We use cookies to understand how people use Depot.
Container Builds

Best practice Dockerfile for Rust with cargo-chef and sccache

Below is an example Dockerfile that we have used and recommend at Depot for building images for Rust applications.

FROM rust:1.75 AS base
RUN cargo install --locked cargo-chef sccache
ENV RUSTC_WRAPPER=sccache SCCACHE_DIR=/sccache
 
FROM base AS planner
WORKDIR /app
COPY . .
RUN cargo chef prepare --recipe-path recipe.json
 
FROM base AS builder
WORKDIR /app
COPY --from=planner /app/recipe.json recipe.json
RUN --mount=type=cache,target=/usr/local/cargo/registry \
    --mount=type=cache,target=/usr/local/cargo/git \
    --mount=type=cache,target=$SCCACHE_DIR,sharing=locked \
    cargo chef cook --release --recipe-path recipe.json
COPY . .
RUN --mount=type=cache,target=/usr/local/cargo/registry \
    --mount=type=cache,target=/usr/local/cargo/git \
    --mount=type=cache,target=$SCCACHE_DIR,sharing=locked \
    cargo build

In addition to the standard best practices when writing Dockerfiles, here we are also leveraging cargo-chef and sccache to speed up our Rust build.

Using cargo-chef for dependency management

When you install multiple crates with one command like cargo build, Docker treats any change in the output of cargo build as a change to the entire command. This means that Docker will attempt to execute that command again (re-downloading and installing **all **crates) every time you make an unrelated change to your source code or Dockerfile.

There are various workarounds online to manually manage and copy individual packages into your container while building in order to avoid invalidating the cache on every build, but these are cumbersome and prone to bugs. Our preferred solution is to use cargo-chef, which allows you to separate building the dependencies and building the source code so that Docker sees them as different steps and can cache them separately.

Using sccache for additional dependency management

Even though cargo-chef separates your third-party dependencies from your source code, compiling and downloading your third-party dependencies is still considered one operation. This means that if a single dependency changes, there will be a cache miss and all of them will have to be re-downloaded and compiled, even though they haven't changed.

If you have a more fine-grained cache, you only have to rebuild the changed dependencies. Enter sccache, which caches individual compilation artifacts so that they can be reused at a more granular level during future compilations. This allows you to recompile individual dependencies only when needed, rather than everything or nothing.

Explanation of the Dockerfile

At a high level, here are the things we're optimizing in our Docker build for a Rust application::

Multi-stage builds via multiple FROM statements

  • cargo-chef for dependency management
  • sccache for dependency caching
  • BuildKit cache mounts for finer-grained caching between builds

Stage 1: FROM rust:1.75 AS base

FROM rust:1.75 AS base
RUN cargo install --locked cargo-chef sccache
ENV RUSTC_WRAPPER=sccache SCCACHE_DIR=/sccache

Here, we use rust:1.75 as our base image and set the stage name to be used in later stages. In addition to this base image, we install sccache and cargo-chef.

We then set the SCCACHE_DIR environment variable so that sccache stores compilation artifacts in the /sccache directory and the RUSTC_WRAPPER environment variable so that Cargo “wraps” the execution of the Rust compiler commands in an sccache call. That way, we can take advantage of sccache's cached dependencies when building the final image.

Stage 2: FROM base AS planner

FROM base AS planner
WORKDIR /app
COPY . .
RUN cargo chef prepare --recipe-path recipe.json

Next, we're building off the base stage by creating the recipe that will later be used to build our application with cargo-chef. We use two commands from cargo-chef that together act as a replacement for the standard cargo build command when building dependencies.

Using cargo-chef to separate building dependencies from building source code

cargo chef prepare looks at your Cargo.toml and auto-generated Cargo.lock files, determines all of your dependencies, and then creates a recipe.json file, which is a dependency tree of your project.

By creating the dependency tree separately from the actual installation of dependencies, we can cache them independently so that all dependencies don't need to be rebuilt whenever the source code changes.

Stage 3: FROM base as builder

FROM base AS builder
WORKDIR /app
COPY --from=planner /app/recipe.json recipe.json
RUN --mount=type=cache,target=/usr/local/cargo/registry \
    --mount=type=cache,target=/usr/local/cargo/git \
    --mount=type=cache,target=$SCCACHE_DIR,sharing=locked \
    cargo chef cook --release --recipe-path recipe.json
COPY . .
RUN --mount=type=cache,target=/usr/local/cargo/registry \
    --mount=type=cache,target=/usr/local/cargo/git \
    --mount=type=cache,target=$SCCACHE_DIR,sharing=locked \
    cargo build

cargo chef cook takes the recipe.json file and runs cargo build under the hood on each package independently. In this stage, we're copying the recipe.json file from the previous planner stage. If the recipe has not changed, then the step to build the dependencies with cargo chef cook can be skipped. After the dependencies have been built, we finally build the source code with the final cargo build command.

Using sccache to retain dependencies between builds

We use a cache mount to attach the sccache directory to the build. This type of cache mount gives you a more fine-grained level of caching that allows you to skip recomputing bits of work when the layer cache invalidates. Using sccache with a cache mount allows you to skip rebuilding compiled artifacts that have already been built even when certain layers in the build are invalidated.

Caching the Cargo registry and git directories

In this stage, we also use additional cache mounts to store the Cargo registry and git directories. The Cargo registry stores the crates that have already been downloaded, and the git directory stores any git dependencies. Usually only one or a few packages have actually changed and need to be re-downloaded and installed. All the other packages can be reused from previous builds with the cache mounts.