Results for tag "wildfly"

Application Server Migration: JBoss EE5 to Wildfly EE7

posted by Roberto Cortez on

A few weeks ago, I’ve posted a blog about moving from Java EE 5 to 7. It was mostly about how you could improve your Java EE 5 code with the new Java EE 7 stuff. Now in this post, I’m going to look a little bit into the migration path on the Application Server side.

If you’re using Java EE 5, there is good chance that you are using one of these servers:

There are many other servers supporting Java EE 5, and you could check them out here.

Prelude

I’ve ended up getting most of my experience with JBoss 4x, since the company I was working on at the time was already using it heavily in most of their projects. I hardly had any vote on the matter and just kept the company direction with JBoss.

When we decided to move one of our client critical applications from Java EE 5 to 7, we were faced with the dilemma of which application server to use. Since I was in a technical management position, I was now able to influence that decision. We end up picking Wildfly for the following reasons:

  • Implemented Java EE 7 Full profile
  • Powerful CLI to manage the server
  • Team already familiar with the Java EE implementations shipped with Wildfly

Even though this post looks into JBoss and Wildfly, some of the principles still apply for Application Servers in general. So I hope that this can be useful for other Application Servers users as well. We are currently using Wildfly 8.2.0, but the content discussed in this post should also work with the latest Wildfly version.

Strategy

Performing an Application Server migration, especially one that involves servers so far apart, is never easy. The migration path is not exactly straightforward, because each application ends up using different features of the Application Server. Worse, the application might even be implementing business code supported in these features that might not be available in the target migration server.

Anyway, there are two strategies that you can follow when working on a migration project:

Feature Freeze

As the name implies, you freeze your project to perform the necessary adjustments to migrate the application. It’s probably easier to deal with complexity, but on the other hand it delays business features and creates a non negotiable deadline. It’s very hard to convince stakeholders to go with this strategy, but if you are able, go for it.

Combined

The other alternative is to keep development going and work the migration at the same time. It’s best for the business, but requires much more discipline and planning. You can always partition and split your application into modules and migrate it in small bits. This in the strategy I usually use.

First Steps

You might need some time to completely migrate your application. During that time, you need to keep the old server running as well as the new. For this, you are required to update and duplicate your environments. It’s like branching the code, but in runtime.

Support tools that you use, might need updating as well. Maven plugins for the new server, Jenkins deployments, whatever interacts with the Application Server. It’s a daunting task, since the complexity to manage all these extra environment and branches is not easy.

Walking the Path

There are a couple of details that you need to worry about when thinking about the migration. This is not an extensive list, but are probably the most common topics that you are going to come across.

Classloading

If you don’t run into ClassNotFoundException, NoClassDefFoundError or ClassCastException you might want to consider to play the lottery and win!

This is especially true with JBoss 4.x Classloader. At the time, class loading was (still is, but even more than) an expensive operation, so JBoss used something called the UnifiedClassLoader. This meant that there was no true isolation between applications. EAR archives could look into each other to load libraries. Of course, this was a major headache to manage. The worst part was when you had to deploy your application into a customer using a JBoss server. If you didn’t have control over it, the current deployment could clash with your own.

Wildfly introduced class loading based on modules instead of the usual hierarchical approach. Usually, an application deployed in Wildfly, doesn’t have access to the Application Server libraries unless is stated explicitly with a file descriptor. For Java EE Applications these modules are loaded automatically.

When changing servers, these are the most common issues related to class loading:

  • Missing libraries that were sitting on other applications.
  • Relaying on libraries sitting on the server that were either removed or updated.
  • Libraries used on the application that are now part of the new server.

To fix this you need to tune your project dependencies by adding your removing the required libraries. There is no step by step guide here. Each case needs to be analyzed and fixed accordingly. It’s kinda like trying to untangle a string full of knots.

If you’re using Dependency Plugin is your friend and you can use mvn dependency:tree and mvn dependency:analyze to help you.

Wildfly also provides you with a specific descriptor file jboss-deployment-structure.xml that allows you to fine tune the class loading, by adding or removing dependencies or change the class loader behaviour. Here is an example:

This custom descriptor is adding dependencies from other deployments, namely app-client.jar and even a sub deployment of another EAR in app-ear.ear.app-entity.jar.

Finally, my advice here is to try to keep with the Standards and only introduce additional libraries if absolutely necessary. This will surely reduce your class loading problem and it would make it easier to move to new versions of the server or even change to another server in the future.

General Configuration

In JBoss 4.x, all the configuration was spread around different files: server.xml, jboss-service.xml, login-config.xml and many others. You had to manually edit the files to change the required configuration. This was a tedious work, especially when you didn’t have access to the server and had to document the set of changes for someone else to perform.

In Wildfly most of the configuration goes into configuration/standalone.xml, but I don’t ever edit the file. Wildfly ships with a very powerful Command Line Interface (CLI) that allows you to script pretty much every change that you need to perform on the server. Here is a sample of Undertow configuration:

This is setting up a virtual host called app, making it the default host, removes the default host that comes with Wildfly and activate Single Sign On.

With scripting and the CLI is very easy to spin up a new server from the ground up. You should always prefer this way of changing configuration on the server.

Datasources

In JBoss 4.x, setting up a Datasource only require you to copy the database driver to the lib folder and create a *-ds.xml file with the Datasource connection details.

In Wildfly, is a little more tricky, but not a big deal. You set up the Datasource as a module and then you can use the CLI to add the Datasource connection details to the server configuration. I even wrote an entire blog post about this in the past: Configure JBoss / Wildfly Datasource with Maven.

Security

Security in JBoss 4.x was set up in conf/login-config.xml. Not many changes were introduced with Wildfly, but if you need to implement a custom login module, the dependencies changed. I’ve also written an entire blog post about it: Custom Principal and LoginModule for Wildfly.

JNDI Bindings

It was common to use @LocalBinding in JBoss 4.x to define the exact JNDI name for your EJB. But Java EE 7 introduced standard JNDI names by scope, meaning that you should follow the convention to lookup EJB’s.

Instead of:

You can:

When Wildfly is starting you can also check the standard bindings in the log:

Other Stuff

This are more specific topics that I’ve also wrote blog posts about, and might be interesting as well:

Spring Batch as Wildfly Module
Wildfly, Apache CXF and @SchemaValidation

Final Words

As stated, migrations never follow a direct path. Still, there are a couple of things that you can do to improve. Write tests, tests and tests. Did I tell you to write tests yet? Do it before working on any migration stuff. Even if everything with the migration seems fine, you might encounter slight behaviour changes between the different versions of the Java EE implementations.

Also, don’t underestimate the job. Keeping your application working with new features being developed, plus changing a server requires you to invest time and effort to make sure that nothing is going to break. Definitely it won’t take you 1 week, unless we are talking about a very tiny application. We took almost 2 years to migrate an application over 1 Million lines. But take these numbers lightly. These are very dependent on your team dynamics.

My final advice: if you are sitting in an old Java EE version, you should definitely migrate. Have a look in my blog about Reduce Legacy from Java EE 5 to 7. The jump is not easy, but with each new version of Java EE release and the concern about standardization, each upgrade should become less painful.

Wildfly, Apache CXF and @SchemaValidation

posted by Roberto Cortez on
tags: , ,

Over the last few days, I have been working on an application migration from JBoss 4 to Wildfly 8. The application is using different technologies, but we are going to focus here on XML Web Services, JAX-WS. Yeah, I know that they are not trendy anymore, but these were developed a long time ago and need to be maintained for compatibility issues.

Anyway, the path to migrate these services was not so easy. I’m sharing some of the problems and fixes with the hope that these could help other developers out there stuck with the same problems.

Sample Definition

Here is a sample of a Web Service definition in the old system, JBoss 4:

Luckily, most of the definition is using standard Java EE annotations. Only @org.jboss.ws.annotation.EndpointConfig and @org.jboss.ws.annotation.SchemaValidation are from the old JBossWS libraries.

We can easily get rid of @org.jboss.ws.annotation.EndpointConfig since we are not going to need it in the new application. For reference, it’s used to set up extra configuration data to be predefined with an endpoint. Check the documentation Predefined client and endpoint configurations.

We want to keep @org.jboss.ws.annotation.SchemaValidation. For reference, this annotation validates incoming and outgoing SOAP messages against the relevant schema in the endpoint wsdl contract. Since the annotation no longer exists in JBossWS we have to use Apache CXF, which is the underlying implementation for JAX-WS on Wildfly.

Problems

Here are a few of the problems I’ve faced:

SchemaValidation Annotation

The annotation @org.jboss.ws.annotation.SchemaValidation doesn’t exist anymore. You have to use the annotation org.apache.cxf.annotations.SchemaValidation from Apache CXF.

Add the following Maven dependency to use the Apache CXF annotation:

Also, notice that in the original annotation we could define an errorHandler property. The old application used a custom error handler to set a custom error message on schema validation errors. There is no equivalent in the new annotation, so we need to do it in another way. To replicate the old behaviour I’ve used Apache CXF Interceptors. Create an interceptor class and extend AbstractPhaseInterceptor. Here is a sample:

And you can use it like this:

Interceptors are used by both CXF clients and CXF servers. There are incoming and outgoing interceptor chains being executed for regular processing and also when an error occurs. In this case, we want to override the Schema Validation message, so we need to bind our interceptor in the error outgoing interceptor chain. You can use the annotation @OutFaultInterceptors for that behaviour. Each chain is split into phases. You define the phase where you want the interceptor to run by passing the Phase.MARSHAL in the constructor. There are other phases, but since we want to change the error message we do it in the MARSHAL phase.

Different WSDL

The old Web Services had the WSDL file being auto generated on deploy time. Unfortunately, in some situations, the WSDL generated by JBoss 4 and Wildfly 8 are different. This can cause problems with your external callers. In this case the main problem was in the Schema Validation. Requests that were valid in JBoss 4 were not valid anymore when being executed in Wildfly 8.

The reason for this behaviour was in the target namespaces. If you are using annotated @XmlRootElement pojos in your Web Service parameters, without defining the namespace property in the annotation, JBoss 4 WS generated the target WSDL element with a black namespace. Apache CXF will use the Web Service default namespace to bind the WSDL elements if they are blank. For reference, this is done in CXF code: org.apache.cxf.jaxws.support.JaxWsServiceConfiguration#getParameterName.

This could be fixed by changing the CXF code, but we opted to place the old generated WSDL file in the migrated application sources and include it in the distribution. It’s not auto generated anymore, meaning that we need to manually generate the WSDL if we change the API. We need to be careful to make sure that we are not breaking anything in the WSDL. This approach seemed better than having to maintain our own CXF version. We could probably submit a fix for this as well, but we believe that JBoss 4 behaviour was not intended.

Start CXF

To use specific API’s from CXF, is not enough to have a project dependency for it. In fact, the first few times I’ve tried the changes, nothing related with CXF seemed to work. This happens because Wildfly it’s only looking for the standard Java EE JAX-WS annotations. To have all the CXF behaviour working, we need to tell Wildfly that our application depends on CXF, even if the libs are already on the server. Yeah, it’s a bit confusing.

The application is deployed in a EAR file. So you need to create a jboss-deployment-structure.xml and add the following content:

Using a MANIFEST.MF in the WAR file apparently doesn’t work if it’s deployed inside an EAR file. For more information, please check Class Loading in WildFly.

If you want to use other CXF features, especially the ones linked with Spring, thing might be a bit trickier. Have a look into this post: Assorted facts about JBoss. Fact 6: JBoss and CXF: match made in heaven.

Final Definition

This should be our final definition for our Web Service:

As you can see, the required changes to migrate a Web Service from JBoss 4 to Wildfly are just a few. However, there are a few minor details that can block you for a long time if you don’t know the details. Maybe you have a different setup and the problems that you face are different. This can also help if you are just trying to setup CXF with Wildfly Anyway, I hope that this post can be useful to you.

Configure JBoss / Wildfly Datasource with Maven

posted by Roberto Cortez on

Most Java EE applications use database access in their business logic, so developers are often faced with the need to configure drivers and database connection properties in the application server. In this post, we are going to automate that task for JBoss / Wildfly and a Postgre database using Maven. The work is based on my World of Warcraft Auctions Batch application from the previous post.

Maven Configuration

Let’s start by adding the following to our pom.xml:

We are going to use the Wildfly Maven Plugin to execute scripts with commands in the application server. Note that we also added a dependency to the Postgre driver. This is for Maven to download the dependency, because we are going to need it later to add it to the server. There is also a ${cli.file} property that is going to be assigned to a profile. This is to indicate which script we want to execute.

Let’s also add the following to the pom.xml:

With the Resources Maven Plugin we are going to filter the script files contained in the src/main/resources/scripts and replace them with the properties contained in ${basedir}/src/main/resources/configuration.properties file.

Finally lets add a few Maven profiles to the pom.xml, with the scripts that we want to run:

Wildfly Script Files

Add Driver

The scripts with the commands to add a Driver:

wildfly-install-postgre-driver.cli

Database drivers are added to Wildfly as a module. In this was, the driver is widely available to all the applications deployed in the server. With ${settings.localRepository} we are pointing into the database driver jar downloaded to your local Maven repository. Remember the dependency that we added into the Wildfly Maven Plugin? It’s to download the driver when you run the plugin and add it to the server. Now, to run the script we execute (you need to have the application server running):

mvn process-resources wildfly:execute-commands -P "install-driver"

The process-resources lifecycle is needed to replace the properties in the script file. In my case ${settings.localRepository} is replaced by /Users/radcortez/.m3/repository/. Check the target/scripts folder. After running the command, you should see the following output in the Maven log:

And on the server:

wildfly-remove-postgre-driver.cli

This script is to remove the driver from the application server. Execute mvn wildfly:execute-commands -P "remove-driver". You don’t need process-resources if you already executed the command before, unless you change the scripts.

Add Datasource

wow-auctions-install.cli
The scripts with the commands to add a Datasource:

We also need a a file to define the properties:

configuration.properties

Default Java EE 7 Datasource

Java EE 7, specifies that the container should provide a default Datasource. Instead of defining a Datasource with the JNDI name java:/datasources/WowAuctionsDS in the application, we are going to point our newly created datasource to the default one with /subsystem=ee/service=default-bindings:write-attribute(name="datasource", value="${datasource.jndi}"). In this way, we don’t need to change anything in the application. Execute the script with mvn wildfly:execute-commands -P "install-wow-auctions". You should get the following Maven output:

And on the server:

wow-auctions-remove.cli

This is the script to remove the Datasource and revert the Java EE 7 default Datasource. Run it by executing mvn wildfly:execute-commands -P "remove-wow-auctions"

Conclusion

This post demonstrated how to automate add / remove Drivers to Wildfly instances and also add / remove Datasources. This is useful if you want to switch between databases or if you’re configuring a server from the ground up. Think about CI environments. These scripts are also easily adjustable to other drivers.

You can get the code from the WoW Auctions Github repo, which uses this setup. Enjoy!

Spring Batch as Wildfly Module

posted by Roberto Cortez on

(TL;DR – Get me to the code)

For a long time, the Java EE specification was lacking a Batch Processing API. Today, this is an essential necessity for enterprise applications. This was finally fixed with the JSR-352 Batch Applications for the Java Platform now available in Java EE 7. The JSR-352 got it’s inspiration from the Spring Batch counterpart. Both cover the same concepts, although the resulting API’s are a bit different.

Since the Spring team also collaborated in the JSR-352, it was only a matter of time for them to provide an implementation based on Spring Batch. The latest major version of Spring Batch (version 3), now supports the JSR-352.

I’m a Spring Batch user for many years and I’ve always enjoyed that the technology had a interesting set of built-in readers and writers. These allowed you to perform the most common operations required by batch processing. Do you need to read data from a database? You could use JdbcCursorItemReader, how about writing data in a fixed format? Use FlatFileItemWriter, and so on.

Unfortunately, JSR-352 implementations do not have the amount of readers and writers available in Spring Batch. We have to remember that JSR-352 is very recent and didn’t have time to catch up. jBeret, the Wildfly implementation for JSR-352 already provides a few custom readers and writers.

What’s the point?

I was hoping that with the latest release, all the readers and writers from the original Spring Batch would be available as well. This is not the case yet, since it would take a lot of work, but there are plans to make them available in future versions. This would allow us to migrate native Spring Batch applications into JSR-352. We still have the issue of the implementation vendor lock-in, but it may be interesting in some cases.

Motivation

I’m one of the main test contributors for the Java EE Samples in the JSR-352 specification. I wanted to find out if the tests I’ve implemented have the same behaviour using the Spring Batch implementation. How can we do that?

Code

I think this exercise is not only interesting because of the original motivation, but it’s also useful to learn about modules and class loading on Wildfly. First we need to decide how are we going to deploy the needed Spring Batch dependencies. We could deploy them directly with the application, or use a Wildfly module. Modules have the advantage to be bundled directly into the application server and can be reused by all deployed applications.

Adding Wildfly Module with Maven

With a bit of work it’s possible to add the module automatically with the Wildfly Maven Plugin and the CLI (command line). Let’s start to create two files that represent the CLI commands that we need to create and remove the module:

wildfly-add-spring-batch.cli

The module --name is important. We’re going to need it to reference it in our application. The --resources is a pain, since you need to indicate a full classpath to all the required module dependencies, but we’re generating the paths in the next few steps.

wildfly-remove-spring-batch.cli

Note wildfly.module.classpath. This property will hold the complete classpath for the required Spring Batch dependencies. We can generate it with Maven Dependency plugin:

This is going to pick all dependencies (including transitive), exclude javax (since they are already present in Wildfly) and exclude test scope dependencies. We need the following dependencies for Spring Batch:

Now, we need to replace the property in the file. Let’s use Maven Resources plugin:

This will filter the configured files and replace the property wildfly.module.classpath with the value we generated previously. This is a classpath pointing to the dependencies in your local Maven repository. Now with Wildfly Maven Plugin we can execute this script (you need to have Wildfly running):

And these profiles:

(For the full pom.xml contents, check here)

We can add the module by executing:
mvn process-resources wildfly:execute-commands -P install-spring-batch.

Or remove the module by executing:
mvn wildfly:execute-commands -P remove-spring-batch.

This strategy works for any module that you want to create into Wildfly. Think about adding a JDBC driver. You usually use a module to add it into the server, but all the documentation I’ve found about this is always a manual process. This works great for CI builds, so you can have everything you need to setup your environment.

Use Spring-Batch

Ok, I have my module there, but how can I instruct Wildfly to use it instead of jBeret? We need to add a the following file in META-INF folder of our application:

jboss-deployment-structure.xml

Since the JSR-352 uses a Service Loader to load the implementation, the only possible outcome would be to load the service specified in org.springframework.batch module. Your batch code will now run with the Spring Batch implementation.

Testing

The github repository code, has Arquillian sample tests that demonstrate the behaviour. Check the Resources section below.

Resources

You can clone a full working copy from my github repository. You can find instructions there to deploy it.

Wildfly – Spring Batch

Since I may modify the code in the future, you can download the original source of this post from the release 1.0. In alternative, clone the repo, and checkout the tag from release 1.0 with the following command: git checkout 1.0.

Future

I’ve still need to apply this to the Java EE Samples. It’s on my TODO list.

Custom Principal and LoginModule for Wildfly

posted by Roberto Cortez on

Did you ever had the need to implement your own custom JAAS Principal and LoginModule for you JEE application? There are a couple of reasons for it. I’ve done it in the following cases:

  • Authenticate the user using different strategies.
  • Have additional user information on the Principal object.
  • Share user information between applications using the Principal object.

Maybe you have your own specific reason, it doesn’t matter. Today’s post will guide you on how to do it for Wildfly. There are a few articles on the topic, but each of them deal with a different aspect of the problem. I got motivated to write this post to aggregate all the steps in a single article, including the Arquillian test.

Wildfly uses PicketBox for Java Application Security and already implements some handy classes to take care of the authentication of the user for you. Have a look into UsersRolesLoginModule, DatabaseServerLoginModule, LdapUsersLoginModule, BaseCertLoginModule and so on. Let’s start by creating a Maven project with the following dependency:

Next, just create a CustomPrincipal and a CustomLoginModule classes:

Note that we’re extending the org.jboss.security.SimplePrincipal present in PicketBox, but you can also implement java.security.Principal instead.

Here again we’re extending a PicketBox class, org.jboss.security.auth.spi.UsersRolesLoginModule. You can code your own login module by implementing javax.security.auth.spi.LoginModule, but I recommend to extend one of the PicketBox classes, since they already have a lot of the behaviour that you will need. org.jboss.security.auth.spi.UsersRolesLoginModule is a very simple login module that authenticates an user by matching his login and password to a file. You shouldn’t use it for production applications, but it’s very handy for prototypes.

CustomLoginModule is also overriding two methods. These are needed to access our CustomPrincipal in a JEE application. The login() method as the name says is called when the user is performing the login action, so in here we create our CustomPrincipal object. On the other hand, the getIdentity() method is called to return the Principal that corresponds to the user primary identity, so we return our own instance if the login was successful.

Ok, great. How do we test it now? We use Arquillian, JUnit and HttpUnit. Start by adding the needed Maven dependencies (these are all the project dependencies):

Note that we also included the javaee-api 7 dependency. Next, we’re creating a simple EJB to access our CustomPrincipal and a Servlet to perform the authentication:

Now the Servlet:

I think the code is self-explanatory, but we still need to wire everything together. The way we configure our login module to be used on Wildfly and on our application is by using Security Domains. You can add a Security Domain by hand using the standalone/configuration/standalone.xml and domain/configuration/domain.xml files on the Wildfly installation folder, but we’re going to do something more interesting.

Using the Command Line Interface (CLI), is really easy to make changes to the server configuration without modifying any XML. This also allows you to setup a test environment and clean up your changes in the end. To achieve that, create the following files:

jboss-add-login-module.cli

As you probably have guessed, jboss-add-login-module.cli contains the CLI commands to add our Security Domain to the Wildfly instance. We first add the Security Domain and then assign the login modules to the domain. Both commands should be able to be executed as a single command, but for some reason I was getting an error so I had to split them apart. The configuration is not available on the server unless you perform a reload command. For the login module, please note the FQN of our CustomLoginModule associated with the configuration. That’s how we wire the custom login module to the Security Domain. The configuration also references two other files: user.properties and roles.properties that are used to perform the user credential verification and load the user roles. Here are examples for both files:

user.properties

Define all valid usernames and their corresponding passwords.

roles.properties

Define the sets of roles for valid usernames.

We still need the CLI commands to remove the Security Domain at the cleanup phase:

jboss-remove-login-module.cli

Almost done? Not yet! We still need to associate our Security Domain to our JEE application so our custom code runs when we perform authentication or execute Principal related behaviour. We need the following files now:

jboss-web.xml

The previous file sets the Security Domain into Servlets.

jboss-ejb3.xml

This file sets the Security Domain in EJB’s.

Uff! Now we’re finally ready to see some action! We required a bit of setup to have this example working! Here is the test class:

That’s it! Now your servlet login method will authenticate using your custom login module and methods like getUserPrincipal() from the servlet request or getCallerPrincipal() from the EJBContext will return the CustomPrincipal instance.

Fire up a Wildfly instance and run the test using mvm test or just use your favourite IDE.

A few problems:

  • It should be possible to add the security domain using only one single CLI command, but for some reason I was getting an erro. I need to have a better look into this.
  • I couldn’t find a way to run code before and after the Arquillian deployment, so the code to add the Security Domain is inside the deployment method. I’ll try to find a way to do it.

If you want additional information, please check the following references:

If you’re too lazy to write the code on your own or just want a working sample, you can download it here. Enjoy!