简单之美 | Hive-0.12.0的Web接口HWI安装、配置、改造及使用

使用Hive的HWI接口，可以通过在Web页面上提交HQL查询操作，并浏览查询结果数据。默认情况下，HWI只支持浏览结果数据，不能够下载查询结果文件（当然，HWI可能也是考虑通过Web下载大量的结果数据，对服务器造成压力，或者处于安全方面的考虑）。我们对HWI进行了简单的改造，改造内容主要是增加了一个内置的文件服务器，可以通过页面进行查询，然后下载结果文件。

HWI安装配置

首先，要保证Hadoop集群正常运行，Hive能够正常使用。

先要安装Ant，如下所示：

1
wget http://mirrors.hust.edu.cn/apache//ant/binaries/apache-ant-1.9.4-bin.tar.gz

2
tar xvzf apache-ant-1.9.4-bin.tar.gz

3
ln -s /usr/local/apache-ant-1.9.4-bin /usr/local/ant

修改Hive的环境配置文件hive-env.sh，增加如下配置内容：

`1`	`export` `ANT_LIB=/usr/local/ant`

将如下JAR文件拷贝到${HIVE_HOME}/lib目录下面：

1
// 用于编译JSP文件

2
jasper-compiler-5.5.23.jar

3
jasper-runtime-5.5.23.jar

4
 
5
// 替换默认的servlet-api-2.5-20081211.jar，我使用的是apache-tomcat-7.0.53/lib下面的servlet-api.jar文件

6
servlet-api.jar

另外，由于Hive 0.12.0版本的HWI的问题，需要将你的${JAVA_HOME}/lib/tools.jar文件也加入到${HIVE_HOME}/lib目录下。这个算是一个Bug吧，大概要到Hive 0.13.0版本能够解决。

接下来，需要在${HIVE_HOME}/conf/hive-site.xml配置文件中，加入如下配置内容：

01
<!-- HWI Configuration -->

02
<property>

03
           <name>hive.hwi.listen.host</name>

04
           <value>0.0.0.0</value>

05
           <description>This is the host address the Hive Web Interface will listen on.</description>

06
   </property>

07
<property>

08
           <name>hive.hwi.listen.port</name>

09
           <value>9999</value>

10
           <description>This is the port the Hive Web Interface will listen on.</description>

11
   </property>

12
<property>

13
           <name>hive.hwi.war.file</name>

14
           <value>/lib/hwi.war</value>

15
           <description>This is the WAR file with the jsp content for Hive Web Interface.</description>

16
   </property>

其中，属性hive.hwi.war.file对应的hwi.war文件，我们会在后面改造之后重新打包为该文件，并拷贝到${HIVE_HOME}/lib目录下面。

HWI改造

默认使用HWI是没有查询结果文件下载功能的，可以增加一个文件服务器，用来存放并提供下载查询结果文件。这里，我发现HWI虽然也提供一个Web容器，但是由于封装的太深，所以放弃修改内部源码，而是直接通过我熟悉的方式，使用Jetty实现了一个文件服务，需要用到下面的一些JAR文件：

01
// 用于文件服务器

02
jetty-continuation-8.1.0.RC5.jar 

03
jetty-io-8.1.0.RC5.jar       

04
jetty-server-8.1.0.RC5.jar  

05
jetty-util-8.1.0.RC5.jar   

06
jetty-xml-8.1.0.RC5.jar

07
jetty-http-8.1.0.RC5.jar         

08
jetty-security-8.1.0.RC5.jar 

09
jetty-servlet-8.1.0.RC5.jar 

10
jetty-webapp-8.1.0.RC5.jar

实现文件服务器

首先，定义一个服务器接口，用来控制文件服务器的启停操作等，如下所示：

01
package org.shirdrn.hadoop.hive.jetty;

02
 
03
public interface JettyServer {

04
 
05
     int getPort();

06
 
07
     /** Start this server. */

08
     void start();

09
 
10
     /** Stop this server. */

11
     void close();

12
 
13
     /** Wait for this server to exit. */

14
     void join() throws InterruptedException;

15
     
16
}

然后，实现一个文件服务器，代码如下所示：

01
package org.shirdrn.hadoop.hive.jetty;

02
 
03
import java.util.HashMap;

04
import java.util.Iterator;

05
import java.util.Map;

06
import java.util.Map.Entry;

07
 
08
import org.apache.commons.logging.Log;

09
import org.apache.commons.logging.LogFactory;

10
import org.eclipse.jetty.server.Server;

11
import org.eclipse.jetty.server.handler.DefaultHandler;

12
import org.eclipse.jetty.server.handler.HandlerList;

13
import org.eclipse.jetty.server.handler.ResourceHandler;

14
import org.eclipse.jetty.server.nio.SelectChannelConnector;

15
 
16
/**

17
* Jetty file server for handling hvie query result file downloading.

18
*

19
* @author yanjun

20
*/

21
public class JettyFileServer implements JettyServer {

22
 
23
     private static final Log LOG = LogFactory.getLog(JettyFileServer.class);

24
     private final Server server;

25
     private final int port = 9722;

26
     
27
     static Map<String, String> users = new HashMap<String, String>();

28
     static Map<String, String> resourceBases = new HashMap<String, String>();

29
     static {

30
          users.put("user", "user");

31
          resourceBases.put("user", "/download");

32
     }

33
     
34
     public JettyFileServer() {

35
          server = new Server(port);

36
        SelectChannelConnector connector = new SelectChannelConnector();

37
        server.addConnector(connector);

38
        // create file context for each authorized user

39
        HandlerList handlers = new HandlerList();

40
        createFileServerContexts(handlers);

41
        
42
        handlers.addHandler(new DefaultHandler());

43
        
44
        server.setHandler(handlers);

45
     }

46
 
47
     private void createFileServerContexts(HandlerList handlers) {

48
          Iterator<Entry<String, String>> iter = users.entrySet().iterator();

49
          while(iter.hasNext()) {

50
               Entry<String, String> entry = iter.next();

51
               String user = entry.getKey();

52
               String fileBase = resourceBases.get(user);

53
               ResourceHandler resourceHandler = createResourceHandler(user, fileBase);

54
               handlers.addHandler(resourceHandler);

55
          }

56
     }

57
     
58
     private ResourceHandler createResourceHandler(String user, String fileBase) {

59
          ResourceHandler resourceHandler = new ResourceHandler();

60
        resourceHandler.setDirectoriesListed(true);

61
        resourceHandler.setWelcomeFiles(new String[]{ "index.html" });

62
        resourceHandler.setResourceBase(fileBase);

63
        return resourceHandler;

64
     }

65
     
66
     @Override

67
     public int getPort() {

68
          return port;

69
     }

70
 
71
     @Override

72
     public void start() {

73
          try {

74
               server.start();

75
          } catch (InterruptedException e) {

76
          } catch (Exception e) {

77
               throw new RuntimeException("Fail to start Jetty file server!", e);

78
          }

79
     }

80
 
81
     @Override

82
     public void close() {

83
          try {

84
               server.stop();

85
          } catch (Exception e) {

86
               throw new RuntimeException("Fail to stop Jetty file server!", e);

87
          }         

88
     }

89
 
90
     @Override

91
     public void join() throws InterruptedException {

92
          server.join();

93
     }

94
 
95
}

这里，为了简化，我没有使用配置的方式，配置文件服务器的resource base目录，直接写死路径在代码里面，默认是/download目录，主要的是，这个目录需要设置一下权限，可以允许任何人写，但是不能覆盖别人已经写的文件（执行查询生成的结果文件）：

`1`	`chmod` `1777 /download`

增加一个ServletContextListener

主要用来启动文件服务器，实现类FileServerContextListener的代码如下所示：

01
package org.shirdrn.hadoop.hive.jetty;

02
 
03
import javax.servlet.ServletContextEvent;

04
 
05
import org.apache.commons.logging.Log;

06
import org.apache.commons.logging.LogFactory;

07
 
08
public class FileServerContextListener implements javax.servlet.ServletContextListener {

09
 
10
  protected static final Log l4j = LogFactory.getLog(FileServerContextListener.class

11
      .getName());

12
  private JettyServer fileServer;

13
  
14
  public void contextInitialized(ServletContextEvent sce) {

15
     // start Jetty file server

16
    l4j.info("Start Jetty file server...");

17
    JettyServer server = new JettyFileServer();

18
    server.start();

19
    l4j.info("Jetty file server started!");

20
  }

21
 
22
  public void contextDestroyed(ServletContextEvent sce) {

23
     // stop Jetty file server

24
    fileServer.close();

25
    l4j.info("Jetty file server stopped!");

26
  }

27
}

然后，需要在web.xml文件中增加如下配置：

1
<listener>

2
     <listener-class>org.shirdrn.hadoop.hive.jetty.FileServerContextListener</listener-class>

3
</listener>

增加下载链接，设置DDL、DML操作限制

因为我们可能需要将HWI暴露给组织内部的其他项目团队使用，只允许他们查询，就应该限制Hive的DDL、DML操作，不运行他们建库建表、

这里，需要修改页面session_manage.jsp，修改后的内容，如下所示：

001
<%@page import="org.apache.hadoop.hive.hwi.*" %>

002
<%@page import="java.util.Arrays" %>

003
<%@page import="java.util.List" %>

004
<%@page errorPage="error_page.jsp" %>

005
<% HWISessionManager hs = (HWISessionManager) application.getAttribute("hs");; %>

006
 
007
<% HWIAuth auth = (HWIAuth) session.getAttribute("auth"); %>

008
<% if (auth==null) { %>

009
     <jsp:forward page="/authorize.jsp" />

010
<% } %>

011
<% String sessionName=request.getParameter("sessionName"); %>

012
<% HWISessionItem sess = hs.findSessionItemByName(auth,sessionName); %>

013
<% String message=null; %>

014
<%

015
     String randomFile = String.valueOf(System.currentTimeMillis()) + ".txt"; // 生成随机文件名称

016
     String errorFile=request.getParameter("errorFile");

017
     String resultFile=request.getParameter("resultFile");

018
     resultFile = "/user/download/" + resultFile; // 结果文件存储的路径

019
     String query = request.getParameter("query");

020
     String silent = request.getParameter("silent");

021
     String start = request.getParameter("start");

022
     
023
     // 简单的HQL操作过滤，限制DDL、DML操作

024
    String[] stoppedKeywords = new String[] {

025
         "INSERT", "DELETE", "TRUNCATE",

026
         "CREATE", "DROP", "ALTER",

027
         "GRANT", "REVOKE", "LOAD"

028
    };

029
     List<String> list = Arrays.asList(stoppedKeywords);

030
    if(query != null) {

031
            String q = query.toUpperCase();

032
            String[] a = q.split("\\s+");

033
            for(String w : a) {

034
                 if(list.contains(w)) { // 如果包含上述关键词，直接抛出异常，限制执行操作

035
                      String err = "Permission denied! Excludes operations: " + list;

036
                      throw new HWIException(err);

037
                 }

038
            }

039
       }

040
%>

041
<%

042
  if (request.getParameter("start")!=null ){

043
    if ( sess.getStatus()==HWISessionItem.WebSessionItemStatus.READY){

044
      sess.setErrorFile(errorFile);

045
      sess.setResultFile(resultFile);

046
      sess.clearQueries();

047
      
048
      for (String q : query.split(";") ){

049
        sess.addQuery(q);

050
      }

051
      if (query.length()==0){

052
        message="You did not specify a query";

053
        start="NO";

054
      }

055
      if (silent.equalsIgnoreCase("YES") )

056
     sess.setSSIsSilent(true);

057
      else

058
     sess.setSSIsSilent(false);

059
             
060
     message="Changes accepted.";

061
     if (start.equalsIgnoreCase("YES") ){

062
       sess.clientStart();

063
          message="Session is set to start.";

064
     }

065
      }

066
  }

067
%>

068
<!DOCTYPE html>

069
<html>

070
<head>

071
<title>Manage Session <%=sessionName%></title>

072
<link href="css/bootstrap.min.css" rel="stylesheet">

073
</head>

074
<body style="padding-top: 60px;">

075
    <jsp:include page="/navbar.jsp"></jsp:include>

076
     <div class="container">

077
          <div class="row">

078
               <div class="span4">

079
                    <jsp:include page="/left_navigation.jsp" />

080
               </div><!-- span4 -->

081
               <div class="span8">

082
                    <h2>

083
                         Manage Session

084
                         <%=sessionName%></h2>

085
 
086
                    <% if (message != null) {  %>

087
                    <div class="alert alert-info"><%=message %></div>

088
                    <% } %>

089
 
090
                    <% if (sess.getStatus()==HWISessionItem.WebSessionItemStatus.QUERY_RUNNING) { %>

091
                    <div class="alert alert-warning">Session is in QUERY_RUNNING

092
                         state. Changes are not possible!</div>

093
                    <% } %>

094
 
095
                    <% if (sess.getStatus()==HWISessionItem.WebSessionItemStatus.QUERY_RUNNING){ %>

096
                    <%--

097
               View JobTracker: <a href="<%= sess.getJobTrackerURI() %>">View Job</a><br>

098
               Kill Command: <%= sess.getKillCommand() %>

099
               Session Kill: <a href="/hwi/session_kill.jsp?sessionName=<%=sessionName%>"><%=sessionName%></a><br>

100
               --%>

101
                    <% } %>

102
 
103
                    <div class="btn-group">

104
                         <a class="btn" href="/hwi/session_history.jsp?sessionName=<%=sessionName%>"><i class="icon-book"></i> History</a>

105
                         <a class="btn" href="/hwi/session_diagnostics.jsp?sessionName=<%=sessionName%>"><i class="icon-cog"></i> Diagnostics</a>

106
                         <a class="btn"href="/hwi/session_remove.jsp?sessionName=<%=sessionName%>"><i class="icon-remove"></i> Remove</a>

107
                         <a class="btn"href="/hwi/session_result.jsp?sessionName=<%=sessionName%>"><i class=" icon-download-alt"></i> Result Bucket</a>

108
                    </div>

109
 
110
                    <form action="session_manage.jsp" class="form-horizontal">

111
                         <input type="hidden" name="sessionName" value="<%=sessionName %>">

112
 
113
                         <fieldset>

114
                              <legend>Session Details     </legend>

115
                              <div class="control-group">

116
                                   <label class="control-label" for="fldresfile">Result File</label>

117
                                   <div class="controls">

118
                                        <input id="fldresfile" type="text" name="resultFile"

119
                                             readonly value="<%=randomFile%>">

120
                                        <% if (sess.getResultFile()!=null) { %>

121
                                             <a href="/hwi/view_file.jsp?sessionName=<%=sessionName%>">View File</a>

122
                                             <%

123
                                                  String[] a = sess.getResultFile().split("/");

124
                                                  String file = a[a.length - 1];

125
                                             %>

126
                                             <!-- 增加下载文件的链接 -->

127
                                             <a target="_blank" href="http://192.168.1.105:9722/<%=file%>">Download File</a>

128
                                        <% } %>

129
                                   </div>

130
                              </div>

131
 
132
                              <div class="control-group">

133
                                   <label class="control-label" for="flderrfile">Error File</label>

134
                                   <div class="controls">

135
                                        <input id="flderrfile" type="text" name="errorFile"

136
                                             value="<%

137
                    if (sess.getErrorFile()==null) { out.print(""); } else { out.print(sess.getErrorFile()); }

138
                 %>">

139
                                   </div>

140
                              </div>

141
 
142
                              <div class="control-group">

143
                                   <label class="control-label" for="fldquery">Query</label>

144
                                   <div class="controls">

145
                                        <textarea id="fldquery" name="query" rows="8" cols="70">

146
                                             <%

147
                    if (sess.getQueries()==null) {

148
                      out.print("");

149
                    } else {

150
                         for (String qu: sess.getQueries() ) {

151
                        out.print(qu); out.print(" ; ");

152
                      }    

153
                    }

154
                    %>

155
                                        </textarea>

156
                                   </div>

157
                              </div>

158
 
159
 
160
                              <div class="control-group">

161
                                   <label class="control-label" for="fldsilent">Silent Mode</label>

162
                                   <div class="controls">

163
                                        <select id="fldsilent" name="silent">

164
                                             <option value="YES"

165
                                                  <% if (sess.getSSIsSilent()==true) { out.print("SELECTED=\"TRUE\""); } %>>YES</option>

166
                                             <option value="NO"

167
                                                  <% if (sess.getSSIsSilent()==false) { out.print("SELECTED=\"TRUE\""); } %>>NO</option>

168
                                        </select>

169
                                   </div>

170
                              </div>

171
 
172
                              <div class="control-group">

173
                                   <label class="control-label" for="fldstart">Start Query</label>

174
                                   <div class="controls">

175
                                        <select id="fldstart" name="start">

176
                                             <option value="YES" SELECTED="TRUE">YES</option>

177
                                             <option value="NO">NO</option>

178
                                        </select>

179
                                   </div>

180
                              </div>

181
 
182
                         </fieldset>

183
 
184
                         <h3>Query Return Codes</h3>

185
                         <p>

186
                              <% for (int i=0; i< sess.getQueryRet().size();++i ){ %>

187
                              <%=i%>

188
                              :

189
                              <%=sess.getQueryRet().get(i)%><br>

190
                              <% } %>

191
                         </p>

192
 
193
                         <% if (sess.getStatus()!=HWISessionItem.WebSessionItemStatus.QUERY_RUNNING) { %>

194
                         <div class="form-actions">

195
                              <button type="submit" class="btn btn-primary">Submit</button>

196
                         </div>

197
 
198
                         <% } %>

199
                    </form>

200
               </div><!-- span8 -->

201
          </div><!-- row -->

202
     </div><!-- container -->

203
</body>

204
</html>

上面注释的地方，说明了修改的内容，可以查看。

HWI重新构建

我们只需要将我们自己实现的代码部分，打包到WAR文件里面就可以，因为默认的HWI的Java实现部分，可以在Hive的软件包中找到，可以查看${HIVE_HOME}/lib/hive-hwi-0.12.0.jar。上面我们改造过程中实现了一个Jetty文件服务器，编译完成后，需要删除默认目录${HWI_HOME}\target\WEB-INF\classes下面，不是我们改造后增加的类的类文件，然后就可以执行如下命令构建：

`1`	`cd` `${HWI_HOME}\target`

`2`	`jar cvf hwi.war *`

然后，将生成的hwi文件，拷贝到部署Hive的对应目录${HIVE_HOME}/lib/下面即可，启动HWI服务：

`1`	`hive --service hwi >> /tmp/hwi/hwi.log &`

然后，就可以通过Web页面访问。

HWI使用

访问页面，例如我的是http://10.10.2.245:9999/hwi，然后，可以看到HWI首页，执行如下操作进行查询：

点击“Authorize”，填写“User”和“Groups”的内容，例如都是hadoop，然后点击提交按钮“Submit”；
点击“Create Session”，填写“Session name”，例如MYSESSION[hadoop]，然后点击提交按钮“Submit”；
这时，进入到Manage Session MYSESSION[hadoop]页面，可以在“Query”中输入HQL查询语句，“Start Query”选择“YES”，然后击提交按钮“Submit”；
可以通过“View File”和“Result Bucket”查看结果内容；
可以通过我们改造后的“Download File”链接，进行查询结果文件的下载。

参考链接

https://cwiki.apache.org/confluence/display/Hive/HiveWebInterface#HiveWebInterface-FeaturesofHWI
https://issues.apache.org/jira/browse/HIVE-5132
http://ant.apache.org/bindownload.cgi
http://mvnrepository.com/artifact/org.eclipse.jetty/jetty-server/8.1.0.RC5
http://mvnrepository.com/artifact/org.eclipse.jetty/jetty-servlet/8.1.0.RC5

来源URL：http://shiyanjun.cn/archives/885.html