Tagging Docker images

How to set registry, repository and tag names on your images

---

Configuring registries

A docker_image target takes an optional registries field, whose value is a list of registry endpoints and aliases:

src/example/BUILD

docker_image(
    name="demo",
    registries=[
        "reg1.company.internal",
        "@company-registry2",
    ]
)

When publishing this image, it will be pushed to these registries by default.

In order to provide registry specific configuration, add them to the Pants configuration under [docker.registries.<alias>] and refer to them by their alias from the docker_image targets, using a @ prefix.

Options for registries in pants.toml:

• address - The registry endpoint.

• default - Use this registry for all docker_image targets that does not provide a value for the registries field. Multiple registries may be used as default at the same time.

• extra_image_tags - Registry specific version tags to apply to the image when using this registry.

• repository - Format the repository part of the image name for this image. See Setting a repository name for details of this option.

• skip_push - Do not push images to this registry during pants publish.

• use_local_alias - Use the registry alias as a shorter name to use locally such as when running an image, useful if the address is unwieldy long. When building images using pants package, the image will be tagged with all image names for the target where as when simply running an image with pants run only the shorter image name will be tagged avoid cluttering the Docker images repository. The shorter image names are automatically skipped for any push operations.

Example:

pants.toml

[docker.registries.company-registry1]
address = "reg1.company.internal"
default = true
extra_image_tags = ["dev"]

[docker.registries.company-registry2]
address = "reg2.company.internal"
skip_push = true

[docker.registries.company-registry3]
address = "reg3.company.internal"
repository = "{parent_directory}/{name}"
use_local_alias = true

src/example/BUILD

docker_image(name="demo")

# This is equivalent to the previous target,
# since company-registry1 is the default registry:
docker_image(
    name="demo",
    registries=["@company-registry1"],
)

# You can mix named and direct registry references.
docker_image(
    name="demo2",
    registries=[
        "@company-registry2",
        "ext-registry.company-b.net:8443",
    ]
)

Setting a repository name

In Docker parlance, an image is identified by a repository and one or more tags within that repository.

You set a repository name using the repository field on docker_image:

src/example/BUILD

docker_image(
    name="demo",
    repository="example/demo",
)

$ pants package src/example:demo
# Will build the image: example/demo:latest

To use a repository only for a specific registry, provide a repository value in the registry configuration, and this can contain placeholders in curly braces that will be interpolated for each image name.

pants.toml

[docker.registries.demo]
address = "reg.company.internal"
repository = "example/{name}"

You can also specify a default repository name in config, and this name can contain placeholders in curly braces that will be interpolated for each docker_image:

pants.toml

[docker]
default_repository = "{directory}/{name}"

src/example/BUILD

docker_image(
    name="demo",
)

The default placeholders are:

• {directory}: The directory the docker_image's BUILD file is in.
• {parent_directory}: The parent directory of {directory}.
• {name}: The name of the docker_image target.
• {build_args.ARG_NAME}: Each defined Docker build arg is available for interpolation under the build_args. prefix.
• {default_repository}: The default repository from configuration.
• {target_repository}: The repository on the docker_image if provided, otherwise the default repository.

Since repository names often conform to patterns like these, this can save you on some boilerplate by allowing you to omit the repository field on each docker_image. But you can always override this field on specific docker_image targets, of course. In fact, you can use these placeholders in the repository field as well, if you find that helpful.

See String interpolation using placeholder values for more information.

Tagging images

When Docker builds images, it can tag them with a set of tags. Pants will apply the tags listed in the image_tags field of docker_image, and any additional tags if defined from the registry configuration (see Configuring registries.

(Note that the field is named image_tags and not just tags, because Pants has its own tags concept, which is unrelated.)

src/example/BUILD

docker_image(
    name="demo",
    repository="example/demo",
    image_tags=["1.2", "example"]
)

When pants builds the src/example:demo target, a single image will be built, with two tags applied:

• example/demo:1.2
• example/demo:example

It's often useful to keep versions of derived images and their base images in sync. Pants helps you out with this by interpolating tags referenced in FROM commands in your Dockerfile into the image_tags in the corresponding docker_image:

src/example/BUILD

# These three are equivalent
docker_image(name="demo1", image_tags=["{tags.upstream}"])
docker_image(name="demo1", image_tags=["{tags.stage0}"])
# The first FROM may also be referred to as "baseimage"
docker_image(name="demo1", image_tags=["{tags.baseimage}"])

# Any stage my be used, and being a format string, you may add extra text as well.
docker_image(name="demo1", image_tags=["{tags.stage1}-custom-suffix"])

src/example/Dockerfile

FROM upstream:1.2 as upstream
# ...
FROM scratch
# ...

This way you can specify a version just once, on the base image, and the derived images will automatically acquire the same version.

You may also use any Docker build arguments (when configured as described in Docker build arguments) for interpolation into the image_tags in the corresponding docker_image:

src/example/BUILD

docker_image(image_tags=["{build_args.ARG_NAME}"])

Using env vars to include dynamic data in tags

You can interpolate dynamic data, such as the current Git commit sha, in an image tag, using environment variables and Docker build args.

For example, you can declare a custom build arg, either in extra_build_args for a specific docker_image target, or for all docker_image targets in pants.toml:

# pants.toml
[docker]
build_args = ["GIT_COMMIT"]

and use this build arg in the image tag:

# src/example/BUILD
docker_image(name="demo", image_tags=["1.2-{build_args.GIT_COMMIT}"])

Then, if you run Pants with the data set in an environment variable of the same name:

$ GIT_COMMIT=$(git rev-parse HEAD) pants package src/example:demo

the value from the environment will be used.

Generating dynamic tags in a plugin

If you don't want to use the environment variable method described above, you'll need to write some custom plugin code. Don't hesitate to reach out for help with this.

We are looking into making some common dynamic data, such as the git sha, automatically available in the core Docker plugin in the future.

Providing additional image tags with a plugin

For cases where more customization is required and using environment variables and interpolation is not enough, the next option is to write a plugin to provide additional tags when building images.

Demonstrated with an example:

example/plugin.py

from pants.backend.docker.target_types import DockerImageTagsRequest, DockerImageTags
from pants.engine.unions import UnionRule
from pants.engine.rules import rule, collect_rules
from pants.engine.target import Target


class CustomDockerImageTagsRequest(DockerImageTagsRequest):
    @classmethod
    def is_applicable(cls, target: Target) -> bool:
        # Optional. Opt-out on a per target basis.
        if some-condition:
            return False
        else:
            return True


@rule
async def custom_image_tags(request: CustomDockerImageTagsRequest) -> DockerImageTags:
    custom_tags = ["some", "tags"]
    return DockerImageTags(custom_tags)


def rules():
    return (
        *collect_rules(),
        UnionRule(DockerImageTagsRequest, CustomDockerImageTagsRequest),
    )

All together: Registries, Repositories and Tags

To illustrate how all the above work together, this target:

src/example/BUILD

docker_image(
    name="demo",
    repository="example/demo",
    registries=["reg1", "reg2"],
    image_tags=["1.0", "latest"]
)

Will create a single image with these full names:

reg1/example/demo:1.0
reg1/example/demo:latest
reg2/example/demo:1.0
reg2/example/demo:latest

String interpolation using placeholder values

As we've seen above, some fields of the docker_image support replacing placeholder values in curly braces with variable text, such as a build arg or base image tag for instance.

The interpolation context (the available placeholder values) depends on which field it is used in. These are the common values available for all fields:

• {tags.<stage>}: The tag of a base image (the FROM instruction) for a particular stage in the Dockerfile. The <stage> is either stageN where N is the numeric index of the stage, starting at 0. The first stage, stage0, is also available under the pseudonym baseimage. If the stage is named (FROM image AS my_stage), then the tag value is also available under that name: {tags.my_stage}.
• {build_args.ARG_NAME}: Each defined Docker build arg is available for interpolation under the build_args. prefix.
• {pants.hash}: This is a unique hash value calculated from all input sources and the Dockerfile. It is effectively a hash of the Docker build context. See note below regarding its stability guarantee.

See Setting a repository name for placeholders specific to the repository field.

The {pants.hash} stability guarantee

The calculated hash value may change between stable versions of Pants for the otherwise same input sources.

Retrieving the tags of an packaged image

When a docker image is packaged, metadata about the resulting image is output to a JSON file artefact. This includes the image ID, as well as the full names that the image was tagged with. This file is written in the same manner as outputs of other packageable targets and available for later steps (for example, a test with runtime_package_dependencies including the docker image target) or in dist/ after pants package. By default, this is available at path.to.target/target_name.docker-info.json.

The structure of this JSON file is:

{
    "version": 1, // always 1, until a breaking change is made to this schema
    "image_id": "sha256:..." // the local Image ID of the computed image
    "registries": [ // info about each registry used for this image
        {
            "alias": "name", // set if the registry is configured in pants.toml, or null if not
            "address": "reg.invalid", // the address of the registry itself
            "repository": "the/repo", // the repository used for the image within the registry
            "tags": [
                {
                    "template": "tag-{...}", // the tag before substituting any placeholders
                    "tag": "tag-some-value", // the fully-substituted tag, actually used to tag the image
                    "uses_local_alias": false, // if this tag used the local alias for the registry or not
                    "name": "reg.invalid/the/repo:tag-some-value", // the full name that the image was tagged with
                }
            ]
        }
    ]
}

This JSON file can be used to retrieve the exact name to place into cloud deploy templates or to use for running locally, especially when using tags with placeholders.