dazelisk 0.1.1

Containerized bazelisk wrapper: run bazelisk transparently inside a Docker container

dazelisk

dazelisk (portmanteau of Docker + Bazelisk) is a containerized bazelisk wrapper focused on simplicity and supply chain safety, targeting modern Bazel versions, that aims to isolate and normalize the build environment across machines and platforms by providing a transparent containerized wrapping of Bazel invocations.

In place of

bazelisk build //...

you run

dazelisk build //...

and the command executes transparently inside a Linux Docker container, giving a consistent build environment across Linux, WSL2, and macOS (Apple Silicon only).

Any argument that is not a --dazelisk-* flag is forwarded verbatim to bazelisk.

Supported platforms

• Linux
• macOS (Apple silicon only)
• Windows (WSL2 only)

Requirements

• Python 3.12+
• Docker (Engine on Linux/WSL2, Desktop on macOS) with a running daemon

dazelisk has no third-party Python dependencies to strengthen supply chain safety.

Installation

uv pip install dazelisk

This installs the dazelisk entry point.

Container image

The container image can be configured. By default dazelisk uses the official image published on Docker Hub, martinopilia/dazelisk — a minimal image containing only what is needed to run bazelisk, so that you do not pay for what you do not use. It is pulled automatically on first use; no configuration is required. dazelisk never builds an image, it only pulls one.

To use a custom image instead, set DAZELISK_IMAGE:

# pin a digest for exact reproducibility
export DAZELISK_IMAGE=docker.io/myorg/my-bazel-image@sha256:<digest>

For reproducibility and supply-chain safety, a custom DAZELISK_IMAGE should also be an immutable reference (a pinned tag or a digest), not a mutable tag.

Usage

dazelisk build //...
dazelisk test //foo:bar --config=ci
dazelisk --dazelisk-shell                 # interactive shell in the container
dazelisk --dazelisk-as-root run //tool    # run this command as root

Supported arguments and environment variables

See

dazelisk --dazelisk-help

Cache

A per-user Docker volume (dazelisk-cache-<uid>) is mounted at $HOME/.cache inside the container and shared across all of your worktrees. A volume (rather than a bind mount) avoids cross-platform permission issues and gives native Linux filesystem performance on macOS.

GPU passthrough

If an NVIDIA GPU and the NVIDIA Container Toolkit (nvidia-container-cli) are both present, dazelisk adds --gpus all automatically. Disable with DAZELISK_GPU_PASSTHROUGH=0.

Container naming & isolation

Containers are scoped by image, user and worktree to avoid collisions, and changing DAZELISK_IMAGE automatically uses a fresh container.

Troubleshooting

• Image pull failure — check your network/registry access and that DAZELISK_IMAGE (if set) is a valid, reachable reference.
• Docker CLI not found / daemon not reachable — install Docker and start it (sudo systemctl start docker, or launch Docker Desktop).
• Permission denied talking to Docker — sudo usermod -aG docker $USER && newgrp docker.
• Invalid DAZELISK_* JSON — dazelisk reports the parse error; fix the JSON.
• --dazelisk-mount-cache on macOS — unsupported; the volume lives inside the Docker Desktop VM with no host-reachable path.

Development

See DESIGN.md for design and security guidelines.

Run checks and tests:

./pre_push.sh

Or individually:

uv run pytest              # tests
uv run ruff check          # lint
uv run ruff format         # auto-format the code
uv run ruff format --check # verify formatting
uv run ty check            # type-check

Build and load the image:

cd image
bazelisk test //...                 # build the image and run its tests
bazelisk run //:load                # load it into the local Docker daemon

Manual push to registry:

bazelisk run //:push -- --repository <repository> --tag <tag>

After changing anything under image/, refresh the pinned default digest:

python scripts/sync_image_digest.py           # rewrite the pinned digest
python scripts/sync_image_digest.py --check   # verify if the digest is up to date

To add a Debian package, edit image/packages.yaml and regenerate the lock:

cd image
bazelisk run @bookworm//:lock

Related projects

• dazel — A transparent proxy that runs Bazel commands inside a Docker container.

License

dazelisk is released under the MIT License. See LICENSE.