Earthfile reference
Earthfiles are comprised of a series of target declarations and recipe definitions. Earthfiles are named Earthfile, regardless of their location in the codebase. For backwards compatibility, the name build.earth is also permitted. This compatibility will be removed in a future version.
Earthfiles have the following rough structure:
<base-recipe>...​<target-name>:<recipe>...​<target-name>:<recipe>...
Each recipe contains a series of commands, which are defined below. For an introduction into Earthfiles, see the Basics page.
FROM <image-name>FROM [--build-arg <key>=<value>] [--platform <platform>] <target-ref>
The FROM command initializes a new build environment and sets the base image for subsequent instructions. It works similarly to the classical Dockerfile FROM instruction, but it has the added ability to use another target's image as the base image for the build. For example: FROM +another-target.
The FROM ... AS ... form available in the classical Dockerfile syntax is not supported in Earthfiles. Instead, define a new Earthly target. For example, the following Dockerfile
# Dockerfile​FROM alpine:3.11 AS build# ... instructions for build​FROM build as another# ... further instructions inheriting build​FROM busybox as yet-anotherCOPY --from=build ./a-file ./
can become
# Earthfile​build:FROM alpine:3.11# ... instructions for buildSAVE ARTIFACT ./a-file​another:FROM +build# ... further instructions inheriting build​yet-another:FROM busyboxCOPY +build/a-file ./
Sets a value override of <value> for the build arg identified by <key>. See also BUILD for more details about the --build-arg option.
Specifies the platform to build on.
For more information see the multi-platform guide.
FROM DOCKERFILE [--build-arg <key>=<value>] [--platform <platform>] [--target <target-name>] <context-path>
The FROM DOCKERFILE command initializes a new build environment, inheriting from an existing Dockerfile. This allows the use of Dockerfiles in Earthly builds.
The <context-path> is the path where the Dockerfile build context exists. It is assumed that a file named Dockerfile exists in that directory. The context path can be either a path on the host system, or an artifact reference, pointing to a directory containing a Dockerfile.
This feature is currently in Beta and it has the following limitations:
This feature only works with files named
Dockerfile. The equivalent of the-foption available indocker buildhas not yet been implemented..dockerignoreis not used.The newer experimental features which exist in the Dockerfile syntax are not guaranteed to work correctly.
Sets a value override of <value> for the Dockerfile build arg identified by <key>. This option is similar to the docker build --build-arg <key>=<value> option.
In a multi-stage Dockerfile, sets the target to be used for the build. This option is similar to the docker build --target <target-name> option.
Specifies the platform to build on.
For more information see the multi-platform guide.
RUN [--push] [--entrypoint] [--privileged] [--secret <env-var>=<secret-ref>] [--ssh] [--mount <mount-spec>] [--] <command>(shell form)RUN [[<flags>...], "<executable>", "<arg1>", "<arg2>", ...](exec form)
The RUN command executes commands in the build environment of the current target, in a new layer. It works similarly to the Dockerfile RUN command, with some added options.
The command allows for two possible forms. The exec form runs the command executable without the use of a shell. The shell form uses the default shell (/bin/sh -c) to interpret the command and execute it. In either form, you can use a \ to continue a single RUN instruction onto the next line.
When the --entrypoint flag is used, the current image entrypoint is used to prepend the current command.
To avoid any ambiguity regarding whether an argument is a RUN flag option or part of the command, the delimiter -- may be used to signal the parser that no more RUN flag options will follow.
Marks the command as a "push command". Push commands are only executed if all other non-push instructions succeed. In addition, push commands are never cached, thus they are executed on every applicable invocation of the build.
Push commands are not run by default. Add the --push flag to the earthly invocation to enable pushing. For example
earthly --push +deploy
Push commands were introduced to allow the user to define commands that have an effect external to the build. This kind of effects are only allowed to take place if the entire build succeeds. Good candidates for push commands are uploads of artifacts to artifactories, commands that make a change to an external environment, like a production or staging environment.
Note that non-push commands are not allowed to follow a push command within a recipe.
Prepends the currently defined entrypoint to the command.
This option is useful for replacing docker run in a traditional build environment. For example, a command like
docker run --rm -v "$(pwd):/data" cytopia/golint .
Might become the following in an Earthfile
FROM cytopia/golingCOPY . /dataRUN --entrypoint .
Allows the command to use privileged capabilities.
Note that privileged mode is not enabled by default. In order to use this option, you need to additionally pass the flag --allow-privileged (or -P) to the earthly command. Example:
earthly --allow-privileged +some-target
Makes available a secret, in the form of an env var (its name is defined by <env-var>), to the command being executed.
The <secret-ref> needs to be of the form +secrets/<secret-id>, where <secret-id> is the identifier passed to the earthly command when passing the secret: earthly --secret <secret-id>=<value>.
Here is an example:
release:RUN --push --secret GITHUB_TOKEN=+secrets/GH_TOKEN github-release upload
earthly --secret GH_TOKEN="the-actual-secret-token-value" +release
An empty string is also allowed for <secret-ref>, allowing for optional secrets, should it need to be disabled.
release:ARG SECRET_ID=+secrets/GH_TOKENRUN --push --secret GITHUB_TOKEN=$SECRET_ID github-release upload
earthly --build-arg SECRET_ID="" +release
See also the Cloud secrets guide.
Allows a command to access the ssh authentication client running on the host via the socket which is referenced by the environment variable SSH_AUTH_SOCK.
Here is an example:
RUN mkdir -p ~/.ssh && \echo 'github.com ssh-rsa AAAAB3NzaC1yc2EAAAABIwAAAQEAq2A7hRGmdnm9tUDbO9IDSwBK6TbQa+PXYPCPy6rbTrTtw7PHkccKrpp0yVhp5HdEIcKr6pLlVDBfOLX9QUsyCOV0wzfjIJNlGEYsdlLJizHhbn2mUjvSAHQqZETYP81eFzLQNnPHt4EVVUh7VfDESU84KezmD5QlWpXLmvU31/yMf+Se8xhHTvKSCZIFImWwoG6mbUoWf9nzpIoaSjB+weqqUUmpaaasXVal72J+UX2B+2RPW3RcT0eOzQgqlJL3RKrTJvdsjE3JEAvGq3lGHSZXy28G3skua2SmVi/w4yCE6gbODqnTWlg7+wC604ydGXA8VJiS5ap43JXiUFFAaQ==' >> ~/.ssh/known_hosts && \echo 'gitlab.com ssh-rsa AAAAB3NzaC1yc2EAAAADAQABAAABAQCsj2bNKTBSpIYDEGk9KxsGh3mySTRgMtXL583qmBpzeQ+jqCMRgBqB98u3z++J1sKlXHWfM9dyhSevkMwSbhoR8XIq/U0tCNyokEi/ueaBMCvbcTHhO7FcwzY92WK4Yt0aGROY5qX2UKSeOvuP4D6TPqKF1onrSzH9bx9XUf2lEdWT/ia1NEKjunUqu1xOB/StKDHMoX4/OKyIzuS0q/T1zOATthvasJFoPrAjkohTyaDUz2LN5JoH839hViyEG82yB+MjcFV5MU3N1l1QL3cVUCh93xSaua1N85qivl+siMkPGbO5xR/En4iEY6K2XPASUEMaieWVNTRCtJ4S8H+9' >> ~/.ssh/known_hostsRUN --ssh git config --global url."[email protected]:".insteadOf "https://github.com/" && \go mod download
Mounts a file or directory in the context of the build environment.
The <mount-spec> is defined as a series of comma-separated list of key-values. The following keys are allowed
Key | Description | Example |
| The type of the mount. Currently only |
|
| The target path for the mount. |
|
| The secret ID for the contents of the |
|
Example:
ENV GOCACHE=/go-cacheRUN --mount=type=cache,target=/go-cache go build main.go
Note that mounts cannot be shared between targets, nor can they be shared within the same target, if the build-args differ between invocations.
COPY [options...] <src>... <dest>(classical form)COPY [options...] <src-artifact>... <dest>(artifact form)
The command COPY allows copying of files and directories between different contexts.
The command may take a couple of possible forms. In the classical form, COPY copies files and directories from the build context into the build environment - in this form, it works similarly to the Dockerfile COPY command. In the artifact form, COPY copies files or directories (also known as "artifacts" in this context) from the artifact environment of other build targets into the build environment of the current target. Either form allows the use of wildcards for the sources.
The parameter <src-artifact> is an artifact reference and is generally of the form <target-ref>/<artifact-path>, where <target-ref> is the reference to the target which needs to be built in order to yield the artifact and <artifact-path> is the path within the artifact environment of the target, where the file or directory is located. The <artifact-path> may also be a wildcard.
Note that the Dockerfile form of COPY whereby you can reference a source as a URL is not yet supported in Earthfiles.
To prevent Earthly from copying unwanted files, you may specify file patterns to be excluded from the build context using an .earthignore file. This file has the same syntax as a .dockerignore file.
The option --dir changes the behavior of the COPY command to copy the directories themselves, rather than the contents of the directories. It allows the command to behave similarly to a cp -r operation on a unix system. This allows the enumeration of several directories to be copied over on a single line (and thus, within a single layer). For example, the following two are equivalent with respect to what is being copied in the end (but not equivalent with respect to the number of layers used).
COPY dir1 dir1COPY dir2 dir2COPY dir3 dir3
COPY --dir dir1 dir2 dir3 ./
Sets a value override of <value> for the build arg identified by <key>, when building the target containing the mentioned artifact. See also BUILD for more details about the --build-arg option.
Instructs Earthly to not overwrite the file creation timestamps with a constant.
Instructs Earthly to keep file ownership information. This applies only to the artifact form and has no effect otherwise.
Although this option is present in classical Dockerfile syntax, it is not supported by Earthfiles. You may instead use a combination of SAVE ARTIFACT and COPY artifact form commands to achieve similar effects. For example, the following Dockerfile
# DockerfileCOPY --from=some-image /path/to/some-file.txt ./
... would be equivalent to final-target in the following Earthfile
# Earthfileintermediate:FROM some-imageSAVE ARTIFACT /path/to/some-file.txt​final-target:COPY +intermediate/some-file.txt ./
In artifact form, it specifies the platform to build the artifact on.
For more information see the multi-platform guide.
Assuming the following directory tree, of a folder named test:
test└── file
Here is how the following copy commands will behave:
# Copies the contents of the test directory.# To access the file, it would be found at ./fileCOPY test .​# Also copies the contents of the test directory.# To access the file, it would be found at ./fileCOPY test/* .​# Copies the whole test folder.# To access the file, it would be found at ./test/fileCOPY --dir test .
For detailed examples demonstrating how other scenarios may function, please see our test suite.
GIT CLONE [--branch <git-ref>] [--keep-ts] <git-url> <dest-path>
The command GIT CLONE clones a git repository from <git-url>, optionally referenced by <git-ref>, into the build environment, within the <dest-path>.
In contrast to an operation like RUN git clone <git-url> <dest-path>, the command GIT CLONE is cache-aware and correctly distinguishes between different git commit IDs when deciding to reuse a previous cache or not. In addition, GIT CLONE can also use Git authentication configuration passed on to earthly, whereas RUN git clone would require additional secrets passing, if the repository is not publicly accessible.
Points the HEAD to the git reference specified by <git-ref>. If this option is not specified, then the remote HEAD is used instead.
Instructs Earthly to not overwrite the file creation timestamps with a constant.
SAVE ARTIFACT [--keep-ts] [--keep-own] <src> [<artifact-dest-path>] [AS LOCAL <local-path>]
The command SAVE ARTIFACT copies a file, a directory, or a series of files and directories represented by a wildcard, from the build environment into the target's artifact environment.
If AS LOCAL ... is also specified, it additionally marks the artifact to be copied to the host at the location specified by <local-path>, once the build is deemed as successful.
If <artifact-dest-path> is not specified, it is inferred as /.
Files within the artifact environment are also known as "artifacts". Once a file has been copied into the artifact environment, it can be referenced in other places of the build (for example in a COPY command), using an artifact reference.
Instructs Earthly to not overwrite the file creation timestamps with a constant.
Instructs Earthly to keep file ownership information.
Assuming the following directory tree, of a folder named test:
test└── file
Here is how the following SAVE ARTIFACT ... AS LOCAL commands will behave:
WORKDIR baseCOPY test .​# This will copy the base folder into the output directory.# You would find file at out-dot/base/file.SAVE ARTIFACT . AS LOCAL out-dot/​# This will copy the contents of the base folder into the output directory.# You would find sub-file at out-glob/file. Note the base directory is not in the output.SAVE ARTIFACT ./* AS LOCAL out-glob/
For detailed examples demonstrating how other scenarios may function, please see our test suite.
SAVE IMAGE [--cache-from=<cache-image>] [--push] <image-name>...(output form)SAVE IMAGE --cache-hint(cache hint form)
In the output form, the command SAVE IMAGE marks the current build environment as the image of the target and assigns an output image name.
In the cache hint form, it instructs Earthly that the current target should be included as part of the explicit cache. For more information see the shared caching guide.
The --push options marks the image to be pushed to an external registry after it has been loaded within the docker daemon available on the host.
If inline caching is enabled, the --push option also instructs Earthly to use the specified image names as cache sources.
The actual push is not executed by default. Add the --push flag to the earthly invocation to enable pushing. For example
earthly --push +docker-image
Adds additional cache sources to be used when --use-inline-cache is enabled. For more information see the shared caching guide.
Instructs Earthly that the current target should be included as part of the explicit cache. For more information see the shared caching guide.
BUILD [--build-arg <key>=<value>] [--platform <platform>] <target-ref>
The command BUILD instructs Earthly to additionally invoke the build of the target referenced by <target-ref>, where <target-ref> follows the rules defined by target referencing.
Sets a value override of <value> for the build arg identified by <key>.
The override value of a build arg may be a constant string
--build-arg SOME_ARG="a constant value"
or an expression involving other build args
--build-arg SOME_ARG="a value based on other args, like $ANOTHER_ARG and $YET_ANOTHER_ARG"
or a dynamic expression, based on the output of a command executed in the context of the build environment. In this case, the build arg becomes a "variable build arg".
--build-arg SOME_ARG=$(find /app -type f -name '*.php')
Specifies the platform to build on.
This flag may be repeated in order to instruct the system to perform the build for multiple platforms. For example
build-all-platforms:BUILD --platform=linux/amd64 --platform=linux/arm/v7 +build
For more information see the multi-platform guide.
ARG <name>[=<default-value>]
The command ARG declares a variable (or arg) with the name <name> and with an optional default value <default-value>. If no default value is provided, then empty string is used as the default value.
This command works similarly to the Dockerfile ARG command, with a few differences regarding the scope and the predefined args (called builtin args in Earthly). The variable's scope is always limited to the current target's recipe and only from the point it is declared onwards. For more information regarding builtin args, see the builtin args page.
The value of an arg can be overridden either from the earthly command
earthly --build-arg <name>=<override-value>
or from a command from another target, when implicitly or explicitly invoking the target containing the ARG
BUILD --build-arg <name>=<override-value> <target-ref>COPY --build-arg <name>=<override-value> <target-ref>/<artifact-path>... <dest-path>FROM --build-arg <name>=<override-value> <target-ref>
for example
BUILD --build-arg PLATFORM=linux +binaryCOPY --build-arg PLATFORM=linux +binary/bin ./FROM --build-arg NAME=john +docker-image
A number of builtin args are available and are pre-filled by Earthly. For more information see builtin args.
WITH DOCKER [--pull <image-name>] [--load <image-name>=<target-ref>] [--compose <compose-file>][--service <compose-service>] [--build-arg <key>=<value>]<commands>...END
The clause WITH DOCKER initializes a Docker daemon to be used in the context of a RUN command. The Docker daemon can be pre-loaded with a set of images using options such as -pull and --load. Once the execution of the RUN command has completed, the Docker daemon is stopped and all of its data is deleted, including any volumes and network configuration. Any other files that may have been created are kept, however.
The clause WITH DOCKER automatically implies the RUN --privileged flag.
The WITH DOCKER clause only supports the command RUN. Other commands (such as COPY) need to be run either before or after WITH DOCKER ... END. In addition, only one RUN command is permitted within WITH DOCKER. However, multiple shell commands may be stringed together using ; or &&.
A typical example of a WITH DOCKER clause might be:
FROM earthly/dind:alpineWORKDIR /testCOPY docker-compose.yml ./WITH DOCKER \--compose docker-compose.yml \--load image-name:latest=+some-target \--pull some-image:latestRUN docker run ... && \docker run ... && \...END
For more examples, see the Docker in Earthly guide and the Integration testing guide.
For performance reasons, it is recommended to use a Docker image that already contains dockerd. If dockerd is not found, Earthly will attempt to install it.
Earthly provides officially supported images such as earthly/dind:alpine and earthly/dind:ubuntu to be used together with WITH DOCKER.
Pulls the Docker image <image-name> from a remote registry and then loads it into the temporary Docker daemon created by WITH DOCKER.
This option may be repeated in order to provide multiple images to be pulled.
It is recommended that you avoid issuing RUN docker pull ... and use WITH DOCKER --pull ... instead. The classical docker pull command does not take into account Earthly caching and so it would redownload the image much more frequently than necessary.
Builds the image referenced by <target-ref> and then loads it into the temporary Docker daemon created by WITH DOCKER. The image can be referenced as <image-name> within WITH DOCKER.
This option may be repeated in order to provide multiple images to be loaded.
Loads the compose definition defined in <compose-file>, adds all applicable images to the pull list and starts up all applicable compose services within.
This option may be repeated, thus having the same effect as repeating the -f flag in the docker-compose command.
Specifies which compose service to pull and start up. If no services are specified and --compose is used, then all services are pulled and started up.
This option can only be used if --compose has been specified.
This option may be repeated in order to specify multiple services.
Sets a value override of <value> for the build arg identified by <key>, when building a <target-ref> (specified via --load). See also BUILD for more details about the --build-arg option.
Specifies the platform for any referenced --load and --pull images.
For more information see the multi-platform guide.
DOCKER PULL <image-name>
DOCKER PULL is now deprecated and will not be supported in future versions of Earthly. Please use WITH DOCKER --pull <image-name> instead.
DOCKER LOAD [--build-arg <name>=<override-value>] <target-ref> <image-name>
DOCKER LOAD is now deprecated and will not be supported in future versions of Earthly. Please use WITH DOCKER --load <image-name>=<target-ref> instead.
CMD ["executable", "arg1", "arg2"](exec form)CMD ["arg1, "arg2"](as default arguments to the entrypoint)CMD command arg1 arg2(shell form)
The command CMD sets default arguments for an image, when executing as a container. It works the same way as the Dockerfile CMD command.
LABEL <key>=<value> <key>=<value> ...
The LABEL command adds label metadata to an image. It works the same way as the Dockerfile LABEL command.
EXPOSE <port> <port> ...EXPOSE <port>/<protocol> <port>/<protocol> ...
The EXPOSE command marks a series of ports as listening ports within the image. It works the same way as the Dockerfile EXPOSE command.
ENV <key> <value>ENV <key>=<value>
The ENV command sets the environment variable <key> to the value <value>. It works the same way as the Dockerfile ENV command.
Do not use the ENV command for secrets used during the build. All ENV values used during the build are persisted within the image itself. See the RUN --secret option to pass secrets to build instructions.
ENTRYPOINT ["executable", "arg1", "arg2"](exec form)ENTRYPOINT command arg1 arg2(shell form)
The ENTRYPOINT command sets the default command or executable to be run when the image is executed as a container. It works the same way as the Dockerfile ENTRYPOINT command.
VOLUME <path-to-target-mount> <path-to-target-mount> ...VOLUME ["<path-to-target-mount>", <path-to-target-mount> ...]
The VOLUME command creates a mount point at the specified path and marks it as holding externally mounted volumes. It works the same way as the Dockerfile VOLUME command.
USER <user>[:<group>]USER <UID>[:<GID>]
The USER command sets the user name (or UID) and optionally the user group (or GID) to use when running the image and also for any subsequent instructions in the build recipe. It works the same way as the Dockerfile USER command.
WORKDIR <path-to-dir>
The WORKDIR command strs the working directory for other commands that follow in the recipe. The working directory is also persisted as the default directory for the image. If the directory does not exist, it is automatically created. This command works the same way as the Dockerfile WORKDIR command.
HEALTHCHECK NONE(disable healthchecking)HEALTHCHECK [--interval=DURATION] [--timeout=DURATION] [--start-period=DURATION] [--retries=N] CMD command arg1 arg2(check container health by running command inside the container)
The HEALTHCHECK command tells Docker how to test a container to check that it is still working. It works the same way as the Dockerfile HEALTHCHECK command, with the only exception that the exec form of this command is not yet supported.
Sets the time interval between health checks. Defaults to 30s.
Sets the timeout for a single run before it is considered as failed. Defaults to 30s.
Sets an initialization time period in which failures are not counted towards the maximum number of retries. Defaults to 0s.
Sets the number of retries before a container is considered unhealthy. Defaults to 3.
The classical SHELL Dockerfile command is not yet supported. Use the exec form of RUN, ENTRYPOINT and CMD instead and prepend a different shell.
The classical ADD Dockerfile command is not yet supported. Use COPY instead.
The classical ONBUILD Dockerfile command is not supported.
The classical STOPSIGNAL Dockerfile command is not yet supported.
