Dockerfiles

虽然只需在 Dockerfile 中写几行就能将 Spring Boot uber jar 转换为 Docker 镜像，但使用 分层特性 能获得更优的镜像。 当你创建包含 layers 索引文件的 jar 时，spring-boot-jarmode-tools jar 会作为依赖添加到你的 jar 中。 有了这个 jar 在类路径上，你可以以特殊模式启动应用，让引导代码运行与应用本身完全不同的内容，例如提取分层。

带有启动脚本的 完全可执行 Spring Boot 存档 不支持 tools 模式。 构建用于 extract tools 模式命令的 jar 文件时，请禁用启动脚本配置。

以下是以 tools jar 模式启动 jar 的方法：

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

这将输出如下内容：

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

Available commands:
  extract      Extract the contents from the jar
  list-layers  List layers from the jar that can be extracted
  help         Help about any command

extract 命令可轻松将应用拆分为可添加到 Dockerfile 的分层。 以下是使用 jarmode 的 Dockerfile 示例。

# Perform the extraction in a separate builder container
FROM bellsoft/liberica-openjre-debian:24-cds AS builder
WORKDIR /builder
# This points to the built jar file in the target folder
# Adjust this to 'build/libs/*.jar' if you're using Gradle
ARG JAR_FILE=target/*.jar
# Copy the jar file to the working directory and rename it to application.jar
COPY ${JAR_FILE} application.jar
# Extract the jar file using an efficient layout
RUN java -Djarmode=tools -jar application.jar extract --layers --destination extracted

# This is the runtime container
FROM bellsoft/liberica-openjre-debian:24-cds
WORKDIR /application
# Copy the extracted jar contents from the builder container into the working directory in the runtime container
# Every copy step creates a new docker layer
# This allows docker to only pull the changes it really needs
COPY --from=builder /builder/extracted/dependencies/ ./
COPY --from=builder /builder/extracted/spring-boot-loader/ ./
COPY --from=builder /builder/extracted/snapshot-dependencies/ ./
COPY --from=builder /builder/extracted/application/ ./
# 启动应用 jar —— 这不是 builder 使用的 uber jar
# 此 jar 仅包含应用代码及对已提取 jar 文件的引用
# 这种布局启动高效，且对 CDS/AOT 缓存友好
ENTRYPOINT ["java", "-jar", "application.jar"]

假设上述 Dockerfile 位于当前目录，可用 docker build . 构建镜像，或如下面例子指定应用 jar 路径：

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

这是一个多阶段 Dockerfile。 builder 阶段会提取后续需要的目录。 每个 COPY 命令都对应 jarmode 提取的分层。

当然，也可以不用 jarmode 编写 Dockerfile。 你可以用 unzip 和 mv 等组合将内容移到正确分层，但 jarmode 简化了这一过程。 此外，jarmode 创建的布局天生支持 CDS 和 AOT 缓存。

CDS

如果你还想启用 CDS，可使用如下 Dockerfile：

# Perform the extraction in a separate builder container
FROM bellsoft/liberica-openjre-debian:24-cds AS builder
WORKDIR /builder
# This points to the built jar file in the target folder
# Adjust this to 'build/libs/*.jar' if you're using Gradle
ARG JAR_FILE=target/*.jar
# Copy the jar file to the working directory and rename it to application.jar
COPY ${JAR_FILE} application.jar
# Extract the jar file using an efficient layout
RUN java -Djarmode=tools -jar application.jar extract --layers --destination extracted

# This is the runtime container
FROM bellsoft/liberica-openjre-debian:24-cds
WORKDIR /application
# Copy the extracted jar contents from the builder container into the working directory in the runtime container
# Every copy step creates a new docker layer
# This allows docker to only pull the changes it really needs
COPY --from=builder /builder/extracted/dependencies/ ./
COPY --from=builder /builder/extracted/spring-boot-loader/ ./
COPY --from=builder /builder/extracted/snapshot-dependencies/ ./
COPY --from=builder /builder/extracted/application/ ./
# 执行 CDS 训练运行
RUN java -XX:ArchiveClassesAtExit=application.jsa -Dspring.context.exit=onRefresh -jar application.jar
# 启动已启用 CDS 的应用 jar —— 这不是 builder 使用的 uber jar
# 此 jar 仅包含应用代码及对已提取 jar 文件的引用
# 这种布局启动高效且对 CDS 友好
ENTRYPOINT ["java", "-XX:SharedArchiveFile=application.jsa", "-jar", "application.jar"]

这与上面的 Dockerfile 基本一致。 最后一步通过训练运行创建 CDS 存档，并将 CDS 参数传递给 java -jar。

AOT cache

如果你还想启用 AOT cache，可使用如下 Dockerfile：

# Perform the extraction in a separate builder container
FROM bellsoft/liberica-openjre-debian:24-cds AS builder
WORKDIR /builder
# This points to the built jar file in the target folder
# Adjust this to 'build/libs/*.jar' if you're using Gradle
ARG JAR_FILE=target/*.jar
# Copy the jar file to the working directory and rename it to application.jar
COPY ${JAR_FILE} application.jar
# Extract the jar file using an efficient layout
RUN java -Djarmode=tools -jar application.jar extract --layers --destination extracted

# This is the runtime container
FROM bellsoft/liberica-openjre-debian:24-cds
WORKDIR /application
# Copy the extracted jar contents from the builder container into the working directory in the runtime container
# Every copy step creates a new docker layer
# This allows docker to only pull the changes it really needs
COPY --from=builder /builder/extracted/dependencies/ ./
COPY --from=builder /builder/extracted/spring-boot-loader/ ./
COPY --from=builder /builder/extracted/snapshot-dependencies/ ./
COPY --from=builder /builder/extracted/application/ ./
# 执行 AOT 缓存训练运行
RUN java -XX:AOTMode=record -XX:AOTConfiguration=app.aotconf -Dspring.context.exit=onRefresh -jar application.jar
RUN java -XX:AOTMode=create -XX:AOTConfiguration=app.aotconf -XX:AOTCache=app.aot -jar application.jar && rm app.aotconf
# 启动已启用 AOT 缓存的应用 jar —— 这不是 builder 使用的 uber jar
# 此 jar 仅包含应用代码及对已提取 jar 文件的引用
# 这种布局启动高效且对 AOT 缓存友好
ENTRYPOINT ["java", "-XX:AOTCache=app.aot", "-jar", "application.jar"]

这与上面的 Dockerfile 基本一致。 最后一步通过训练运行创建 AOT 缓存文件，并将 AOT 缓存参数传递给 java -jar。