Saturday, June 16, 2018

The next generation of Java application deployments

A lot is going on in the Java world. So let's first have a look at where we currently are.

Oracle discontinues Java Web Start with Java 11

After discontinuing the Applet technology, which required browser plug-ins, Oracle now announced that they will also discontinue the Java Web Start technoloy.

Java Web Start, however, was a great way to easily deploy Java applications. It made sure Java application deployments don't fall too far behind the instant deployment feature of web applications - a point which already let many companies to choose web applications over Java based applications.

With Java Web Start gone, it will become harder again to compete with web applications, if we don't find a good alternative (please read on).

Oracle changed to a biannual major Java release cadence

Every 6 month there will be now a new "major" Java version, though the term "major" likely won't be used anymore and differences will be smaller than with the 2-3 year cadence (see Java 8 -> Java 9 vs Java 9 -> Java 10).

With every "major" release the support of the previous release seems to be discontinued.

Every 3 years there will be a LTS (Long Term Support) version, starting with Java 11 targeted for September 2018 (there is a small video and a FAQ about this from Oracle).

As far as I understand (please correct me if I'm wrong) you will need to buy this LTS. If you don't, then it's recommended to upgrade to the next major version or find another distributor which provides updates (more on this see below).

Interestingly and to some surprise, the long awaited Java SE 9 release with the new module system didn't become a LTS version and already has been superseded by Java SE 10.

Oracle also mentions that developers often package the JRE with their applications nowadays. End users should not need to install a JRE anymore.

Oracle and OpenJDK binaries

While Sun started to open source the JDK with the OpenJDK project back in 2006 and Oracle continued this process, there are still some differences between the OpenJDK builds and the Oracle Java builds.

Now Oracle plans to ship OpenJDK binaries for non-LTS releases and with this shift they also plan to finally close those technical differences, which currently still exist.

Once these differences are closed, this will make projects such as AdoptOpenJDK much more appealing, as it seems they will provide free updates for LTS versions in the future.

Modular JRE and JLink

Together with the JPMS (Java Platform Module System), which modularizes the JRE and can be used to modularize applications, Java 9 introduced a new tool called JLink.

JLink allows you to create custom runtime images which only consists of your application modules and those JRE modules which your application requires.

The result is likely a smaller runtime image which uses less resources than using a default JRE.

This should also make Java more appealing to IoT devices.

JavaFX

Since JavaFX 2.2 and Java SE 7 update 6, JavaFX has been included in Oracle's JDK. With JavaFX 8 and Java SE 8 the version has been aligned and JavaFX has become available on the default classpath.

Now Oracle announced (see the link at the top) that with Java SE 11 JavaFX will be removed again from the JDK.

With a modularized JRE it might be more natural to let feature modules have their own cycles.

But since JavaFX also includes native parts, it seems to me WORA (Write Once, Run Anywhere) will become a bit trickier - at least the packaging part of JavaFX applications.

Please read on for a possible solution.

The Java Packager Tool

JavaFX also introduced a new tool to create platform-specific native packages with an embedded JRE. With JavaFX 8 the tool has been added to the JDK as the javapackager tool.

Since Java SE 9 the javapackager tool integrates with the jlink tool.

In Java SE 11 the javapackager tool will be removed again from the JDK along with JavaFX.

There is a JEP request (JDK Enhancement Proposal) for a new jpackager tool, however.

While I think we do need a tool such as javapackager/ jpackager, javapackager and the proposed jpackager also have some major drawbacks:
  • no cross-compilation support: You have to create the native packages on the target platforms, e.g. msi-packages on Windows, deb-packages on Linux. Creating a one-click release pipeline is non-trivial.
  • platform specific configurations: Each native package has slightly different requirements.
  • no automatic security updates: The users have to rely on the application vendors to ship a new application version which contains a JRE with the latest security fixes (though not all security bugs might affect all applications in the same way).
  • expensive managed deployments: In heavily managed enterprises, where the average users don't have local admin rights and can't install applications themselves, the processes of application deployments tend to get very expensive (up to several thousand CHF per deployment). This is another reason why enterprises nowadays tend to either Java Web Start or to web applications.
  • slow managed deployments: In these heavily managed enterprises, deployments also tend to be slow and can take up to several weeks until the application is deployed to the user machine. This also makes it hard to coordinate frontend deployments with backend deployments and is definitely not DevOps-friendly. This is yet another reason why enterprises nowadays tend to either Java Web Start or to web applications.

Drombler JStore


Given the above mentioned issues with tools such as  javapackager/ jpackager, which create application packages with an embedded JRE, I don't think this is the solution for the majority of Java applications.

On the other hand, I too think that we should not require anymore from end-users to pre-install a JRE, which likely doesn't fit for all applications anyway.

It's however unlikely that Oracle will provide another solution than the jpackager.

If we need something different then it will have to be community driven.

So I'm proposing here a new deployment model with the working title (likely to change): Drombler JStore

At the core of this deployment model is a new application store with first class support for Java applications.

In general, most applications should be able to define a platform-independent package which also describes the dependencies to some managed native components such as the JRE.

The tooling will then make sure that the required native components are available on the end-user machine. The tooling will also configure the environment for the process which starts the application.

The currently proposed solution consists of 4 major software parts.
I've already written a PoC (Proof of Concept) for each of those, which I think can serve as a starting point. Please note that everything is still subject to change.
  • Drombler JStore Client Agent:
    • Current goals:
      • headless background service
      • targets: end-user devices (desktops, NAS, IoT devices,...)
      • download and update applications from Drombler JStore
      • download and update applications using a JNLP-like file
      • download and update native components:
        • JREs (supporting multiple parallel major version installations)
        • custom runtimes created by JLink (supporting multiple parallel major version installations)
        • JavaFX native components?
        • Other native components, e.g. SWT?
      • support Security Manager
      • start applications in a configured environment
      • OS integration: 
        • desktop shortcuts
        • URL protocol registration
        • run as native OS service/ deamon
    • PoC:
      • Access a first Drombler JStore REST service
      • Download a JRE from Oracle
    • GitHub: https://github.com/Drombler/drombler-jstore-client-agent
  • Drombler JStore:
    • Current goals:
      • a free service (backend) run by Drombler
      • Provide access to applications / application updates
      • Provide access to native components and updates (JREs etc.)
      • Search applications
      • Rate applications
      • Comment on applications
      • Register vendors
      • Provide vendor contact options
    • PoC:
      • a first deployed REST service is available (likely to change)
    • Swagger: http://drombler-jstore-staging.us-east-1.elasticbeanstalk.com/swagger-ui.html
  • Drombler JStore Client
    • Current goals:
      • JavaFX-based rich client
      • Remote management of multiple Drombler JStore Client Agents
      • Discover applications on Drombler JStore
      • Rate applications on Drombler JStore
      • Comment on applications on Drombler JStore
      • Register vendors
    • PoC:
      • Drombler FX-based rich client
      • Initial multiple Drombler JStore Client Agent management support
      • Plugable feature support
      • Connect to the Drombler JStore REST service
    • GitHub: https://github.com/Drombler/drombler-jstore-client
  • JAP Maven Plugin
    • Current goals:
      • Define a platform-independent application packaging format and meta data
      • Provide Maven tooling
    • PoC
      • Creates a ZIP-based package, which contains the application in a nested ZIP-file. 
    • GitHub: https://github.com/Drombler/jap-maven-plugin 
    • in the future there could be similar solutions for other build tools such as Gradle

Call to the Community

As mentioned above, if we need a different deployment model then it will have to be community driven.

What do think about this proposed solution?
Do you have a need for an alternative deployment model?
Do you think we're running in the right direction with this proposed solution or did we miss something?

Contributions and Donations 

Even for a MVP (Minimal Viable Product) there is quite some work to do.
Contributions are highly welcome! The current code base is not very complex. You should be able to get started quickly. Please don't hesitate to ask me, if you have any questions.

We can use your inputs especially in the following areas:
  • What should the platform-independent application packaging format and meta data look like to support your applications? What information is needed by the agent? What information is only needed by the discovery feature of the rich client?
  • Help us to make sure the agents have the necessary native OS integration to support your applications.
  • Help us to create a great user experience with the rich client (some UX and JavaFX skills required)
  • Help translate the texts of the rich client to different languages (i18n/ l10n)
  • Help us to get the basic features of the agent right
  • File enhancement requests for the various components
  • Join the discussions in the issues or on Gitter.
  • Help with the documentation (GitHub Pages, README files)

Since the proposed solution requires a backend, this will also require some resources to run the service. Any donations are welcome!

Further Readings

Here are some further blog posts about similar topics:

Conclusion

A lot is going on in the Java world. But these changes also provide opportunities for new solutions.

Join and help to shape the next generation of Java application deployments!

3 comments:

  1. What are the advantages of using a separate store of, for example, Maven repositories?

    ReplyDelete
    Replies
    1. The main purpose of the JStore backend will be to manage native components such as JREs. It will allow to install several versions of the JRE in parallel, provide JRE security updates without any actions from the application vendors, propose alternative JREs if a JRE reaches end of life and so on.
      The application could either be deployed using a JNLP like approach or by registering the application in the JStore backend, which would allow applications to be discoverd by end-users and can provide additional services such end-user comments and rating even accros os platforms.

      While the implementation of the JStore backend will likely use a private Maven repository, at least Maven Central is more for (Open Source) libraries rather than applications which can vary in size from a few mega bytes to some tens of mega bytes and even up to a few gigabytes.

      But maybe I didn't fully get your question?

      Delete
  2. I stumbled upon this while trying to package my JavaFX application build with Java11. JMod and JLink maven plugins seem to be full of errors and still in Beta since 2017. I agree that the community urgently needs a solid packaging solution.
    I would be glad to lend a hand.

    ReplyDelete