skip to Main Content

I have a docker image being built as a multi stage build

from ubuntu:latest as base

# do some stuff

from base as llvm 

# build llvm (takes ~1.5 hours)

from llvm as independent_big_project

# build independent big project (takes ~1 hours)

I build llvm which takes a long time, and then another large project which is not dependent on llvm. I may have to tweak the settings for both llvm and my other project. If I put llvm first I will have to rebuild the other project as it is based on the previous stage, if I build the project first then I will have to rebuild llvm for the same reason.

Is there a way to split this into independent sections and combine both projects into one image at the end?

I could use multiple docker images and copy the completed builds across, but I think this would need to be managed manually?

Is there a way to automate this process where I can just run docker build on a top level image, and the dependencies are built if required, but separately?

2

Answers


  1. Consider copy --from, along:

    from ubuntu:latest as base
    
    from base as llvm 
    
    from base as independent_big_project
    
    from ubuntu:latest as the_result
    copy --from independent_big_project /the/result /
    copy --from llvm /the/result /
    
    Login or Signup to reply.
  2. You need to COPY --from files from one build stage to another. If you’re using the current BuildKit Docker image builder, it will try to build all of the build stages in parallel, and only block at the point where it wants to execute that COPY.

    For C and C++ programs, both the Autoconf and CMake tool sets generally support building a program to install into an arbitrary filesystem prefix. That will give you a self-contained directory path that you can COPY into later image. LLVM uses CMake and you should be able to do something like

    FROM ubuntu:24.04 AS base
    ...
    
    FROM base AS llvm
    RUN apt-get update 
     && DEBIAN_FRONTEND=noninteractive
        apt-get install --no-install-recommends --assume-yes 
          build-essential 
          cmake 
          git-core 
          libz-dev 
          ninja-build
    
    RUN git clone --depth 1 https://github.com/llvm/llvm-project.git
    WORKDIR /llvm-project
    RUN cmake -S llvm -B build -G Ninja 
        -DCMAKE_INSTALL_PREFIX=/opt/llvm  # <--
    RUN cmake --build build
    
    FROM base AS independent_big_project
    ...
    
    FROM base AS final
    COPY --from=llvm /opt/llvm /opt/llvm
    

    Calling out the three important things from this fragment:

    1. We build LLVM with -DCMAKE_INSTALL_PREFIX=/opt/llvm so that it installs into its own filesystem tree.
    2. The second build stage, with the independent build, is FROM base; it does not depend on the LLVM build.
    3. We COPY --from=llvm /opt/llvm /opt/llvm in the final build stage. It’s often important to use the same filesystem path when copying install trees like this.

    You’ll probably need to similarly COPY --from=independent_big_project. If LLVM is installing C shared libraries, you may need to set ENV LD_LIBRARY_PATH=/opt/llvm/lib or to append to /etc/ld.so.conf in the final build image so those can be used at runtime.

    Login or Signup to reply.
Please signup or login to give your own answer.
Back To Top
Search