Apache in front of Tomcat and JBoss

Recently I have to deploy two Java web applications on a single server (one IP address). Where static content was to be served by Apache Http web server instead of Java application servers. When we have multiple applications deployed on an instance of Tomcat or JBoss we have their URLs something like http://server:port/appA, http://server:port/appB etc. This kind of URLs does not look good when we publish it to external world. So instead of 'http://domainA.tld/appA' we wanted to have URL as 'http://domainA.tld/'. This can be achieved by running applications on different instances of Tomcat / JBoss and fronting them with Apache using virtual hosts.

Running multiple instances of JBoss on a machine with single IP address is a nightmare, while it is much easier to run several Tomcat instances, as one has to change just four ports (8005, 8080, 8443, 8009). An application can be run at root context in Tomcat by adding a line like below, inside <host> element in $TOMCATHOME/conf/server.xml-

    <context path="" docbase="appA.war" unpackWAR="false" />

There are pretty decent documentation available on Apache, Tomcat and JBoss' website about their installation and configuration. So I will leave that discussion here itself. Though there is lots of documentation available for Apache virtual hosting also but it takes lots of effort for a newbie to do the configuration and put it in front of application servers. Here I am assuming that Apache Tomcat Connector (mod_jk) is also installed along with Apache web server.

To establish communication between Apache and application server, first we need to define connector workers as shown below-

    worker.list=appA,appB

    # Define appA
    worker.appA.port=9009
    worker.appA.host=localhost
    worker.appA.type=ajp13
    worker.appA.lbfactor=1

    # Define appB
    worker.appB.port=8009
    worker.appB.host=localhost
    worker.appB.type=ajp13
    worker.appB.lbfactor=1

In the above configuration appA's server's connector is listening on port 9009 and appB's on 8009. This connector port is different from http port (8080 by default for Tomcat). Now define the virtual hosts, this can be done either in Apache's httpd.conf or in a separate file and include that in httpd.conf.

   NameVirtualHost *:80

   LoadModule jk_module modules/mod_jk.so

   # Where to find workers defined above
   JkWorkersFile conf/extra/worker.properties

   <virtualhost *:80>
       ServerAdmin someone@domainA.tld
       DocumentRoot "/path/to/the/content"
       ServerName domainA.tld
       ErrorLog "logs/domainA-error.log"
       CustomLog "logs/domainA-access.log" common

       JkMount / appA
       JkMount /* appA
       JkUnMount /static|/* appA

       JkLogFile logs/mod_jk_appA.log
       JkLogLevel info
       JkLogStampFormat "[%a %b %d %H:%M:%S %Y]"
       JkOptions +ForwardKeySize +ForwardURICompatUnparsed -ForwardDirectories
       JkRequestLogFormat "%w %V %T"
   </virtualhost>

   <virtualhost *:80>
       ServerAdmin someone@domainB.tld
       DocumentRoot "/path/to/the/content"
       ServerName domainB.tld
       ErrorLog "logs/domainB-error.log"
       CustomLog "logs/domainB-access.log" common

       JkMount / appB
       JkMount /* appB
       JkUnMount /static|/* appB

       JkLogFile logs/mod_jk_appB.log
       JkLogLevel info
       JkLogStampFormat "[%a %b %d %H:%M:%S %Y]"
       JkOptions +ForwardKeySize +ForwardURICompatUnparsed -ForwardDirectories
       JkRequestLogFormat "%w %V %T"
   </virtualhost>

With above configuration all requests to domainA.tld are served by worker appA and domainB.tld is served by worker appB. The 'JkUnMount /static|/* appA' causes any URL like http://domainA.tld/static/* to be served by Apache instead of the application server. As Apache performs far better while serving the static content like images, html, css, java script etc. so it is good to unmount their URLs from the mod_jk worker. Most of the Jk* properties can be defined out of the <virtualhost> element at global level, that will avoid duplicate entries. Though defining them inside <virtualhost> element gives flexibility to have different values for each virtual host. Detailed information about Jk* is available here.

I have tested above configuration with following environment-

• Java: 1.6.0_10
• Tomcat: 6.0.18
• JBoss: 5.0.0.GA
• Apache: 2.2.11
• OS: Fedora 9, Red Hat Enterprise Linux 5 and Windows Vista

Hope this will be useful to others and save their time.

JNDI lookup on Tomcat and JBoss using Spring

Unfortunately Java EE specs does not specify any standard way of JNDI naming conventions, hence most of the application servers have their own way of JNDI naming. On specifying a Datasource's JNDI name as 'jdbc/myDatasource', Tomcat (6) binds that as 'java:/comp/env/jdbc/myDatasource' while JBoss (5) binds as 'java:/jdbc/myDatasource'. So if one wants to deploy the application on multiple application servers then at least JNDI names has to be changed. I was wondering if Spring Framework has a solution for this, so tried to do lookup using <jee:jndi-lookup ...> as shown below-

<jee:jndi-lookup id="dataSource" jndi-name="jdbc/myDatasource" resource-ref="true"/>

it works well with Tomcat but fails on JBoss. Reason, the JNDI prefix is hard coded in Spring to 'java:comp/env/'.

package org.springframework.jndi;

public abstract class JndiLocatorSupport extends JndiAccessor {

    /** JNDI prefix used in a J2EE container */
    public static final String CONTAINER_PREFIX = "java:comp/env/";
    . . .
}

So it is obvious that it will fail on JBoss. To overcome this limitation, I came up with following solution-

public final class ServiceLocator {

    private static final Map<String, Object> services = new ConcurrentHashMap<String, Object>();

    private static ServiceLocator instance;

    private static Context context;

    static {
        try {
            Context initContext = new InitialContext();
            if (ServerDetector.isJBoss()) {
                context = (Context) initContext.lookup("java:");
            } else if (ServerDetector.isTomcat()) {
                context = (Context) initContext.lookup("java:/comp/env");
            } else {
                context = initContext;
                // or add more 'else if' blocks according to servers to be supported
            }
        } catch (Exception e) {
            throw new ExceptionInInitializerError(e);
        }
    }

    public DataSource getDataSource(String name) throws Exception {
        if (name == null || name.length() <= 0)
            throw new IllegalArgumentException("name");

        if (services.containsKey(name))
            return (DataSource) services.get(name);

        DataSource ds = (DataSource) context.lookup(name);

        services.put(name, ds);
        return ds;
    }
}

Here ServerDetector is a good utility class I found in Liferay code base. Now ServiceLocator takes care of application server specific JNDI prefixes and avoids hassles of changing configuration or code to deploy it on a specific server. In case of Spring instead of using <jee:jndi-lookup ...> one can do the lookup using above-mentioned ServiceLocator as shown below-

    <bean id="serviceLocator" class="com.vinodsingh.ServiceLocator" factory-method="getInstance" />

    <bean id="dataSource" factory-bean="serviceLocator" factory-method="getDataSource">
        <constructor-arg value="jdbc/myDatasource" />
    </bean>

Now entire code (including configuration files) becomes truly portable, at least for JNDI lookups :-)

Let Spring load service class for JAX-WS

During deployment JAX-WS RI by default creates an instance (singleton) of web service implementation class and uses that to serve requests. At times we may want to customize the instantiation of the web service implementation class e.g. let Spring configure and load the object.

JAX-WS Commons has an extension for Spring integration, which delegates total configuration to Spring and makes RI specific files (sun-jaxws.xml) redundant. Currently I am happy with the way JAX-WS RI exposes web services and do not want everything to move to Spring or something else. I was just looking for a way to load the service implementation class by my code instead of RI doing so. JAX-WS RI has a poorly documented (or may be I am not able find it easily) feature called InstanceResolver, which lets developers to customize the way service implementation class is loaded. The code snippets below demostrate how to so-

public class MyResolver extends InstanceResolver<Object> {

    private final Object serviceImpl;

    /**
     * @param clazz
     */
    public MyResolver(Class<?> clazz) {
        serviceImpl= . . . load the class here or get it from Spring or somewhere else . . .;
    }

    /**
     * @see com.sun.xml.ws.api.server.InstanceResolver#resolve(com.sun.xml.ws.api.message.Packet)
     */
    @Override
    public Object resolve(Packet request) {
        return serviceImpl;
    }

}

// Now create an Annotation, which has @InstanceResolverAnnotation with MyResolver.class
@Retention(RetentionPolicy.RUNTIME)
@Target(ElementType.TYPE)
@Documented
@InstanceResolverAnnotation(MyResolver.class)
public @interface MyWs {
}

// Now annotate the service class with above-mentioned annotation
@WebService(endpointInterface = "com.vinodsingh.Address")
@MyWs
public class AddressImpl implements Address {
    . . .
}

During deployment (WSServletContextListener) JAX-WS RI iterates through all annotations on the service class and sees if there is any annotation annotated with @InstanceResolverAnnotation. On finding one it will delegate the task of creating instance of the service class to the user defined InstanceResolver. With this I am able to load service classes using Spring.