Raible Designs

JA-SIG Keynote: Comparing Java Web Frameworks

This morning I did my first keynote at the JA-SIG Summer Conference in Denver. My talk was on Comparing Java Web Frameworks. I told attendees I'd post it here afterwards, so here it is:Download Comparing Java Web Frameworks Presentation (1.1 MB)

In addition, I mentioned my Java Web Frameworks Sweetspots Whitepaper.

Will I be comparing web frameworks at conferences for the rest of my life? Possibly. I've been submitting 2-3 proposals to conferences and it's the only one that keeps getting selected. I'll be delivering it at OSCON, JavaZone, Colorado Software Summit and ApacheCon US.

The Colorado Software Summit wants to have an original presentation - so I may need to drop a framework or two and add in Seam, Grails and GWT. If you are planning on attending one of these talks, which frameworks would you like to see compared?

Related: Comments after I delivered this presentation at ApacheCon EU.

JAR Hell with XFire 1.2.6

I discovered something somewhat disturbing last week. As part of AppFuse 2.0 M5, we added "xfire-all" as a dependency so web services could be supported out-of-the-box. What I didn't know is that xfire-all has transitive dependencies to 40 other libraries, which total 13.4 MB in size. Yikes!

Does XFire take the cake for the most bloated library you can use or are we just including too much (xfire-all vs. fine-grained dependencies)? I tried changing to the recommended Maven configuration and there's still 28 JARs added by XFire.

The WEB-INF/lib directory of a basic Struts 2 + Spring + Hibernate AppFuse application is already 19.2 MB to start. Adding XFire for web services increases the size to 29.2 MB. While disk space may be cheap, some users have noticed "mvn jetty:run" is much slower with XFire (presumably from the JAR processing that happens at startup). Is there an uber XFire JAR we can use instead?

OSCache vs. EhCache for Hibernate's 2nd Level Cache

Hibernate has a number of options for configuring its second level cache. For more information on configuring this, you might want to read John Ferguson Smart's article titled Speed Up Your Hibernate Applications with Second-Level Caching.

Up until today, I thought EhCache was the default cache provider, but apparently not anymore. From Hibernate's documentation:

Note that versions prior to 3.2 defaulted to use EhCache as the default cache provider; that is no longer the case as of 3.2.

So what's the default now? It can't be Hashtable since that's not for production use. I doubt it's OSCache since OSCache can't even get its patches into Hibernate. Looking through the release notes, I found out it's NoCacheProvider - seemingly because of an issue with EhCache:

Due to the upgrade to EhCache1.2 and its new non-singleton cache setup, we should no longer default the cache provider to be ehcache. Instead, default to NoCacheProvider.

That's reasonable I guess. EhCache added support for distributed caching in 1.2. It's a shame they didn't maintain backwards compatibility or they'd still be the default caching provider. Regardless, it doesn't matter who the default caching provider is because it's very easy to change it. Here's how it's configured on one of my projects:

<bean id="sessionFactory" 
    class="org.springframework.orm.hibernate3.annotation.AnnotationSessionFactoryBean">
    <property name="dataSource" ref="dataSource"/>
    <property name="configLocation" value="classpath:hibernate.cfg.xml"/>
    <property name="hibernateProperties">
        <value>
            hibernate.dialect=${hibernate.dialect}
            hibernate.query.substitutions=true 'Y', false 'N'
            hibernate.cache.use_second_level_cache=true
            hibernate.cache.provider_class=org.hibernate.cache.EhCacheProvider
        </value>
    </property>
</bean>

Of course, you can also configure it directly in hibernate.cfg.xml or a hibernate.properties file.

This leads me to the reason for this post:

What is the best 2nd level (clustered) cache to use for Hibernate?
I'm sure some folks will say Coherence, so let's narrow the question to what's the best open source option?

I've used OSCache in the past. It worked well, but it was kind of annoying that I had to patch Hibernate to make it work. The Hibernate folks say it's OSCache's fault, the OSCache guys say it's Hibernate's fault - so this issue will likely never get resolved. So what about EhCache? I don't know, I've only used it in a single JVM environment and haven't tried it in a clustered environment. Is there anyone using Hibernate + EhCache in production that can verify its effectiveness?

Of the options listed in Hibernate's documentation, the only other options seem to be JBoss TreeCache and SwarmCache. You can quickly eliminate SwarmCache since it never made it past 1.0 RC2 in October of 2003.

That leaves JBoss TreeCache, EhCache and OSCache as choices for a clusterable 2nd-level cache. OSCache is an invalidating cache, which definitely works - but might not work as you expect it to. JBoss Cache only seems to allow a replicated cache which also works. EhCache seems to support both. I don't know if invalidating or replicating is better, but I imagine replicating can get quite chatty if you're dealing with large amounts of data.

But wait - is there another open source option? According to Terracotta's CTO, Terracotta is much faster than JBoss Cache. However, if you read about it on DZone, you'll see that JBoss Cache has no "official" benchmarks.

So what's a developer to do? My current client likes OSCache, but I'm leaning towards EhCache. Which would you recommend?

Of course, if Coherence is only $1K per CPU, maybe that's the obvious choice? Unfortunately, I couldn't find their pricing using Google.

Message Driven POJOs by Mark Fisher

Last night, I attended the New England Java Users Group to see Mark Fisher talk about Message Driven POJOs. This was the first JUG I've been to (outside of Denver's) where I was an attendee instead of a speaker. It was interesting to see how they do things. They have one main speaker who speaks for two hours. After the first hour, they have a break, offer pizza and do a raffle. Then the speaker continues. They require you to "register" at least 48 hours before the meeting starts. This is because the meeting is held at Sun and they (apparently) need to do it for security purposes. They actually checked my ID and made sure I was registered at the door. After I passed their verification test, I received a name tag. While I like Denver's Basic Concepts followed by Main Speaker setup, I liked that this JUG meeting was over at 8:15. Below are my notes from the event.

Topics in this Session: Overview of JMS, Spring's JmsTemplate and implementing a Message Driven POJO.

Goals of JMS: provide a vendor-neutral abstraction for accessing Message Oriented Middleware from Java. It provides for enterprise messaging systems what JDBC provides for relational databases. See Wikipedia's definition of JMS for more information.

The API is what insulates your code from the JMS implementation you're using. The JMS implementation will use a message broker to communicate between servers. Tonight, Mark will be using ActiveMQ in his examples. He plans on doing most of his presentation in Eclipse because he's been spending 14 hours per day in PowerPoint revamping Interface21's training courses (I don't miss that at all).

The JMS Message is the central object to JMS. There's various types of messages, TextMessage, ObjectMessage (I missed the rest). Two types of JMS Destinations are available: Queues and Topics. A JMS Session is used to create messages as well as producers and consumers. Examples calls:

• session.createTextMessage(String)
• session.createConsumer(dest)
• session.createProducter(dest)

A JMS Connection is obtained from a JMS ConnectionFactory. The ConnectionFactory is typically accessed by a JNDI Lookup.

JMS is most commonly used for internal application communications (not public facing).

Templates are common in the Spring Framework and are used to simply API usage. Their main goal is to reduce boilerplate code (resource management, try/catch, etc.). Examples include JdbcTemplate, JpaTemplate, JndiTemplate, TransactionTemplate and (you guessed it) JmsTemplate. Spring also translates exceptions to consistent runtime hierarchies. In addition to one-line methods, Spring's templates supply callbacks that alleviate try/catch blocks - but give you full power of the API.

JmsTemplate has a couple of capabilities: it handles acquisition and release of resources and translates checked JMS Exceptions to a parallel hierarchy of RuntimeExceptions. It's also capable of converting a payload to the corresponding JMS message type with a MessageConverter strategy. Lastly, it provides convenience methods to allow sending asynchronous messages.

Now Mark is showing us how you configure a ConnectionFactory, Queue and JmsTemplate in a Spring context file. The first couple of beans only take 3 lines of code to configure, the 3rd one takes 4 because it has a dependency on the first two. Pretty easy configuration if you ask me. After composing bean definitions, Mark created a JUnit test and called jmsTemplate.convertAndSend() to send a message in 2 lines of Java code.

To receive messages with JmsTemplate (pre Spring 2.0), you could use synchronous receive calls: receive(), receive(Destination) and receive(String). There are also receiveAndConvert() methods.

In Spring 2.0, they added the ability to do MessageListener containers to enable asynchronous reception in a non application server environment. Implementations include SimpleMessageListenerContainer, DefaultMessageListenerContainer and ServerSessionMessageListenerContainer. The DMLC adds transactional behavior, and the SSMLC hooks into the server's SPI. Spring's MessageListenerAdapter enables the delegation to a POJO for handling the payload.

For the next 1/2 hour or so, Mark wrote a bunch of Java and XML to create a simple Trader application. About the same time, I got managed to get an internet connection from somewhere and started browsing the net and answering e-mail. Every once in a while I looked up to see Mark's code - it all looks very simple and straight forward. In Spring 2.1, a <jms:*> namespace will be added to simplify the XML configuration.

For those of you out there using Spring's JMS support - are there any issues you've run into? Are you using it in production? It's always pimped as "awesome", so I'm looking for pain points that folks might encounter when using it.

Related: Spring's JMS Documentation and ActiveMQ's JmsTemplate Gotchas.

[TSE] Keynote: The Bigger Picture with Adrian Colyer

We've seen a lot of things over the last few days, but what about the big picture? It's not just about the Spring Framework anymore, but there's also a lot of sub-projects: SFW, SWF, SWS, S-OSGi. Then there's Enterprise services: clustering, persistence, messaging and scheduling. Industry trends: SOA, Web 2.0/RIA, RAD stacks.

Agenda

• Spring portfolio: unifying themes, fitting the pieces together (by layer) and future direction
• Facing the feature: my boss says I need a SOA, from auto-suggest to RIA and the quest for ever-increasing productivity

[Read More]

[TSE] Using Dynamic Languages with Spring with Rod Johnson and Guillaume LaForge

Spring 2.0 has dynamic language support. To make it work, you do need a Java interface as a contract between callers and dynamic beans. There's no special requirements on the interface. It's a "POJI" and doesn't have to extend or implement anything. For example:

There's 3 ways of configuring Groovy beans:

1. GroovyScriptFactory <bean> element defining source location and properties
2. <lang:groovy> element from a <lang> namespace
3. POBD (Plain old <bean> definition) - this is unique for Groovy since it can be compiled into Java bytecode

[Read More]

Continuum, Luntbuild, Pulse and NetBeans

Last night, I did a bit of playing with technologies new to me. First of all, I got AppFuse 2.0 running on Continuum. This was was easy enough, I just had to add <scm> information to each pom.xml. Thanks to those who recommended this approach. I thought it was a silly solution until I realized "mvn site" produced the wrong information when <scm> wasn't present for sub-modules.

Since I was playing with Continuous Integration tools, I decided to give Cerberus, LuntBuild, and Pulse a spin. My goal was to give each server the old "college try" and see if I could get them running with minimal effort. I don't know where I heard about Pulse, but it was somehow included in my tests.

Cerberus didn't work with my Cygwin/Ruby setup, so I was done with it quickly. LuntBuild worked pretty well, but the interface and configuration seemed kinda clunky. I also found it strange that it uses a 4.x version of Jetty - seems kinda old. I was surprised to see that it uses Tapestry for its web framework. Pulse was the nicest one with a kick-ass (ajaxified) user inferface, powered by Acegi, WebWork and Hibernate (according to its JARs). It was definitely the easiest to setup and use. While Pulse isn't free for commercial use, it is free for open source projects, as well as small teams.

Summary: Continuum, LuntBuild and Pulse seem to be the best tools for building Maven 2 projects. While CruiseControl works, and works well, it does require you to customize XML from the command line, whereas these tools allow you to do everything through a web interface.

Toward the end of the night, I downloaded NetBeans 5.5 and installed its Maven 2 Plugin. I was surprised at how full-featured this plugin is. I was able to build, test and run the AppFuse web modules in the embedded Tomcat without issues. It's definitely a cool plugin. As for NetBeans, it seemed pretty sluggish and I couldn't figure out how to get Ctrl+Shift+R functionality, which is a must for me these days. Also, I couldn't get the JSF support working for the AppFuse JSF Module, seemingly caused by the Maven plugin (project properties only has Maven options). Since NetBeans works so well with Maven 2, and it's much more full-featured than Eclipse, it seems natural to recommend it to AppFuse 2 users. Of course, I like IDEA a lot more, but there's no Maven 2 plugin that I know of.

Comparing Open Source Application Servers

With all the recent hubbub about GlassFish, I decided to do a quick performance test this morning. I downloaded all the most recent versions of the various open source application servers, deployed AppFuse 1.9.3 (Struts version) on them, and ran "ant test-canoo" to see if any of them were faster than the other. This was by no means a scientific, isolated test. It also didn't take into account any performance tuning you should do on these servers, I just used the out-of-the-box settings.

I ran these tests on my MacBook Pro (2.16 GHz Intel Core Duo, 2 GB DDR2 SDRAM) with my JAVA_OPTS set to:

-Xms768M -Xmx768M -XX:MaxPermSize=512m -Djava.awt.headless=true

When typing "java -version" at the command line, I got:

java version "1.5.0_06"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.5.0_06-103)
Java HotSpot(TM) Client VM (build 1.5.0_06-57, mixed mode, sharing)

Servers tested (in no particular order):

• JBoss 4.0.4
• GlassFish B48
• JOnAS 4.7.6
• Resin 3.0.21
• Geronimo 1.1

I'm pleased to note that all servers allowed me to deploy appfuse.war without using a console or command-line tool. They all support dropping the WAR in some sort of auto-deploy directory. Very cool! Secondly, I was able to successfully deploy AppFuse on all of them with no changes to AppFuse nor the server. Quite impressive.

My test consisted of the following:

• Copying appfuse.war into the appropriate directory
• Starting the server
• Running "ant test-canoo" from my $APPFUSE_HOME directory once
• Running "ant test-canoo" 3 times, recording the numbers for each run

Here's what I found:

Since I know you're going to ask about Jetty and Tomcat (the two main servlet-only containers), I ran the numbers on those too. First off, I tried Jetty 6 RC0. No dice - I got the following error when trying to start the server.

java.lang.IllegalStateException: Cannot initialize context because there is already a root application 
context present - check whether you have multiple ContextLoader* definitions in your web.xml!
        at org.springframework.web.context.ContextLoader.initWebApplicationContext(ContextLoader.java:173)

Since AppFuse deploys on all the above app servers, as well as Jetty 5.1.x, I'll chalk this up to a bug in Jetty 6. I used Jetty 5.1.11 for this test because I already had it installed on my machine.

I don't know that these numbers mean anything, but it was a fun experiment. For those of you who think these numbers might mean something, here's the rankings:

1. Tomcat 5.5.17
2. Resin 3.0.21
3. JBoss 4.0.4
4. Jetty 5.1.11/GlassFish B48 (tie)
5. Geronimo 1.1
6. JOnAS 4.7.6

Of course, the better test would be hammering each server with 1000 concurrent users (or a number higher than that) and comparing how each server holds up.

RE: What Web Application framework should you use?

Tim O'Brien has an interesting post titled What Web Application framework should you use?. The first thing I noticed about this post is the permalink. It looks like he started with "Isn't Rails supposed to change...", which makes me wonder what the rest of the title was. In this post, he rags on Java Web Frameworks and the lack of a clear path for choosing one. He ends up predicting that many will stick with Struts 1.x (poor bastards) and those that aren't tied to Java should move to Rails. I don't have a problem with folks moving to Rails, but I would like to comment on the Java Web Framework space and Tim's comments.

He says:

Prediction: The confusion over what is happening over at Struts is going to discourage people from continuing to use it. The Struts team did the right thing in recognizing that Struts 1.x was a dead-end, but that project needs a single public message. Is it Struts Action or is it Struts Faces? Or is it two frameworks capitalizing on the Struts brand name?

I think what is going on in the Struts project is definitely two frameworks capitalizing on a brand name. That was a concious choice on the project's part when they chose to start creating sub-projects. The interesting thing about Struts Shale is it's largely a prototype for JSF 2.0. Furthermore, it was rejected by many Struts developers as becoming Struts 2.0. Why? Because JSF sucks. Especially when used with JSP - which is what most folks are doing.

JSF continues to be the most over-hyped under-used framework in Javaland. If you read the blogs of first-time users, you'll find many complaints and issues on how things work. Granted, most of these problems are with JSP and the implementation, but still. If I were in charge of JSF, I'd dump JSP altogether, bundle Facelets with it and allow more flexible page navigation (including controller-to-page). Don't get me wrong, I like the ideas behind JSF, I just don't like the implementation (or the fact I have to wait years for things to be fixed in the spec).

That being said, I've yet to meet an unhappy WebWork fan. If you find someone that still likes Struts, ask them if they've used WebWork. Chances are they'll say no. As far as Tapestry is concerned, the learning curve is too high. It's been rejected time and time again by my clients because of the learning curve. Are they going to fix this? Yep, they're going to re-write the whole damn thing - again! Every major point release of Tapestry throws backwards-compatibility out the window. Furthermore, I've heard once you get over the learning curve, it's a joy to work with. I've also met people at conferences that've used it over a year and say they're still struggling with its concepts.

Spring MVC - I wish I had bad things to say about it, but I don't. It (obviously) has the best Spring integration, but I've found WebWork much more pleasurable to work with. Sure, Spring has a ThrowawayController, but with a name like that, you can tell it's a second-class citizen.

Inspired by Tim's post, here's my prediction:

Struts Action 2 will be the best choice for developing Java-based web frameworks. Not only does it support JSF, but it's easy to learn, test and use. Furthermore, it seems to be the most often used framework in major software products and web sites.

How's that for a clear message? Struts Action 2 is the shiznit, now let's get back to developing applications.

Disclaimer: This is my opinion with a lot of stuff thrown in to get folks riled up. I've never put a JSF, Tapestry or Spring MVC application into production (except for AppFuse of course), so most of my opinions are likely without foundation. In wonder how many applications Mr. O'Brien has put into production with these frameworks?

How do you determine a good MaxPermSize?

I know I'll probably get beat up for not knowing my JVM Turning parameters. I admit that I should know them better than I do. Hopefully this post will help us all understand them a bit better.

Ever since I upgraded appfuse.org to AppFuse 1.9.1, it's been experiencing OOM issues. They've been so bad that the site is lucky if it stays up for more than an hour. I've done a fair amount of performance testing on a single AppFuse application (and gotten very good numbers), so I was pretty puzzled by the whole situation.

To reproduce the problem, I downloaded all 5 demos to my machine and began profiling with JProfiler. Nothing stood out, but I was able to reproduce the problem by clicking through all the different applications. While testing, I had my JAVA_OPTS set to -Xms256M -Xmx384M.

After staring at JProfiler for hours, I gave up and sent my findings to the AppFuse mailing list. After going back and forth with several ideas, Sanjiv came up with the winner.

Did you try increasing the max perm size (-XX:MaxPermSize=256m)? Max Perm size is running out of memory and not necessarily the main memory. Class metadata stuff is placed in the perm memory (google for more details) and since we're using Spring, Hibernate and Tapestry which all use a lot of reflection, proxying etc, it's not surprising that max perm size is running out of memory.

Based on his advice, I added -XX:MaxPermSize=256m to my JAVA_OPTS, fired up JProfiler/Tomcat and began hammering my local instance with WAPT. 15 minutes later, with 20 simultaneous users, the heap and memory were humming along nicely with no issues. I made the change on appfuse.org and it's been up every since.

This experience has motivated me to start adding "-XX:MaxPermSize=256m" to all my JAVA_OPTS. Is this a good idea? If so, is 256m a good value to use? If not, what's the best way to determine (or guess) the proper value for this setting?