GraalVM with JPA-based applications (Part 1)

First of all, Happy New Year 2019 to each and every single one of you, dear readers!!!

I wanted to start the first technical blog post of this new year with one particular piece of JVM-related project I discovered last year and am really enthusiastic about: GraalVM, from Oracle.

In a nutshell, GraalVM is a (still experimental) very powerful polyglot virtual machine aiming at supporting and mixing different programming languages in a same application. It currently supports languages such as:

• JavaScript. Actually, JavaScript in the JVM has been supported for a long time via lightweight runtimes, first with the Rhino engine (released in JDK 6) and later with the Nashorn engine (released in JDK 8 but deprecated in JDK 11). GraalVM's support for JavaScript in the JVM is a much more complete and faster implementation of the latest JavaScript standards, and also provides a full support for the Node.js server framework.
• Ruby
• Python
• R
• JVM-based languages, like Java, Kotlin, Groovy, Scala, Clojure
• LLVM-based languages, such as C and C++

You see how awesome this can be?! Using GraalVM, it should be possible for example to call Ruby code from a C++ program, and share cross-language data structures.

For JVM-based applications, GraalVM also provides the following capabilities:

• a brand-new efficient Just-In-Time (JIT) compiler. As a reminder, Java code is compiled to bytecode (the famous .class files), which can be seen as an idealized platform-agnostic code, which is then interpreted by the JRE, and later compiled on the fly into machine code as the application runs; hence the terms "just in time". Unlike the default HotSpot JIT compiler, which is written in C++ (but criticized as hard to maintain and extend), the Graal JIT compiler is written in Java, as a concrete implementation of a new interface in the JVM called the JVMCI - the JVM compiler interface - introduced in JDK 9.
• an Ahead-Of-Time (AOT) compiler to improve startup times. Unlike a JIT compiler, an AOT compiler analyzes the code up-front (along with all its dependencies) and produces a standalone platform-specific binary program. To emulate the capabilities of the JVM (such as garbage collection), the Graal AOT compiler produces Native Images that leverage Substrate VM, a Java-based framework that is also shipped with GraalVM.

The goal of this blog post is to go beyond the basic HelloWorld program, and see how we can leverage GraalVM against a sample real-world JPA-based application, able to interact with a database. We will walk through what can be done along with the potential limitations.

As always, the code for this blog post is available at Github://rm3l/jpa-graalvm.

We will be using DataNucleus as the JPA implementation, though the example project contains a set of Maven Profiles allowing to use other JPA implementations, such as EclipseLink (the reference one) or Hibernate.

Installing GraalVM

I highly recommend SDKMAN! as a tool to manage multiple versions of JVM-related Development Kits. Please head to the official instructions to see how to install SDKMAN!

Once done, you can list all JDKs available with the following command:

❯ sdk list java

================================================================
Available Java Versions
================================================================
     13.ea.02-open       8.0.202.j9-adpt                                        
     12.ea.26-open       8.0.202.hs-adpt                                        
     11.0.2-zulu         8.0.201-zulu                                           
     11.0.2-open         8.0.201-oracle                                         
     11.0.2.j9-adpt      8.0.192-zulufx                                         
     11.0.2.hs-adpt      7.0.201-zulu                                           
   + 11.0.1-open         6.0.119-zulu                                           
     11.0.1-zulufx     * 1.0.0-rc-12-grl                                        
     10.0.2-zulu         1.0.0-rc-11-grl                                        
     10.0.2-open         1.0.0-rc-10-grl                                        
     9.0.7-zulu          1.0.0-rc-9-grl                                         
     9.0.4-open          1.0.0-rc-8-grl                                         
   + 9.0.1-oracle                                                               
 > + 8u151-oracle                                                               
     8.0.202-amzn                                                               

================================================================
+ - local version
* - installed
> - currently in use
================================================================

Installing GraalVM is then just as easy as picking the right JDK, e.g.:

❯ sdk install java 1.0.0-rc-12-grl

You may choose to make it your default JDK system-wide, or use it temporarily just for testing:

❯ sdk use java 1.0.0-rc-12-grl

Let's check the JDK we have just installed and confirm we are using GraalVM:

❯ java -version
openjdk version "1.8.0_192"
OpenJDK Runtime Environment (build 1.8.0_192-20181024121959.buildslave.jdk8u-src-tar--b12)
GraalVM 1.0.0-rc12 (build 25.192-b12-jvmci-0.54, mixed mode)

Using the Graal Compiler

Our sample project makes use of the Java Persistence API - JPA - to persist and update few people data in an in-memory database. It then pretty-prints such records in JSON on the standard output, using the Ruby programming language, in order to showcase how polyglot GraalVM is.

This is done in the following code snippet:

import org.graalvm.polyglot.*;

//...

private static void prettyPrintWithRuby(final Collection<Person> people) {
        try (final Context context = Context.create("ruby")) {
            final Value json = context.eval("ruby",
                    "require 'json'; " +
                    "JSON.pretty_generate(" +
                    " JSON.parse('" +
                    "   [" + people.stream()
                    .map(Person::toJsonString)
                    .collect(Collectors.joining(", ")) + "]'))");
            System.out.println(json.asString());
        }
    }

What this code does is pretty straightforward:

1. creating an execution context for the "ruby" language. For this, we may need to install the engine for such language using the gu (Graal Updater) command. More on that below.
2. evaluating the provided Ruby script and wrapping the result in a language-agnostic org.graalvm.polyglot.Value object, which is then manipulated back in Java and printed to screen

We can also do the same using the JavaScript language engine, by evaluating multiple native JS functions:

import org.graalvm.polyglot.*;

//...

private static void prettyPrintWithJS(final Collection<Person> people) {
        try (final Context context = Context.create("js")) {
            final Value parse = context.eval("js", "JSON.parse");
            final Value stringify = context.eval("js", "JSON.stringify");
            final Value parseResult = parse.execute("[" + people.stream()
                    .map(Person::toJsonString)
                    .collect(Collectors.joining(", ")) + "]");
            System.out.println(stringify.execute(parseResult, null, 2).asString());
        }
    }

First, let's build the project:

❯ mvn package -P datanucleus

Running with the optimized Graal JIT compiler is then straightforward:

❯ java \
  -Dgraal.ShowConfiguration=info \
  -XX:+UseJVMCICompiler \
  -XX:+EagerJVMCI \
  -cp "target/jpa-graalvm-5.2.jar:target/lib/*" \
  -Dprofile=datanucleus \
  mydomain.MySampleApplication

If we run the command above, we may get an error indicating that the 'ruby' language is not installed:

❯ java -Dgraal.ShowConfiguration=info -XX:+UseJVMCICompiler -XX:+EagerJVMCI -cp "target/jpa-graalvm-5.2.jar:target/lib/*" -Dprofile=datanucleus mydomain.MySampleApplication      
[Use -Dgraal.LogFile=<path> to redirect Graal log output to a file.]                                                                                                              
Using Graal compiler configuration 'community' provided by org.graalvm.compiler.hotspot.CommunityCompilerConfigurationFactory loaded from jar:file:/home/rm3l/.sdkman/candidates/ja
va/1.0.0-rc-12-grl/jre/lib/jvmci/graal.jar!/org/graalvm/compiler/hotspot/CommunityCompilerConfigurationFactory.class                                                              
Exception in thread "main" java.lang.IllegalStateException: Failed test : A language with id 'ruby' is not installed. Installed languages are: [js, llvm].                        
        at mydomain.MySampleApplication.executeInTransaction(MySampleApplication.java:132)                                                                                        
        at mydomain.MySampleApplication.main(MySampleApplication.java:76)

In this case, we need to install the Ruby language engine, using Graal's gu (Graal Updater) command:

❯ gu install ruby

Re-runing the java command above will then pretty-print our database records using Ruby code in our Java code:

❯ java -Dgraal.ShowConfiguration=info -XX:+UseJVMCICompiler -XX:+EagerJVMCI -cp "target/jpa-graalvm-5.2.jar:target/lib/*" -Dprofile=datanucleus mydomain.MySampleApplication


[Use -Dgraal.LogFile=<path> to redirect Graal log output to a file.]
Using Graal compiler configuration 'community' provided by org.graalvm.compiler.hotspot.CommunityCompilerConfigurationFactory loaded from jar:file:/home/rm3l/.sdkman/candidates/java/1.0.0-rc-12-grl/jre/lib/jvmci/graal.jar!/org/graalvm/compiler/hotspot/CommunityCompilerConfigurationFactory.class                                                              
Using Graal compiler configuration 'community' provided by org.graalvm.compiler.hotspot.CommunityCompilerConfigurationFactory loaded from jar:file:/home/rm3l/.sdkman/candidates/java/1.0.0-rc-12-grl/jre/lib/jvmci/graal.jar!/org/graalvm/compiler/hotspot/CommunityCompilerConfigurationFactory.class                                                              
[
  {
    "uuid": "c5e02d7f-348f-4e0f-85d4-a17c8613bc62",
    "uniqueName": "person 2 (renamed)",
    "lastAddress": {
      "uuid": "61121487-1671-4c08-a556-ae29bf81d5e7",
      "zipCode": 22,
      "number": 2,
      "street": "Avenue FHB",
      "city": "Abidjan",
      "countryCode": "CI"
    }
  },
  {
    "uuid": "a1727ea1-c5c2-41d7-b44b-13ed8c1543ca",
    "uniqueName": "person 1 (renamed)",
    "lastAddress": {
      "uuid": "eefffc6a-9866-4421-8ccf-8206faffb215",
      "zipCode": 21,
      "number": 1,
      "street": "Avenue FHB",
      "city": "Lyon",
      "countryCode": "FR"
    }
  }
]

That's it for the moment, now that we have seen how we can easily leverage GraalVM to write polyglot code that extends JPA-based applications. In the second part of this blog post, we'll see how we can further improve startup time by building a Native Image of our JPA-based application, if at all possible.

Stay tuned for the upcoming blog post, and as always, your feedback is more than welcome.