-cloud/i_event/select?q=*:*&fl=log_id,start_time,end_time,prov_id,city_id,area_id,idt_id,cnt,net_type&fq=net_type:1 AND (idt_id:12011 OR idt_id:5004 OR idt_id:6051 OR idt_id:6056 OR idt_id:8002) AND time_type:1 AND start_time:[1373598465 TO 1374055254]&fq =-start_time:1374055254&sort=log_id asc,start_time asc,idt_id desc&start=0&rows=30
或
http://slave1:8888/solr-cloud/i_event/select?q=*:*&fl=log_id,start_time,end_time,prov_id,city_id,area_id,idt_id,cnt,net_type&fq=net_type:1 AND (idt_id:12011 OR idt_id:5004 OR idt_id:6051 OR idt_id:6056 OR idt_id:8002) AND time_type:1 AND start_time:[1373598465 TO 1374055254] AND -start_time:1374055254&sort=log_id asc,start_time asc,idt_id desc&start=0&rows=30
或
http://slave1:8888/solr-cloud/i_event/select?q=*:*&fl=log_id,start_time,end_time,prov_id,city_id,area_id,idt_id,cnt,net_type&fq=net_type:1&fq=idt_id:12011 OR idt_id:5004 OR idt_id:6051 OR idt_id:6056 OR idt_id:8002&fq =time_type:1&fq=start_time:[1373598465 TO 1374055254]&fq =-start_time:1374055254&sort=log_id asc,start_time asc,idt_id desc&start=0&rows=30
5.多个字段分组统计(只支持count函数)
SQL查询语句:
SELECT city_id, area_id, COUNT(cnt) AScount_cnt
FROM v_i_event
WHERE prov_id = 1 AND net_type = 1
GROUP BY city_id, area_id;
查询结果:
Solr查询语句:
http://slave1:8888/solr-cloud/i_event/select?q=*:*&facet=true&facet.pivot=city_id,area_id&fq=prov_id:1 AND net_type:1&rows=0&indent=true
对比上面结果,Solr查询结果,需要从上面的各组中进行合并,得到最终的统计结果,结果和SQL结果是一致的。
6.多个字段分组统计
(支持count、sum、max、min等函数)
一次对多个字段进行独立分组统计,Solr可以很好的支持。这相当于执行两个带有GROUP BY子句的SQL,这两个GROUP BY分别只对一个字段进行汇总统计。
SQL查询
SELECT city_id, area_id, COUNT(cnt) AS count_cnt
FROM v_i_event
WHERE prov_id = 1 AND net_type = 1
GROUP BY city_id;
SELECT city_id, area_id, COUNT(cnt) AS count_cnt
FROM v_i_event
WHERE prov_id = 1 AND net_type = 1
GROUP BY area_id;
Solr查询语句
http://slave1:8888/solr-cloud/i_event/select?q=*:*&stats=true&stats.field=cnt&f.cnt.stats.facet=city_id&&f.cnt.stats.facet=area_id&fq=prov_id:1 AND net_type:1&rows=0&indent=true
结果:
0
72
city_id
103
678
area_id
10307
298
area_id
10315
120
area_id
10317
86
area_id
10304
67
area_id
10310
49
area_id
70104
48
area_id
10308
6
area_id
0
2
area_id
10311
2
city_id
0
463
area_id
0
395
area_id
10307
68
7.多个字段联合分组统计
支持count、sum、max、min等函数
SQL查询语句:
SELECT city_id, area_id, SUM(cnt) ASsum_cnt, AVG(cnt) AS avg_cnt, MAX(cnt) AS max_cnt, MIN(cnt) AS min_cnt,COUNT(cnt) AS count_cnt
FROM v_i_event
WHERE prov_id = 1 AND net_type = 1
GROUP BY city_id, area_id;
查询结果:
Solr目前不能简单的支持这种查询,如果想要满足这种查询统计,需要在schema的设计上,将一个字段设置为多值,然后通过多个值进行分组统计。如果应用中查询统计分析的模式比较固定,预先知道哪些字段会用于联合分组统计,完全可以在设计的时候,考虑设置多值字段来满足这种需求。
说明: "facet.limit"设置为-1,不然得不到facet所有结果。