[b][color=olive][size=large]对于用惯数据库的我们,多表进行join连接,是非常常见的一个需求,但是在我们的索引中,对join的支持,却不是很完美,当然这并不是由于我们的Lucene或Solr不够强大,而是全文检索与数据库的定位不是在同一个目标上,全文检索,主要定位在搜索引擎上,通常是对一个大的索引进行高效检索,而数据库则是定位在结构化数据的存储于与检索,检索功能比较薄弱,那我们的索引是不是就不支持join了,实事并非如此,Lucene里面支持join操作,这种join定位在同一份索引里,而Solr作为Lucene的扩展,又提供了两core join的功能,下面散仙给出一个例子,尽量简单,清晰的描述出如何使用它们和理解它们的工作方式。
[/size][/color][/b]
[b][color=green][size=large]散仙,有2个core,分别是collection1,和collection2,里面的数据分别是:
collection1:总共有3条数据
collection1:schema 都是字符串string ; 有id,name两个字段 ;
collection1: {1, Apple}, {2, Samsung}, {3, HTC}
collection2:总共有5条数据
collection2:schema 都是字符串string ;有id,name,brand_id 两个字段;
collection2: {1, iPhone, 1}, {2, iPad, 1}, {3, Galaxy S3, 2}, {4, Galaxy Note, 2}, {5, One X, 3}
下面,先来看下单core的join,以collection2作为例子,测试代码如下:
[/size][/color][/b]
/***
* join测试
*
*
* ***/
public static void joinquery2()throws Exception{
SolrServer server1=new HttpSolrServer("http://localhost:9003/solr/collection2");
SolrQuery sq=new SolrQuery();
//sq.set("fl", "id,name");//过滤只需要返回的字段
sq.set("q", "{!join from=id to=brand_id }brand_id:*");
QueryResponse qr=server1.query(sq, METHOD.POST);
SolrDocumentList list=qr.getResults();
System.out.println("命中结果集:"+qr.getResults().size());
for(SolrDocument s:list){
System.out.println(s.toString());
}
}
[b][color=olive][size=large]运行结果如下:[/size][/color][/b]
五月 14, 2014 9:03:58 下午 org.apache.solr.client.solrj.impl.HttpClientUtil createClient
INFO: Creating new http client, config:maxConnections=128&maxConnectionsPerHost=32&followRedirects=false
命中结果集:5
SolrDocument{id=1, name=iPhone, brand_id=1, _version_=1468079557386960896}
SolrDocument{id=2, name=iPad, brand_id=1, _version_=1468079557408980992}
SolrDocument{id=3, name=Galaxy, brand_id=2, _version_=1468079557412126720}
SolrDocument{id=4, name=Galaxy Note, brand_id=2, _version_=1468079557416321024}
SolrDocument{id=5, name=One X, brand_id=3, _version_=1468079557420515328}
[b][color=green]改变,条件后,再测:[/color][/b]
/***
* join测试
*
*
* ***/
public static void joinquery2()throws Exception{
SolrServer server1=new HttpSolrServer("http://localhost:9003/solr/collection2");
SolrQuery sq=new SolrQuery();
//sq.set("fl", "id,name");//过滤只需要返回的字段
sq.set("q", "{!join from=id to=brand_id }brand_id:2");
QueryResponse qr=server1.query(sq, METHOD.POST);
SolrDocumentList list=qr.getResults();
System.out.println("命中结果集:"+qr.getResults().size());
for(SolrDocument s:list){
System.out.println(s.toString());
}
}
[b][color=olive][size=large]运行结果如下:[/size][/color][/b]
五月 14, 2014 9:10:04 下午 org.apache.solr.client.solrj.impl.HttpClientUtil createClient
INFO: Creating new http client, config:maxConnections=128&maxConnectionsPerHost=32&followRedirects=false
命中结果集:1
SolrDocument{id=5, name=One X, brand_id=3, _version_=1468079557420515328}
[size=large][color=olive]分析运行原理,类似sql中的写法:
SELECT *
FROM collection1
WHERE brand_id IN (SELECT id FROM collection1 where brand_id = * )
第一步,先执行子查询SELECT id FROM collection1 where brand_id = *
会返回所有的id分别是,1,2,3,4,5
第二步,执行主查询就是
SELECT *
FROM collection1
WHERE brand_id in (1,2,3,4,5)
而brand_id去重完之后,就只有1,2,3了,所以转换成如下查询:
SELECT *
FROM collection1
WHERE brand_id(1,2,3) in (1,2,3,4,5)
取并集后结果,就会命中brand_id=1,2,3的文档,所以就命中了所有的文档
再来分析下,第二个查询,指定查询id的join:
第一步,先执行子查询SELECT id FROM collection1 where brand_id = 2
会返回所有的id分别是,3,4,
第二步,执行主查询就是
SELECT *
FROM collection1
WHERE brand_id in (3,4)
而brand_id去重完之后,就只有1,2,3了,所以转换成如下查询:
SELECT *
FROM collection1
WHERE brand_id(1,2,3) in brand_id(3,4)
取并集后的结果,就会命中brand_id=3的文档了,所以就会返回ID为5的文档;
[/color][/size]
[color=olive][size=large]下面,来测下,两个core的join,代码如下:[/size][/color]
/***
* join测试
*
*
* ***/
public static void joinquery2()throws Exception{
SolrServer server1=new HttpSolrServer("http://localhost:9003/solr/collection1");
SolrQuery sq=new SolrQuery();
//sq.set("fl", "id,name");//过滤只需要返回的字段
sq.set("q", "{!join from=brand_id to=id fromIndex=collection2}name:iPad");
QueryResponse qr=server1.query(sq, METHOD.POST);
SolrDocumentList list=qr.getResults();
System.out.println("命中结果集:"+qr.getResults().size());
for(SolrDocument s:list){
System.out.println(s.toString());
}
}
[color=green][size=large]结果如下:[/size][/color]
五月 14, 2014 9:30:41 下午 org.apache.solr.client.solrj.impl.HttpClientUtil createClient
INFO: Creating new http client, config:maxConnections=128&maxConnectionsPerHost=32&followRedirects=false
命中结果集:1
SolrDocument{id=1, name=Apple, _version_=1468079556974870528}
[color=olive][size=large]分析类似如下SQL:
SELECT b.* FROM collection1 b
INNER JOIN collection2 p ON b.id=p.brand_id
WHERE p.name="iPad";
注意collection名的先后顺序,如上solrj里面的执行,跟上面的sql的运行规则是一样的,所以我们最终的结果里,会返回,如果我们的条件是下面的相反组合:
[/size][/color]
public static void joinquery()throws Exception{
SolrServer server1=new HttpSolrServer("http://localhost:9003/solr/collection2");
SolrQuery sq=new SolrQuery();
sq.set("q", "{!join from=id to=brand_id fromIndex=collection1}id:1");
QueryResponse qr=server1.query(sq, METHOD.POST);
SolrDocumentList list=qr.getResults();
System.out.println("命中结果集:"+qr.getResults().size());
for(SolrDocument s:list){
//s.toString();
System.out.println(s.toString());
}
[color=olive][size=large]则运行结果如下所示:[/size][/color]
五月 14, 2014 9:43:46 下午 org.apache.solr.client.solrj.impl.HttpClientUtil createClient
INFO: Creating new http client, config:maxConnections=128&maxConnectionsPerHost=32&followRedirects=false
命中结果集:2
SolrDocument{id=1, name=iPhone, brand_id=1, _version_=1468079557386960896}
SolrDocument{id=2, name=iPad, brand_id=1, _version_=1468079557408980992}
[color=olive][size=large]原理,依旧与如上的sql一样。只不过位置相反,调整了:
SELECT b.* FROM collection2 b
INNER JOIN collection1 p ON b.id=p.brand_id
WHERE b.id=1;
[/size][/color]