Skip to content

Reduce container registry data transfers

DETAILS: Tier: Free, Premium, Ultimate Offering: SaaS, self-managed

Depending on the frequency with which images or tags are downloaded from the container registry, data transfers can exceed the GitLab.com limit. This page offers several recommendations and tips for reducing the amount of data you transfer with the container registry.

Check data transfer use

Transfer usage is not available within the GitLab UI. GitLab-#350905 is the epic tracking the work to surface this information. In the interim, GitLab team members may reach out to customers that have significantly exceeded their transfer limits to better understand their use case and offer ways to reduce data transfer usage.

Determine image size

Use these tools and techniques to determine your image's size:

  • Skopeo: use the Skopeo inspect command to examine layer count and sizes through API calls. You can therefore inspect this data prior to running docker pull IMAGE.

  • Docker in CI: examine and record the image size when using GitLab CI prior to pushing an image with Docker. For example:

    docker inspect "$CI_REGISTRY_IMAGE:$IMAGE_TAG" \
          | awk '/"Size": ([0-9]+)[,]?/{ printf "Final Image Size: %d\n", $2 }'
  • Dive is a tool for exploring a Docker image, layer contents, and discovering ways to reduce its size.

Reduce image size

Use a smaller base image

Consider using a smaller base image, such as Alpine Linux. An Alpine image is around 5 MB, which is several times smaller than popular base images such as Debian. If your application is distributed as a self-contained static binary, such as for Go applications, you can also consider using the Docker scratch base image.

If you need to use a specific base image OS, look for -slim or -minimal variants, as this helps to reduce the image size.

Also be mindful about the operating system packages you install on top of your base image. These can add up to hundreds of megabytes. Try keeping the number of installed packages to the bare minimum.

Multi-stage builds can be a powerful ally in cleaning up transient build dependencies.

You may also consider using tools such as these:

  • DockerSlim provides a set of commands to reduce the size of your container images.
  • Distroless images contain only your application and its runtime dependencies. They don't contain package managers, shells, or any other programs you would expect to find in a standard Linux distribution.

Minimize layers

Every instruction in your Dockerfile leads to a new layer, which records the file system changes applied during such an instruction. In general, more or larger layers lead to larger images. Try to minimize the number of layers to install the packages in the Dockerfile. Otherwise, this may cause each step in the build process to increase the image size.

There are multiple strategies to reduce the number and size of layers. For example, instead of using a RUN command per operating system package that you want to install (which would lead to a layer per package), you can install all the packages on a single RUN command to reduce the number of steps in the build process and reduce the size of the image.

Another useful strategy is to ensure that you remove all transient build dependencies and disable or empty the operating system package manager cache before and after installing a package.

When building your images, make sure you only copy the relevant files. For Docker, using a .dockerignore helps ensure that the build process ignores irrelevant files.

You can use other third-party tools to minify your images, such as DockerSlim. Be aware that if used improperly, such tools may remove dependencies that your application needs to operate under certain conditions. Therefore, it's preferable to strive for smaller images during the build process instead of trying to minify images afterward.

Use multi-stage builds

With multi-stage builds, you use multiple FROM statements in your Dockerfile. Each FROM instruction can use a different base, and each begins a new build stage. You can selectively copy artifacts from one stage to another, leaving behind everything you don't want in the final image. This is especially useful when you need to install build dependencies, but you don't need them to be present in your final image.

Use an image pull policy

When using the docker or docker+machine executors, you can set a pull_policy parameter in your runner config.toml that defines how the runner works when pulling Docker images. To avoid transferring data when using large and rarely updated images, consider using the if-not-present pull policy when pulling images from remote registries.

Use Docker layer caching

When running docker build, each command in Dockerfile results in a layer. These layers are kept as a cache and can be reused if there haven't been any changes. You can specify a tagged image to be used as a cache source for the docker build command by using the --cache-from argument. Multiple images can be specified as a cache source by using multiple --cache-from arguments. This can speed up your builds and reduce the amount of data transferred. For more information, see the documentation on Docker layer caching.

Check automation frequency

We often create automation scripts bundled into container images to perform regular tasks on specific intervals. You can reduce the frequency of those intervals in cases where the automation is pulling container images from the GitLab Registry to a service outside of GitLab.com.

Move to GitLab Premium or Ultimate

GitLab.com data transfer limits are set at the tier level. If you need a higher limit, consider upgrading to GitLab Premium or Ultimate.

Purchase additional data transfer

Read more about managing your data transfer limits.

Related issues

  • You may want to rebuild your image when the base Docker image is updated. However, the pipeline subscription limit is too low to leverage this feature. As a workaround, you can rebuild daily or multiple times per day. GitLab-#225278 proposes raising the limit to help with this workflow.