1. Efficient container images

It is easily possible to package a Spring Boot fat jar as a docker image. However, there are various downsides to copying and running the fat jar as is in the docker image. There’s always a certain amount of overhead when running a fat jar without unpacking it, and in a containerized environment this can be noticeable. The other issue is that putting your application’s code and all its dependencies in one layer in the Docker image is sub-optimal. Since you probably recompile your code more often than you upgrade the version of Spring Boot you use, it’s often better to separate things a bit more. If you put jar files in the layer before your application classes, Docker often only needs to change the very bottom layer and can pick others up from its cache.

1.1. Unpacking the fat jar

If you are running your application from a container, you can use an executable jar, but it is also often an advantage to explode it and run it in a different way. Certain PaaS implementations may also choose to unpack archives before they run. For example, Cloud Foundry operates this way. One way to run an unpacked archive is by starting the appropriate launcher, as follows:

$ jar -xf myapp.jar
$ java org.springframework.boot.loader.JarLauncher

This is actually slightly faster on startup (depending on the size of the jar) than running from an unexploded archive. At runtime you should not expect any differences.

Once you have unpacked the jar file, you can also get an extra boost to startup time by running the app with its "natural" main method instead of the JarLauncher. For example:

$ jar -xf myapp.jar
$ java -cp BOOT-INF/classes:BOOT-INF/lib/* com.example.MyApplication
Using the JarLauncher over the application’s main method has the added benefit of a predictable classpath order. The jar contains a classpath.idx file which is used by the JarLauncher when constructing the classpath.

1.2. Layering Docker Images

To make it easier to create optimized Docker images, Spring Boot supports adding a layer index file to the jar. It provides a list of layers and the parts of the jar that should be contained within them. The list of layers in the index is ordered based on the order in which the layers should be added to the Docker/OCI image. Out-of-the-box, the following layers are supported:

  • dependencies (for regular released dependencies)

  • spring-boot-loader (for everything under org/springframework/boot/loader)

  • snapshot-dependencies (for snapshot dependencies)

  • application (for application classes and resources)

The following shows an example of a layers.idx file:

- "dependencies":
  - BOOT-INF/lib/library1.jar
  - BOOT-INF/lib/library2.jar
- "spring-boot-loader":
  - org/springframework/boot/loader/JarLauncher.class
  - org/springframework/boot/loader/jar/JarEntry.class
- "snapshot-dependencies":
  - BOOT-INF/lib/library3-SNAPSHOT.jar
- "application":
  - META-INF/MANIFEST.MF
  - BOOT-INF/classes/a/b/C.class

This layering is designed to separate code based on how likely it is to change between application builds. Library code is less likely to change between builds, so it is placed in its own layers to allow tooling to re-use the layers from cache. Application code is more likely to change between builds so it is isolated in a separate layer.

Spring Boot also supports layering for war files with the help of a layers.idx.

For Maven, see the packaging layered jar or war section for more details on adding a layer index to the archive. For Gradle, see the packaging layered jar or war section of the Gradle plugin documentation.

2. Dockerfiles

While it is possible to convert a Spring Boot fat jar into a docker image with just a few lines in the Dockerfile, we will use the layering feature to create an optimized docker image. When you create a jar containing the layers index file, the spring-boot-jarmode-layertools jar will be added as a dependency to your jar. With this jar on the classpath, you can launch your application in a special mode which allows the bootstrap code to run something entirely different from your application, for example, something that extracts the layers.

The layertools mode can not be used with a fully executable Spring Boot archive that includes a launch script. Disable launch script configuration when building a jar file that is intended to be used with layertools.

Here’s how you can launch your jar with a layertools jar mode:

$ java -Djarmode=layertools -jar my-app.jar

This will provide the following output:

Usage:
  java -Djarmode=layertools -jar my-app.jar

Available commands:
  list     List layers from the jar that can be extracted
  extract  Extracts layers from the jar for image creation
  help     Help about any command

The extract command can be used to easily split the application into layers to be added to the dockerfile. Here is an example of a Dockerfile using jarmode.

FROM adoptopenjdk:11-jre-hotspot as builder
WORKDIR application
ARG JAR_FILE=target/*.jar
COPY ${JAR_FILE} application.jar
RUN java -Djarmode=layertools -jar application.jar extract

FROM adoptopenjdk:11-jre-hotspot
WORKDIR application
COPY --from=builder application/dependencies/ ./
COPY --from=builder application/spring-boot-loader/ ./
COPY --from=builder application/snapshot-dependencies/ ./
COPY --from=builder application/application/ ./
ENTRYPOINT ["java", "org.springframework.boot.loader.JarLauncher"]

Assuming the above Dockerfile is in the current directory, your docker image can be built with docker build ., or optionally specifying the path to your application jar, as shown in the following example:

$ docker build --build-arg JAR_FILE=path/to/myapp.jar .

This is a multi-stage dockerfile. The builder stage extracts the directories that are needed later. Each of the COPY commands relates to the layers extracted by the jarmode.

Of course, a Dockerfile can be written without using the jarmode. You can use some combination of unzip and mv to move things to the right layer but jarmode simplifies that.

3. Cloud Native Buildpacks

Dockerfiles are just one way to build docker images. Another way to build docker images is directly from your Maven or Gradle plugin, using buildpacks. If you’ve ever used an application platform such as Cloud Foundry or Heroku then you’ve probably used a buildpack. Buildpacks are the part of the platform that takes your application and converts it into something that the platform can actually run. For example, Cloud Foundry’s Java buildpack will notice that you’re pushing a .jar file and automatically add a relevant JRE.

With Cloud Native Buildpacks, you can create Docker compatible images that you can run anywhere. Spring Boot includes buildpack support directly for both Maven and Gradle. This means you can just type a single command and quickly get a sensible image into your locally running Docker daemon.

See the individual plugin documentation on how to use buildpacks with Maven and Gradle.

The Paketo Spring Boot buildpack has also been updated to support the layers.idx file so any customization that is applied to it will be reflected in the image created by the buildpack.
In order to achieve reproducible builds and container image caching, Buildpacks can manipulate the application resources metadata (such as the file "last modified" information). You should ensure that your application does not rely on that metadata at runtime. Spring Boot can use that information when serving static resources, but this can be disabled with spring.web.resources.cache.use-last-modified

4. What to Read Next

Once you’ve learned how to build efficient container images, you can read about deploying applications to a cloud platform, such as Kubernetes.