当前，许多应用重度依赖于搜索功能。从电子商务网站中寻找合适的产品，到社交网络中搜索寻人，再到地图网站中寻找POI和地址，依赖于搜索的应用非常广泛。
亚马逊新推出的云搜索服务，为自行实现搜索功能或定制安装Apache Lucene、Apache Solr和elasticsearch等流行产品提供了可行的替代方式。他们这样描述该服务：
数据定义尽管亚马逊提供了数据上传和搜索响应的数据定义（以XML和JSON两种方式），但数据上传的文档中仅定义了Relax NG 模式，而搜索响应则未定义任何模式。 在我们的实现方式中，我们决定使用XML数据格式而不是JSON，这是因为进行XML数据封装更加简单——XML使用规范的数据格式，而JSON则是动态的（JSON的标签是动态定义，每个请求各异）。我们分别用下边的两种模式（列表1和列表2）来上传数据和搜索结果。 <?xml version="1.0" encoding="UTF-8"?>
<xsd:schema xmlns:xsd="http://www.w3.org/2001/XMLSchema" elementFormDefault="qualified">
………………………………………………………………………………………….
<xsd:complexType name="fieldType">
<xsd:simpleContent>
<xsd:extension base="xsd:string">
<xsd:attribute name="name" type="field_nameType" />
</xsd:extension>
</xsd:simpleContent>
</xsd:complexType>
<xsd:complexType name="addType">
<xsd:sequence>
<xsd:element name="field" type="fieldType" maxOccurs="unbounded" />
</xsd:sequence>
<xsd:attribute name="id" type="IDType" />
<xsd:attribute name="version" type="versionType" />
<xsd:attribute name="lang" type="xsd:language" />
</xsd:complexType>
<xsd:complexType name="deleteType">
<xsd:attribute name="id" type="IDType" />
<xsd:attribute name="version" type="versionType" />
</xsd:complexType>
<xsd:complexType name="batchType">
<xsd:sequence>
<xsd:element name="add" type="addType" minOccurs="0" maxOccurs="unbounded" />
<xsd:element name="delete" type="deleteType" minOccurs="0" maxOccurs="unbounded" />
</xsd:sequence>
</xsd:complexType>
<xsd:element name="batch" type="batchType" />
<xsd:simpleType name="statusType">
<xsd:restriction base="xsd:string">
<xsd:enumeration value="success"/>
<xsd:enumeration value="error" />
</xsd:restriction>
</xsd:simpleType>
<xsd:complexType name="errorsType">
<xsd:sequence>
<xsd:element name="error" type="xsd:string" maxOccurs="unbounded" />
</xsd:sequence>
</xsd:complexType>
<xsd:complexType name="warningsType">
<xsd:sequence>
<xsd:element name="warning" type="xsd:string" maxOccurs="unbounded" />
</xsd:sequence>
</xsd:complexType>
<xsd:complexType name="responseType">
<xsd:sequence>
<xsd:element name="errors" type="errorsType" minOccurs="0" />
<xsd:element name="warnings" type="warningsType" minOccurs="0" />
</xsd:sequence>
<xsd:attribute name="status" type="statusType"/>
<xsd:attribute name="adds" type="xsd:int"/>
<xsd:attribute name="deletes" type="xsd:int"/>
</xsd:complexType>
<xsd:element name="response" type="responseType" />
</xsd:schema>
Listing 1 Upload data schema
<xsd:schema xmlns:xsd="http://www.w3.org/2001/XMLSchema"
targetNamespace="http://cloudsearch.amazonaws.com/2011-02-01/results"
xmlns="http://cloudsearch.amazonaws.com/2011-02-01/results"
elementFormDefault="qualified">
<xsd:complexType name="constraintType">
<xsd:attribute name="value" type="xsd:string"/>
<xsd:attribute name="count" type="xsd:int"/>
</xsd:complexType>
<xsd:complexType name="facetType">
<xsd:sequence>
<xsd:element name="constraint" type="constraintType" maxOccurs="unbounded"/>
</xsd:sequence>
<xsd:attribute name="name" type="xsd:string" />
</xsd:complexType>
<xsd:complexType name="facetsType">
<xsd:sequence>
<xsd:element name="facet" type="facetType" maxOccurs="unbounded"/>
</xsd:sequence>
</xsd:complexType>
<xsd:complexType name="infoType">
<xsd:attribute name="rid" type="xsd:string" />
<xsd:attribute name="time-ms" type="xsd:int" />
<xsd:attribute name="cpu-time-ms" type="xsd:int" />
</xsd:complexType>
<xsd:complexType name="dType">
<xsd:simpleContent>
<xsd:extension base="xsd:string">
<xsd:attribute name="name" type="xsd:string" />
</xsd:extension>
</xsd:simpleContent>
</xsd:complexType>
<xsd:complexType name="hitType">
<xsd:sequence>
<xsd:element name="d" type="dType" maxOccurs="unbounded"/>
</xsd:sequence>
<xsd:attribute name="id" type="xsd:string" />
</xsd:complexType>
<xsd:complexType name="hitsType">
<xsd:sequence>
<xsd:element name="hit" type="hitType" maxOccurs="unbounded"/>
</xsd:sequence>
<xsd:attribute name="found" type="xsd:int" />
<xsd:attribute name="start" type="xsd:int" />
</xsd:complexType>
<xsd:complexType name="resultsType">
<xsd:sequence>
<xsd:element name="rank" type="xsd:string" />
<xsd:element name="match-expr" type="xsd:string" />
<xsd:element name="hits" type="hitsType" minOccurs="0"/>
<xsd:element name="facets" type="facetsType" minOccurs="0"/>
<xsd:element name="info" type="infoType" />
</xsd:sequence>
</xsd:complexType>
<xsd:element name="results" type="resultsType"/>
<xsd:complexType name="messageType">
<xsd:attribute name="severity" type="xsd:string" />
<xsd:attribute name="code" type="xsd:string" />
<xsd:attribute name="message" type="xsd:string"/>
</xsd:complexType>
<xsd:complexType name="errorType">
<xsd:sequence>
<xsd:element name="error" type="xsd:string" />
<xsd:element name="rid" type="xsd:string" />
<xsd:element name="time-ms" type="xsd:int" />
<xsd:element name="cpu-time-ms" type="xsd:int" />
<xsd:element name="messages" type="messageType" maxOccurs="unbounded" />
</xsd:sequence>
</xsd:complexType>
<xsd:element name="error" type="errorType" />
</xsd:schema>
Listing 2 Search results data schema
我们使用xjc binding compiler 生成上述的两种模式的Java类，这样就能通过Java Architecture for XML Binding (JAXB)进行自动封装/解封装。 查询定义 除了数据定义，实现搜索API还需要查询定义。我们已经创建了一组类，用来实现亚马逊的查询定义。 这个搜索查询的核心是过滤器。我们引入了SearchQueryFilter 接口，并提供了两种实现方式——Search Query Value Filter（列表3）和Search Query Filter Operation（列表4）。 public class SearchQueryValueFilter implements SearchQueryFilter{
private String _field;
private String _value;
private boolean _isExclude;
private boolean _isNumeric;
public SearchQueryValueFilter(){}
public SearchQueryValueFilter(String field, String value, boolean isNumeric, boolean isExclude){
_field = field;
_value = value;
_isExclude = isExclude;
_isNumeric = isNumeric;
}
public String getField() {
return _field;
}
public void setField(String field) {
_field = field;
}
public String getValue() {
return _value;
}
public void setValue(String value) {
_value = value;
}
public boolean isExclude() {
return _isExclude;
}
public void setExclude(boolean isExclude) {
_isExclude = isExclude;
}
public boolean isNumeric() {
return _isNumeric;
}
public void setNumeric(boolean isNumeric) {
_isNumeric = isNumeric;
}
@Override
public String toString(){
StringBuffer sb = new StringBuffer();
if(_isExclude){
sb.append("(not ");
}
if(_field != null){
sb.append(_field);
sb.append(":");
}
if(!_isNumeric){
sb.append("'");
}
sb.append(_value);
if(!_isNumeric){
sb.append("'");
}
if(_isExclude){
sb.append(")");
}
return sb.toString();
}
}
Listing 3 Value filter implementation
public class SearchQueryFilterOperation implements SearchQueryFilter {
List<SearchQueryFilter> _filters;
FilterOperation _operation;
public SearchQueryFilterOperation(){
_operation = FilterOperation.and;
_filters = new LinkedList<SearchQueryFilter>();
}
public List<SearchQueryFilter> getFilters() {
return _filters;
}
public void setFilters(List<SearchQueryFilter> filters) {
_filters = filters;
}
public void addFilters(SearchQueryFilter filter) {
_filters.add(filter);
}
public FilterOperation getOperation() {
return _operation;
}
public void setOperation(FilterOperation operation) {
_operation = operation;
}
@Override
public String toString() {
StringBuffer sb = new StringBuffer();
sb.append("(");
sb.append(_operation);
for(SearchQueryFilter f : _filters){
sb.append(" ");
sb.append(f);
}
sb.append(")");
return sb.toString();
}
public enum FilterOperation{
and, or
}
}
Listing 4 Operation filter implementation
Search Query Value Filter类支持开发者使用等于、小于、大于、区间（同样支持负值比较）等运算符设置单个字段的限制。而Search Query Filter Operation类还支持开发者使用AND/OR操作符，将多个Search Query Value Filters和Search Query Filter Operations组合使用。通过这两个类的组合使用，就能实现亚马逊云搜索所支持的任意查询过滤器的表达式了。 亚马逊云搜索支持分面分类（Faceted classification）：
分面分类应用于分面搜索系统，用户在这种系统中能够从多方面进行信息的导航（译者注：如书籍可以从作者、主题、出版日期等不同的分面），多方面对应于不同顺序的分面。 AWS支持按分面控制搜索执行以及对搜索结果排序。同时还支持开发者控制返回的搜索结果中包含的分面数量。所有的分面操作由Search Query Facet（列表5）这个类来实现。 public class SearchQueryFacet {
private String _name;
private int _maxFacets;
private List<String> _constraints;
private FacetSort _sort;
public SearchQueryFacet(String name){
_name = name;
_maxFacets = -1;
_constraints = null;
_sort = FacetSort.none;
}
public SearchQueryFacet(String name, int maxFacets){
_name = name;
_maxFacets = maxFacets;
_constraints = null;
_sort = FacetSort.none;
}
public SearchQueryFacet(String name, int maxFacets, FacetSort sort){
_name = name;
_maxFacets = maxFacets;
_constraints = null;
_sort = sort;
}
public SearchQueryFacet(String name, int maxFacets, List<String> constraints){
_name = name;
_maxFacets = maxFacets;
_constraints = constraints;
_sort = FacetSort.none;
}
public SearchQueryFacet(String name, List<String> constraints){
_name = name;
_maxFacets = -1;
_constraints = constraints;
_sort = FacetSort.none;
}
public SearchQueryFacet(String name, FacetSort sort, List<String> constraints){
_name = name;
_maxFacets = -1;
_constraints = constraints;
_sort = sort;
}
public SearchQueryFacet(String name, FacetSort sort){
_name = name;
_maxFacets = -1;
_constraints = null;
_sort = sort;
}
public String getName() {
return _name;
}
public void setName(String name) {
_name = name;
}
public int getMaxFacets() {
return _maxFacets;
}
public void setMaxFacets(int maxFacets) {
_maxFacets = maxFacets;
}
public FacetSort getSort() {
return _sort;
}
public void setSort(FacetSort sort) {
_sort = sort;
}
public int get_maxFacets() {
return _maxFacets;
}
public void set_maxFacets(int _maxFacets) {
this._maxFacets = _maxFacets;
}
public List<String> getConstraints() {
return _constraints;
}
public void setConstraints(List<String> constraints) {
_constraints = constraints;
}
public void addConstraint(String constraint) {
if(_constraints == null)
_constraints = new LinkedList<String>();
_constraints.add(constraint);
}
@Override
public String toString(){
StringBuffer sb = new StringBuffer();
sb.append("&facet=");
sb.append(_name);
if(_maxFacets > 0){
sb.append("&facet-");
sb.append(_name);
sb.append("-top-n=");
sb.append(_maxFacets);
}
if((_constraints != null) && (_constraints.size() > 0)){
sb.append("&facet-");
sb.append(_name);
sb.append("-constraints=");
boolean first = true;
for(String c : _constraints){
if(!first)
sb.append("%2C");
else
first = false;
sb.append("%27");
sb.append(c);
sb.append("%27");
}
}
if(!_sort.equals(FacetSort.none)){
sb.append("&facet-");
sb.append(_name);
sb.append("-sort=");
sb.append(_sort);
}
return sb.toString();
}
public enum FacetSort{
none, alpha, count, max, sum
}
}
Listing 5 Facets control class
最后Search Query Sort类（列表6）实现了开发者对结果排序的控制。 public class SearchQuerySort {
private List<SearchRank> _ranks;
public SearchQuerySort(){
_ranks = new LinkedList<SearchRank>();
}
public void addRank(SearchRank rank){
_ranks.add(rank);
}
@Override
public String toString(){
if(_ranks.size() == 0)
return null;
StringBuffer sb = new StringBuffer();
sb.append("&rank=");
boolean first = true;
for(SearchRank r : _ranks){
if(!first)
sb.append("%2C");
else
first = false;
sb.append(r);
}
return sb.toString();
}
public static class SearchRank{
private String _name;
private boolean _ascending;
public SearchRank(){
_ascending = true;
}
public SearchRank(String name){
_ascending = true;
_name = name;
}
public SearchRank(String name, boolean ascending){
_ascending = ascending;
_name = name;
}
@Override
public String toString(){
if(_ascending)
return _name;
return "-" + _name;
}
}
}
Listing 6 Sort control class
CloudSearch查询除了将所有的参数汇总到一起，还增加了页码信息和一组返回字段。 这个查询类还提供了一个方法——HTTP查询转换（列表7），将搜索查询的所有部分汇总，并生成能被搜索处理的HTTP字符串。 public String toHttpQuery() throws Exception{
StringBuffer sb = new StringBuffer();
sb.append("?results-type=xml");
if(_size > 0){
sb.append("&size=");
sb.append(_size);
}
if(_start > 0){
sb.append("&start=");
sb.append(_start);
}
if((_fields != null) && (_fields.size() > 0)){
sb.append("&return-fields=");
boolean first = true;
for(String f : _fields){
if(!first)
sb.append("%2C");
else
first = false;
sb.append(f);
}
}
if(_filter != null){
if(_filter instanceof SearchQueryValueFilter)
sb.append("&q=");
else
sb.append("&bq=");
sb.append(URLEncoder.encode(_filter.toString(), "UTF8"));
}
if((_facets != null) && (_facets.size() > 0)){
for(SearchQueryFacet f : _facets){
sb.append(f);
}
}
if((_sorts != null) && (_sorts.size() > 0)){
for(SearchQuerySort s : _sorts){
sb.append(s);
}
}
return sb.toString();
}
Listing 7 Convert to HTTP query method
我们使用Apache HttpComponents来实现与亚马逊云搜索的通信。 测试我们的API我们使用亚马逊提供的IMDB样例来进行验证。首次单元测试（列表8）用于验证我们实现的搜索API。 public class SearchAPITester extends TestCase {
private static final String SearchURL = "search-imdb-movies-ab4fpqw4eocczpgsnrtlu4rn7i.us-east-
1.cloudsearch.amazonaws.com";
private CloudSearchClient client;
protected void setUp() throws Exception {
client = new CloudSearchClient(SearchURL);
}
protected void tearDown() {
client.close();
}
public void testSearch() throws Exception{
SearchQueryValueFilter f1 = new SearchQueryValueFilter("title", "star", false, false);
SearchQueryValueFilter f11 = new SearchQueryValueFilter("title", "war", false, true);
SearchQueryValueFilter f2 = new SearchQueryValueFilter("year", "..2000", true, false);
SearchQueryFilterOperation f12 = new SearchQueryFilterOperation();
f12.setOperation(FilterOperation.or);
f12.addFilters(f1);
f12.addFilters(f11);
SearchQueryFilterOperation f3 = new SearchQueryFilterOperation();
f3.addFilters(f12);
f3.addFilters(f2);
CloudSearchQuery query = new CloudSearchQuery(f3);
query.addField("actor");
query.addField("director");
query.addField("title");
query.addField("year");
SearchQueryFacet sf = new SearchQueryFacet("genre", 5, FacetSort.alpha);
sf.addConstraint("Drama");
sf.addConstraint("Sci-Fi");
query.addFacet(sf);
SearchQuerySort sort = new SearchQuerySort();
SearchRank r1 = new SearchRank("title");
SearchRank r2 = new SearchRank("year", false);
sort.addRank(r1);
sort.addRank(r2);
query.addSort(sort);
try {
System.out.println("Test 1 ");
SearchResults result = client.search(query);
System.out.println(result);
} catch (Exception e) {
e.printStackTrace();
}
}
}
Listing 8 Search API tester
该测试获得的结果（列表9），和直接通过亚马逊REST API获得的结果相同。 SearchResults ] 第二次测试（列表10）用来验证文档的添加和删除。 public class DocumentAPITester extends TestCase {
private static final String DocumentURL = "doc-imdb-movies-ab4fpqw4eocczpgsnrtlu4rn7i.us-east-
1.cloudsearch.amazonaws.com";
private CloudSearchDocumentClient client;
private BatchType batch;
protected void setUp() throws Exception {
client = new CloudSearchDocumentClient(DocumentURL);
FieldType title = new FieldType();
title.setName("title");
title.setValue("The Seeker: The Dark Is Rising");
FieldType director = new FieldType();
director.setName("director");
director.setValue("Cunningham, David L.");
FieldType genrea = new FieldType();
genrea.setName("genre");
genrea.setValue("Adventure");
FieldType genred = new FieldType();
genred.setName("genre");
genred.setValue("Drama");
FieldType genref = new FieldType();
genref.setName("genre");
genref.setValue("Fantasy");
FieldType genret = new FieldType();
genret.setName("genre");
genret.setValue("Thriller");
FieldType actor1 = new FieldType();
actor1.setName("actor");
actor1.setValue("McShane, Ian");
FieldType actor2 = new FieldType();
actor2.setName("actor");
actor2.setValue("Eccleston, Christopher");
FieldType actor3 = new FieldType();
actor3.setName("actor");
actor3.setValue("Conroy, Frances");
FieldType actor4 = new FieldType();
actor4.setName("actor");
actor4.setValue("Conroy, Frances");
FieldType actor5 = new FieldType();
actor5.setName("actor");
actor5.setValue("Ludwig, Alexander");
FieldType actor6 = new FieldType();
actor6.setName("actor");
actor6.setValue("Crewson, Wendy");
FieldType actor7 = new FieldType();
actor7.setName("actor");
actor7.setValue("Warner, Amelia");
FieldType actor8 = new FieldType();
actor8.setName("actor");
actor8.setValue("Cosmo, James");
FieldType actor9 = new FieldType();
actor9.setName("actor");
actor9.setValue("Hickey, John Benjamin");
FieldType actor10 = new FieldType();
actor10.setName("actor");
actor10.setValue("Piddock, Jim");
FieldType actor11 = new FieldType();
actor11.setName("actor");
actor11.setValue("Lockhart, Emma");
AddType add = new AddType();
add.setId("tt0484562");
add.setVersion(1l);
add.setLang("en");
add.getField().add(title);
add.getField().add(director);
add.getField().add(genrea);
add.getField().add(genred);
add.getField().add(genref);
add.getField().add(genret);
add.getField().add(actor1);
add.getField().add(actor2);
add.getField().add(actor3);
add.getField().add(actor4);
add.getField().add(actor5);
add.getField().add(actor6);
add.getField().add(actor7);
add.getField().add(actor8);
add.getField().add(actor9);
add.getField().add(actor10);
add.getField().add(actor11);
DeleteType delete = new DeleteType();
delete.setId("tt0301199");
delete.setVersion(1l);
batch = new BatchType();
batch.getAdd().add(add);
batch.getDelete().add(delete);
}
protected void tearDown() {
client.close();
}
public void testSearch() throws Exception{
try {
System.out.println("Test 1 ");
ResponseType result = client.index(batch);
System.out.println("Status " + result.getStatus() + " Added " + result.getAdds() + " Deleted " +
result.getDeletes());
} catch (Exception e) {
e.printStackTrace();
}
}
}
Listing 10 Document upload tester
测试也获得了预期的结果（列表11） Status SUCCESS Added 1 Deleted 1 Listing 11 Document upload test results 总结上面这些简单的Java API实现了亚马逊云搜索的功能，显著简化了亚马逊云搜索功能在已有Java应用中的使用，必然会扩大应用的影响范围。 关于作者Boris Lublinsky博士是诺基亚首席架构师，主要从事大数据、SOA、BPM、中间件的实现。在此之前Boris曾经是 Herzum软件公司的首席架构师，为客户设计大规模的SOA系统，曾负责CNA保险公司的企业架构，参与了CNA的系统集成与SOA策略的设计和实现， 构建了应用框架并实现了面向服务的架构。Boris在企业、技术架构，软件工程方面有超过25年的经验。他还是OASIS SOA参考模型技术委员会的活跃会员，也是《Applied SOA：Service-Oriented Architecture and Design Strategies》，一书的共同作者。他还发表了大量架构、编程、大数据、SOA、BPM的相关文章。 查看英文原文：Using AWS Cloud Search |
