XAP

Web Application Support

GigaSpaces's integration with the Service Grid allows you to deploy web applications (packaged as WAR files) onto the Service Grid. The integration is built on top of the Service Grid Processing Unit Container.

The integration allows you to make use of the following Service Grid features:

  • Dynamic allocation of several instances of a web application (probably fronted by a load balancer).
  • Management of the instances running (if a GSC fails, the web application instances running on it will be instantiated on a different GSC).
  • SLA monitor- based dynamic allocation and de-allocation of web application instances.

The web application itself is a pure, JEE-based web application. The application can be the most generic web application, and automatically make use of the Service Grid features. The web application can define a Space (either embedded or remote) very easily (either using Spring or not).

The web container used behind the scenes is Jetty. This page will list the common usage and configuration of web containers. Jetty specific configuration and usage can be found here.

Deployment

The integration can deploy either a packaged or an exploded WAR file. In order to deploy packaged WAR file, it can be specified using one of the deployment mechanisms (UI/CLI/Programmatic, see more here).

The deploy client, the GSMs, and the GSCs can run on different machines.

When deploying a WAR file, it goes through the following steps until it gets to the GSC:

  1. Point the deployment tool to the WAR file (UI/CLI/Programmatic).
  2. The WAR file itself is uploaded to the chosen GSM (which acts as the primary GSM of the deployment).
  3. The WAR is extracted under the GSM deploy directory with the provided Processing Unit name. The default directory location is GSRoot/deploy/[processing unit name].
  4. The GSM decides (based on the SLA) how many instances of the web application must be deployed, and deploys them to the available GSCs.
  5. Each GSC that is supposed to run an instance of the web application downloads the web application into its own local file system. By default, the web application is downloaded to GSRoot/work/deployed-processing-units/[processing unit name]_[unique identifier].
  6. The appropriate web container is configured to run the web application using the local file system location.

Deploying an exploded WAR is similar to deploying a packaged WAR, and goes through the following steps:

  1. The exploded WAR file should be copied (manually) over to all the GSMs deploy directory. The default location is GSRoot/deploy.
  2. A deploy command is issued with the Processing Unit name (the name of the directory under the deploy directory).
  3. The GSM decides (based on the SLA) how many instances of the web application must be deployed, and deploys them to the available GSCs.
  4. Each GSC that is supposed to run an instance of the web application downloads the web application into its own local file system. By default, the web application is downloaded to GSRoot/work/deployed-processing-units/[processing unit name]_[unique identifier].
  5. The appropriate web container is configured to run the web application using the local file system location.

The directory where the web applications are extracted (up to the work directory) on the GSC side can be controlled using the com.gs.work system property.

The deploy directory location (up to the deploy directory) used on the GSM side can be controlled using the com.gs.deploy system property.

Web Application Structure

A web application deployed to the Service Grid is just another type of Processing Unit. As such, it inherits all the options that a Processing Unit has, including the ability to define an optional META-INF/spring/pu.xml configuration file as any other Processing Unit. However, that class definitions and libraries that the web application depends on are placed in their standard JEE web application location (i.e. WEB-INF/classes and WEB-INF/lib respectively).

Class Loaders

Below is the structure of the class loaders when several web applications are deployed on the Service Grid:

              Bootstrap (Java)
                  |
               System (Java)
                  |
               Common (Service Grid)
                  |
            JEE Container
             /        \
        WebApp1     WebApp2

The following table shows which user controlled locations end up in which class loader, and the important JAR files that exist within each one:

Class Loader User Locations Built-in JAR Files
Common [GSRoot]/lib/platform/ext/*.jar xap-datagrid.jar
JEE Container JEE container specific jars [GSRoot]/lib/optional/jetty/*.jar
Webapp [PU]/WEB-INF/classes, [PU]/WEB-INF/lib/*.jar xap-openspaces.jar, spring/*.jar

The idea behind the class loaders is to create a completely self-sufficient web application. All relevant JAR files or classes should exist within the web application (as if running it standalone), so that deploying it to the Service Grid is a seamless experience.

The xap-datagrid.jar is a special case, and is automatically removed from WEB-INF/lib if it exists there because it has already been defined in the common class loader.

In terms of class loader delegation model, the web application class loader uses a parent last delegation mode. This means that the web application first tries to load classes from its own class loader, and only if they are not found, delegates up to the parent class loader. This is the recommended way to work with this class loader model.

The Service Class Loader is still used in order to load the JEE container, but it is hidden from the user. The web application class loader is created with its parent class loader being the JEE-container-specific class loader and not the Service Class loader. Users should not worry in this case about the Service Class Loader and how it is used, as the above class loader model provides exactly the same semantics as a plain web container class loader model.

The JEE Class Loader is created lazily when the first web application deployed into the GSC. This eliminates the overhead involved when creating none-web processing units. There can also be several JEE class loaders per web container type, allowing support for both jetty and tomcat (for example) in the future quite easily.

Bootstrap Context Listener

When deploying a web application to the GigaSpaces Service Grid, the web.xml of the web application is automatically changed to include a BootstrapWebApplicationContextListener. The Bootstrap Context Listener provides the following services:

  • Automatically puts the ClusterInfo and BeanLevelProperties that the system was deployed with into the web application ServletContext. The ClusterInfo class is put under a context attribute clusterInfo, and the BeanLevelProperties is put under a context attribute named beanLevelProperties.

  • If there is a META-INF/spring/pu.xml, it loads it using Spring (this is why the Spring and OpenSpaces JARs are automatically added to the web application class loader). The ApplicationContext created is put in the Servlet Context under an attribute named applicationContext. All the beans defined within the META-INF/spring/pu.xml that can be instantiated (because the are either not abstract, or require additional arguments to be created) are bound to the ServletContext as well. Each bean name is the ServletContext attribute name, and each bean is the attribute value.

  • If there is an org.springframework.web.context.ContextLoaderListener it is automatically replaced with ProcessingUnitContextLoaderListener. The ProcessingUnitContextLoaderListener is exactly the same as the Spring one, except that it uses the (optional) ApplicationContext loaded from the META-INF/spring/pu.xml as the parent of the created Application Context. It also adds the ClusterInfo and BeanLevelProperties post processors to the created ApplicationContext so they can be injected to any bean requiring it (such as the embedded Space bean).

Using a Space

There are several ways that a Space (and other components) can be used, and configured within a web application. Some common scenarios are listed below.

Pure Remote Space

A typical usage pattern is connecting remotely to a Space. The following is an example (either using Spring within the web application Spring context file, or using pure Java code):

<os-core:space-proxy id="space" space-name="mySpace" />
<os-core:giga-space id="gigaSpace" space="space"/>
<bean id="space" class="org.openspaces.core.space.SpaceProxyFactoryBean">
    <property name="name" value="space" />
</bean>
<bean id="gigaSpace" class="org.openspaces.core.GigaSpaceFactoryBean">
    <property name="space" ref="space" />
</bean>
SpaceProxyConfigurer configurer = new SpaceProxyConfigurer("mySpace");

GigaSpace gigaSpace = new GigaSpaceConfigurer(configurer).gigaSpace();

// ...

// shutting down .. closing the Space
spaceConfigurer.destroy();

Using pu.xml

A web application is simply a Processing Unit. This means that a META-INF/spring/pu.xml can be added, which can be used to define a Space. Accessing the beans is relatively simple as they are automatically added to the web application context and can be accessed from there. The key they are stored under is the bean ID that each bean corresponds to.

The following is an example that starts an embedded Space as part of the web application within the pu.xml (it is the content of the pu.xml).

<os-core:embedded-space id="space" space-name="mySpace"/>
<os-core:giga-space id="gigaSpace" space="space"/>
<os-core:giga-space id="clusteredGigaSpace" space="space" clustered="true"/>
<bean id="space" class="org.openspaces.core.space.EmbeddedSpaceFactoryBean">
    <property name="name" value="space" />
</bean>
<bean id="gigaSpace" class="org.openspaces.core.GigaSpaceFactoryBean">
    <property name="space" ref="space" />
</bean>
<bean id="clusteredGigaSpace" class="org.openspaces.core.GigaSpaceFactoryBean">
    <property name="space" ref="space" />
    <property name="clustered" ref="true" />
</bean>

The following is an example of a simple JSP that uses it:

GigaSpace gigaSpace = (GigaSpace) getServletContext().getAttribute("clusteredGigaSpace");

Embedded Space

The previous section described several ways of starting an embedded Space within the web application. The recommended approach to working with embedded Space, is to work with its clustered proxy (the clustered flag in GigaSpace set to true) for interactions that originate from a web request. This is mainly because the load balancer doesn't know about routing specific classes to each cluster member.

Event-driven operations should still work with non-clustered embedded Spaces (usually). For example, a web request that results in writing an Order (using the clustered proxy) to the Space, and a polling container that picks it up and processes it asynchronously. The polling container should work with the non-clustered, co-located proxy of the cluster member Space.

Load Balancer

When deploying a highly available web site, a load balancer is usually used to load balance requests between at least two instances of a web container that actually runs the web application. When using GigaSpaces to deploy a web application, running more than one instance of a web application is simple, and also simplifies the manageability and virtualized nature of running web applications.

In order to create a single point of view (in terms of clients connecting to a server), a load balancer is usually used. While there are many different types of load balancers (both hardware and software), solving the load balancing problem is not specific to using GigaSpaces to deploy the web application. Examples of how to configure load balancers can be found in specific web container sections.

GigaSpaces also comes with a built-in integration with the Apache httpd load balancer, as described in the Apache Load Balancer Agent section.