I've been building some custom search components for SOLR lately, so wanted to share a couple of things I learned in the process. Most likely this is old hat to people who have been doing this for a while, but thought I'd share, just in case it benefits someone...

Passing State

In a previous post, I described a custom SOLR search handler returns layered search results for a given query term (and optional filters). As I went further, though, I realized that I needed to return information relating to facets and category clusters as well. Of course, I could have added this stuff into the handler itself, but splitting the logic across a chain of search components seemed to be more preferable, readability and reusability wise, so I went that route.

So the first step was to refactor my custom SearchHandler into a SearchComponent. Not much to do there, except to subclass SearchComponent instead of RequestHandlerBase and move the handleRequestBody(SolrQueryRequest,SolrQueryResponse) to a process(ResponseBuilder) method. The request and response objects are accessible from the ResponseBuilder as properties, ie, ResponseBuilder.req and ResponseBuilder.rsp. I then declared this component and an enclosing handler in solrconfig.xml, something like this:

 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
  <!-- this used to be my search handler -->
<searchComponent name="component1"
class="org.apache.solr.handler.component.ext.MyComponent1">
<str name="prop1">value1</str>
<str name="prop2">value2</str>
</searchComponent>
<searchComponent name="component2"
class="org.apache.solr.handler.component.ext.MyComponent2">
<lst name="facets">
<str name="prop1">1</str>
<str name="prop2">2</str>
</lst>
</searchComponent>
<requestHandler name="/mysearch2"
class="org.apache.solr.handler.component.SearchHandler">
<lst name="defaults">
<str name="echoParams">explicit</str>
<str name="fl">*,score,id</str>
<str name="wt">xml</str>
</lst>
<arr name="components">
<str>component1</str>
<str>component2</str>
<!-- ... more components as needed ... -->
</arr>
</requestHandler>

I've also added a second component to the chain above (just so I don't have to show this snippet again later), hope its not too confusing. Obviously there can be multiple components before and after my search handler turned search component, but for the purposes of this discussion, I'll keep things simple and just concentrate on this one other component and pretend that it has multiple unique (and pertinent) requirements.

Now, assume that the second component needed data that was already available, or can be easily generated by component1. Its actually true in my case, since I needed a BitSet of document ids in the search results in my second component, which I could easily get by collecting them while looping through the SolrDocumentList of results in my first component. So it seemed kind of wasteful to compute this again. So I updated this snippet of code in component1's process() method (what used to be my handleRequestBody() method):

 1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
  public void process(ResponseBuilder rb) throws IOException {
...
// build and write response
...
OpenBitSet bits = new OpenBitSet(searcher.maxDoc());
List<SolrDocument> slice = new ArrayList<SolrDocument>();
for (Iterator<SolrDocument> it = results.iterator(); it.hasNext(); ) {
SolrDocument sdoc = it.next();
...
bits.set(Long.valueOf((Integer) sdoc.get("id")));
if (numFound >= start && numFound < start + rows) {
slice.add(sdoc);
}
numFound++;
}
...
rsp.add("response", results);
rsp.add("_bits", bits);
}

In my next component (component2), I simply grab the OpenBitSet data structure by name from the NamedList, use them to generate the result for this component, stick the result back into the response, and discard the temporary data. The last is so that the data does not appear on the response XML (for both aesthetic and performance reasons).

 1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
  public void process(ResponseBuilder rb) throws IOException {
Map<String,Object> cres = new HashMap<String,Object>();
NamedList nl = rb.rsp.getValues();
OpenBitSet bits = (OpenBitSet) nl.get("_bits");
if (bits == null) {
logger.warn("Component 1 must write _bits into response");
rb.rsp.add(COMPONENT_NAME, cres);
return;
}
// do something with bits and generate component response
doSomething(bits, cres);
// stick the result into the response and delete temp data
rb.rsp.add("component2_result", cres);
rb.rsp.getValues().remove("_bits");
}

Before I did this, I investigated if I could subclass the XmlResponseWriter to ignore NamedLists with "hidden" names (ie names prefixed with underscore), but the XmlResponseWriter calls XMLWriter which does the actual XML generation, and XMLWriter is final (at least in SOLR 1.4.1). Good thing too, forced me to look for and find a simpler solution :-).

So there you have it - a simple way to pass data between components in a SOLR Search RequestHandler. Note that it does mean that component2 is always dependent on component1 (or some other component that produces the same data) upstream to it, so these components are no longer truly reusable pieces of code. But this can be useful if you really need it and you document the requirement (or complain about it if not met, as I've done here).

Reacting to a COMMIT

The second thing I needed to do in component2 was to give it some reference data that it would need to compute its results. The reference data is generated from the contents of the index, and the generation is fairly heavyweight, so you don't want to do this on every request.

Now one of the cool things about SOLR is its built-in incremental indexing feature (one of the main reasons we considered using SOLR in the first place), so you can POST data to a running SOLR instance followed by a COMMIT, and voila: your searcher re-opens with the new data.

Of course, this also means that if we want to provide accurate information, the reference data should be regenerated whenever the searcher is reopened. The way I went about doing this is mostly derived from how the SpellCheckerComponent does it, in order to regenerate its dictionaries -- by hooking into the SOLR event framework.

To do this, my component2 implements SolrCoreAware in addition to extending SearchComponent. This requires me to implement the inform(SolrCore) method, which is invoked by SOLR after the init(NamedList) but before prepare(ResponseBuilder) and process(ResponseBuilder). In the inform(SolrCore) method, I register a listener for the firstSearcher and newSearcher events (described in more detail here).

I then build the inner listener class, which implements SolrEventListener, which requires me to provide implementations for newSearcher() and postCommit() methods. Since my listener is a query-side listener, I provide an empty implementation for postCommit(). The newSearcher() method contains the code to generate the reference sets. Here is the relevant snippet of code from the component.

 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
public class MyComponent2 extends SearchComponent implements SolrCoreAware {

  private RefData refdata; // this needs to be regenerated on COMMIT

  @Override
public void init(NamedList args) {
...
} @Override
public void inform(SolrCore core) {
listener = new MyComponent2Listener();
core.registerFirstSearcherListener(listener);
core.registerNewSearcherListener(listener);
} @Override
public void prepare(ResponseBuilder rb) throws IOException {
...
} @Override
public void process(ResponseBuilder rb) throws IOException {
...
// do something with refdata
...
} private class MyComponent2Listener implements SolrEventListener { @Override
public void init(NamedList args) { /* NOOP */ } @Override
public void newSearcher(SolrIndexSearcher newSearcher,
SolrIndexSearcher currentSearcher) {
RefData copy = new RefData();
copy = generateRefData(newSearcher);
refdata.clear();
refdata.addAll(copy);
} @Override
public void postCommit() { /* NOOP */ }
}
...
}

Notice that I have registered the listener to listen on both firstSearcher and newSearcher events. This way, it gets called on SOLR startup (reacting to a firstSearcher event), and again each time the searcher is reopened (reacting to a newSearcher event).

One other thing... since the generation of RefData takes some time, its best to have the listener's newSearcher method build a copy and then repopulate the refdata variable from the copy, that way the component continues to use the old data until the new one is available.

And thats pretty much it for today. Till next time.

http://sujitpal.blogspot.com/2011/04/custom-solr-search-components-2-dev.html

Custom SOLR Search Components - 2 Dev Tricks的更多相关文章

  1. How to Reuse Old PCs for Solr Search Platform?

    家裡或公司的舊電腦不夠力? 效能慢到想砸爛它們? 朋友或同事有電腦要丟嗎? 我有一個廢物利用的方法, 我收集了四台舊電腦, 組了一個Fully Distributed Mode的Hadoop Clus ...

  2. solr search基础知识(控制符及其参数)

    1.^ 控制符 (1)查询串上用^ 搜索: 天后王菲,如果希望将王菲的相关度加大,用^控制符. 天后  王菲^10.5  结果就会将含有王菲的document权重加大分数提高,排序靠前,10.5为权重 ...

  3. Solr Cloud - SolrCloud

    关于 Solr Cloud Zookeeper 入门,介绍 原理 原封不动转自 http://wiki.apache.org/solr/SolrCloud/ ,文章的内存有些过时,但是了解原理. Th ...

  4. Solr: a custom Search RequestHandler

    As you know, I've been playing with Solr lately, trying to see how feasible it would be to customize ...

  5. Importing/Indexing database (MySQL or SQL Server) in Solr using Data Import Handler--转载

    原文地址:https://gist.github.com/maxivak/3e3ee1fca32f3949f052 Install Solr download and install Solr fro ...

  6. Solr 6.7学习笔记(03)-- 样例配置文件 solrconfig.xml

    位于:${solr.home}\example\techproducts\solr\techproducts\conf\solrconfig.xml <?xml version="1. ...

  7. Solr基础知识二(导入数据)

    上一篇讲述了solr的安装启动过程,这一篇讲述如何导入数据到solr里. 一.准备数据 1.1 学生相关表 创建学生表.学生专业关联表.专业表.学生行业关联表.行业表.基础信息表,并创建一条小白的信息 ...

  8. Solr 02 - 最详细的solrconfig.xml配置文件解读

    目录 1 luceneMatchVersion - 指定Lucene版本 2 lib - 配置扩展jar包 3 dataDir - 索引数据路径 4 directoryFactory - 索引存储工厂 ...

  9. solr 5.3.1安装配置

    1.下载Solr5.3.1 http://mirror.bit.edu.cn/apache/lucene/solr/5.3.1/ wget http://mirror.bit.edu.cn/apach ...

随机推荐

  1. Julia - 函数运算符

    Julia 中,大多数运算符都是支持特定语法的函数 && . || 等短路运算是例外,它们不是函数,因为短路求值先算前面的值,再算后面的值 对于函数运算符,可以像其它函数一样,把参数列 ...

  2. 【转】详解Data Binding 通过几个简单示例深入了解WinForm数据绑定特性

    原文地址:http://www.cnblogs.com/lichence/archive/2012/02/17/2356001.html

  3. nginx的下载和安装

    安装 下载必要组件 nginx下载地址 http://nginx.org/en/download.html  pcre库下载地址,nginx需要[解析正则] http://sourceforge.ne ...

  4. 1、linux软件包管理

    linux软件包管理分为两种:RPM包管理和源码包管理,其中RPM包管理又有两种方式:①RPM命令管理,②YUM在线命令管理. RPM包依赖关系: 1.树形依赖 2.环形依赖 (用一条命令同时装来解决 ...

  5. leetcode357

    public class Solution { public int CountNumbersWithUniqueDigits(int n) { ) { ; } ; ; ; && av ...

  6. 如何设置tomcat,直接通过IP 访问

    找到tomcat的主目录,进入conf文件夹,找到server.xml文件,并打开:   修改tomcat的监听端口为80端口: 在server.xml文件中找到: <Connector por ...

  7. #if (DEBUG)

    //DEBUG必须大写,其它是不认的#if (DEBUG)            Console.WriteLine("Debug");#else            Conso ...

  8. Texture Format全解析

    [Texture Format全解析] What internal representation is used for the texture. This is a tradeoff between ...

  9. Python创建单例模式的5种常用方法-乾颐堂

    所谓单例,是指一个类的实例从始至终只能被创建一次. 方法1 如果想使得某个类从始至终最多只有一个实例,使用__new__方法会很简单.Python中类是通过__new__来创建实例的: 1 2 3 4 ...

  10. 深入理解yield-乾颐堂

    yield的英文单词意思是生产,刚接触Python的时候感到非常困惑,一直没弄明白yield的用法. 只是粗略的知道yield可以用来为一个函数返回值塞数据,比如下面的例子: 1 2 3 def ad ...