dockerize2

About this fork. dockerize2 continues larsks/dockerize — picking up where upstream paused in 2020 to refresh the toolchain (PEP 621 packaging, type hints, uv, Python 3.11+) and add new capabilities: UPX compression, OCI-archive output, SBOM generation, a doctor health check, and a multi-arch container image. Original copyright is preserved (see NOTICE and LICENSE.txt); the project remains GPL-3.0-licensed.

dockerize2 packs up your dynamically linked ELF binaries and all their dependencies and turns them into a minimal FROM scratch Docker image — optionally UPX-compressed, with a generated SBOM, emitted as either a daemon push or an OCI archive.

Some example images built with the original tool are available from:

• https://hub.docker.com/u/dockerizeme/

Installation

dockerize2 is a standard Python package. Until v0.3.0 lands on PyPI you can install it from this repository:

pip install git+https://github.com/schubydoo/dockerize2

The installed console script is still called dockerize so existing scripts continue to work unchanged.

Run from a container

A pre-built multi-arch image is available at ghcr.io/schubydoo/dockerize2. Supported architectures:

• linux/amd64
• linux/arm64
• linux/arm/v7 (32-bit hardware-float ABI — Raspberry Pi 32-bit, etc.)

Image variants

Two flavours are published:

Tags	Contents	Use for
:latest, :X.Y, :X.Y.Z, :sha-<short>	dockerize2 + docker CLI + syft + upx	the full CLI: --runtime docker, --sbom, --compress, --output-oci, staging
:slim, :X.Y-slim, :X.Y.Z-slim, :sha-<short>-slim	dockerize2 only (pure Python)	daemonless --output-oci and -n -o staging in multi-stage builds; smallest footprint

:slim drops the docker CLI, syft, and upx, so --runtime docker, --sbom, and --compress are unavailable there — but the fully daemonless paths still work: --output-oci PATH and the -n -o DIR staging pattern below.

Stage a binary in a multi-stage build (:slim)

Use dockerize2 as a build stage to copy a binary and its shared libraries into a directory, then assemble a FROM scratch final image from it:

FROM ghcr.io/schubydoo/dockerize2:slim AS stage
RUN apt-get update && apt-get install -y --no-install-recommends jq
# -n = no build, -o = output dir: stage jq + its libs into /out
RUN dockerize -n -o /out "$(command -v jq)" \
 && rm -f /out/Dockerfile          # generated build artifact, not part of the image

FROM scratch
COPY --from=stage /out/ /
ENTRYPOINT ["/usr/bin/jq"]

dockerize2 resolves glibc binaries (it runs on Debian), which suits typical dynamically-linked Linux executables; it is not the right tool for musl/Alpine binaries. The rm -f /out/Dockerfile matters because -n -o writes a generated Dockerfile into the output dir that you don't want copied into the final image.

OCI-archive output — produces a portable OCI image-layout tarball instead of loading the image into a local store. The difference from classic mode is the self-contained .oci.tar (plus a matching SBOM) that you load yourself.

--output-oci is fully daemonless: dockerize assembles the single-layer OCI image in pure Python — no Docker socket, no buildx, no podman. Load the resulting archive with skopeo, oras, podman load, or docker load (the last needs the containerd image store enabled).

docker run --rm \
  -v "$PWD":/work \
  -v /usr/sbin/mini_httpd:/usr/sbin/mini_httpd:ro \
  ghcr.io/schubydoo/dockerize2:latest \
    -t httpd \
    --output-oci /work/httpd.oci.tar \
    --sbom /work/httpd.sbom.spdx.json \
    --compress \
    /usr/sbin/mini_httpd

Then on the host:

docker load -i httpd.oci.tar

Classic mode — build straight into the daemon's local image store:

docker run --rm \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v /usr/sbin/mini_httpd:/usr/sbin/mini_httpd:ro \
  ghcr.io/schubydoo/dockerize2:latest \
    -t httpd /usr/sbin/mini_httpd

Run the health check:

docker run --rm ghcr.io/schubydoo/dockerize2:latest doctor

Synopsis

usage: dockerize [-h] [--tag TAG] [--cmd CMD] [--entrypoint ENTRYPOINT]
                 [--no-build] [--output-dir OUTPUT_DIR] [--add-file SRC DST]
                 [--symlinks SYMLINKS] [--user USER] [--group GROUP]
                 [--filetools] [--no-host-lookup] [--allow-sensitive]
                 [--nss-modules NSS_MODULES] [--label KEY=VALUE] [--compress]
                 [--compress-level {normal,best,ultra}] [--compress-libs]
                 [--sbom PATH]
                 [--sbom-format {spdx-json,cyclonedx-json,syft-json}]
                 [--output-oci PATH] [--runtime RUNTIME] [--buildcmd BUILDCMD]
                 [--verbose] [--debug] [--version]
                 ...

positional arguments:
  paths

options:
  -h, --help            show this help message and exit
  --add-file, -a SRC DST
                        Add file <src> to image at <dst>
  --symlinks, -L SYMLINKS
                        One of preserve, copy-unsafe, skip-unsafe, copy-all
  --user, -u USER       Add user to /etc/passwd in image
  --group, -g GROUP     Add group to /etc/group in image
  --filetools           Add common file manipulation tools
  --runtime, -R RUNTIME
                        Set container engine for building
  --buildcmd, -B BUILDCMD
                        Set command for building
  --version             show program's version number and exit

Docker options:
  --tag, -t TAG         Tag to apply to Docker image
  --cmd, -c CMD
  --entrypoint, -e ENTRYPOINT

Output options:
  --no-build, -n        Do not build Docker image
  --output-dir, -o OUTPUT_DIR

Security options:
  --no-host-lookup      Reject bare user/group names; require colon-delimited
                        entries.
  --allow-sensitive     Allow copying known-sensitive host paths (/etc/shadow,
                        ~/.ssh/*, etc.).
  --nss-modules NSS_MODULES
                        Comma-separated list of nss modules to copy into the
                        image (default: files,dns). Limits CVE surface vs.
                        copying every libnss*.
  --label KEY=VALUE     Add an OCI image label. Repeatable.

Compression options:
  --compress            Apply UPX compression to ELF executables in the image.
  --compress-level {normal,best,ultra}
                        UPX level when --compress is set (default: best).
  --compress-libs       Also compress shared libraries (deprecated UPX
                        feature; increases incompatibility risk — use at your
                        own risk).

Output options (advanced):
  --sbom PATH           Write an SBOM of the build context to PATH (requires
                        syft).
  --sbom-format {spdx-json,cyclonedx-json,syft-json}
                        SBOM format (default: spdx-json).
  --output-oci PATH     Write the image to PATH as an OCI image-layout
                        archive instead of building it through a container
                        engine. Assembled in pure Python — no daemon, no
                        buildx, no socket. Load the archive with skopeo,
                        oras, podman load, or docker load (the last needs
                        the containerd image store).

Logging options:
  --verbose
  --debug

A simple example

Create a sed image:

dockerize -t sed /bin/sed

Use it:

$ echo hello world | docker run -i sed s/world/jupiter
hello jupiter

A more complicated example

Stage some default content, then create an image named httpd:

mkdir -p /tmp/www && echo '<h1>dockerize2</h1>' > /tmp/www/index.html

dockerize -t httpd \
  -a /tmp/www /var/www \
  --entrypoint '/usr/sbin/mini_httpd -D -d /var/www -p 80' \
  /usr/sbin/mini_httpd

Serve the baked-in content:

docker run --rm -p 8080:80 httpd

Serve your own content instead:

docker run --rm -p 8080:80 -v /my/content:/var/www httpd

Acknowledgements

See NOTICE for credit to the original dockerize project and Lars Kellogg-Stedman, on whose work this fork builds.

Development of dockerize2 is assisted by Claude (Anthropic), used as a pair-programming and code-review tool. The maintainer directs all work, reviews each pull request, and retains editorial control over what is merged.