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.
Earthfiles have the following rough structure:
Each recipe contains a series of commands, which are defined below. For an introduction into Earthfiles, see the Basics page.
FROM
Synopsis
FROM <image-name>
FROM [--platform <platform>] [--allow-privileged] <target-ref> [--<build-arg-key>=<build-arg-value>...]
Description
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.
Examples:
Classical reference:
FROM alpine:latest
Local reference:
FROM +another-target
Relative reference:
FROM ./subdirectory+some-target
orFROM ../otherdirectory+some-target
Absolute reference:
FROM /absolute/path+some-target
Remote reference from a public or private git repository:
FROM github.com/example/project+remote-target
The FROM
command does not mark any saved images or artifacts of the referenced target for output, nor does it mark any push commands of the referenced target for pushing. For that, please use BUILD
.
Note
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
can become
Options
--<build-arg-key>=<build-arg-value>
Sets a value override of <build-arg-value>
for the build arg identified by <build-arg-key>
. See also BUILD for more details about build args.
--platform <platform>
Specifies the platform to build on.
For more information see the multi-platform guide.
--allow-privileged
Allows remotely-referenced targets to request privileged capabilities; this flag has no effect when referencing local targets.
Additionally, for privileged capabilities, earthly must be invoked on the command line with the --allow-privileged
(or -P
) flag.
For example, consider two Earthfiles, one hosted on a remote GitHub repo:
and a local Earthfile:
then one can build my-target
by invoking earthly with the --allow-privileged
(or -P
) flag:
--build-arg <key>=<value>
(deprecated)
This option is deprecated. Use --<build-arg-key>=<build-arg-value>
instead.
RUN
Synopsis
RUN [--push] [--entrypoint] [--privileged] [--secret <env-var>=<secret-id>] [--ssh] [--mount <mount-spec>] [--] <command>
(shell form)RUN [[<flags>...], "<executable>", "<arg1>", "<arg2>", ...]
(exec form)
Description
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.
Options
--push
Marks the command as a "push command". 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
Push commands were introduced to allow the user to define commands that have an effect external to the build. 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.
--no-cache
Force the command to run every time; ignoring the layer cache. Any commands following the invocation of RUN --no-cache
, will also ignore the cache. If --no-cache
is used as an option on the RUN
statement within a WITH DOCKER
statement, all commands after the WITH DOCKER
will also ignore the cache.
Auto-skip
Note that RUN --no-cache
commands may still be skipped by auto-skip. For more information see the Caching in Earthfiles guide.
--entrypoint
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
Might become the following in an Earthfile
--privileged
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:
--secret <env-var>=<secret-id> | <secret-id>
Makes available a secret, in the form of an env var (its name is defined by <env-var>
), to the command being executed. If you only specify <secret-id>
, the name of the env var will be <secret-id>
and its value the value of <secret-id>
.
Here is an example that showcases both syntaxes:
An empty string is also allowed for <secret-id>
, allowing for optional secrets, should it need to be disabled.
It is also possible to mount a secret as a file with RUN --mount type=secret,id=secret-id,target=/path/of/secret,chmod=0400
. See --mount
below.
For more information on how to use secrets see the build arguments and secrets guide. See also the Cloud secrets guide.
--ssh
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:
Note that RUN --ssh
option is only used for creating a tunnel to the host's ssh-agent's socket (set via $SSH_AUTH_SOCK
); it is not related to the git section of the earthly configuration file.
--mount <mount-spec>
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
type
The type of the mount. Currently only cache
, tmpfs
, and secret
are allowed.
type=cache
target
The target path for the mount.
target=/var/lib/data
mode
, chmod
The permission of the mounted file, in octal format (the same format the chmod unix command line expects).
chmod=0400
id
The cache ID for a global cache mount to be used across other targets or Earthfiles, when type=cache
. The secret ID for the contents of the target
file, when type=secret
. Use VERSION --global-cache 0.7
to enable.
id=my-shared-cache
, id=my-password
sharing
The sharing mode (locked
, shared
, private
) for the cache mount, only applicable for type=cache
.
sharing=shared
For cache mounts, the sharing mode can be one of the following:
locked
(default) - the cache mount is locked for the duration of the execution, other concurrent builds will wait for the lock to be released.shared
- the cache mount is shared between all concurrent builds.private
- if another concurrent build attempts to use the cache, a new (empty) cache will be created for the concurrent build.
Examples
Persisting cache for a single RUN
command, even when its dependencies change:
Note that mounts cannot be shared between targets, nor can they be shared within the same target, if the build-args differ between invocations.
Mounting a secret as a file:
The contents of the secret /root/.netrc
file can then be specified from the command line as:
or by passing the contents of an existing file from the host filesystem:
--interactive
/ --interactive-keep
Opens an interactive prompt during the target build. An interactive prompt must:
Be the last issued command in the target, with the exception of
SAVE IMAGE
commands. This also means that you cannotFROM
a target containing aRUN --interactive
.Be the only
--interactive
target within the run.Not be within a
LOCALLY
-designated target.
Examples:
Start an interactive python REPL:
Start bash
to tweak an image by hand. Changes made will be included:
COPY
Synopsis
COPY [options...] <src>... <dest>
(classical form)COPY [options...] <src-artifact>... <dest>
(artifact form)COPY [options...] (<src-artifact> --<build-arg-key>=<build-arg-value>...) <dest>
(artifact form with build args)
Description
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.
The COPY
command does not mark any saved images or artifacts of the referenced target for output, nor does it mark any push commands of the referenced target for pushing. For that, please use BUILD
.
Multiple COPY
commands issued one after the other will build the referenced targets in parallel, if the targets don't depend on each other. The resulting artifacts will then be copied sequentially in the order in which the COPY
commands were issued.
The classical form of the COPY
command differs from Dockerfiles in three cases:
URL sources are not yet supported.
Absolute paths are not supported - sources in the current directory cannot be referenced with a leading
/
The Earthly
COPY
is a classicalCOPY --link
. It uses layer merging for the copy operations.
Note
To prevent Earthly from copying unwanted files, you may specify file patterns to be excluded from the build context using an .earthlyignore
file. This file has the same syntax as a .dockerignore
file.
Options
--dir
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).
If the directories were copied without the use of --dir
, then their contents would be merged into the destination.
--<build-arg-key>=<build-arg-value>
Sets a value override of <build-arg-value>
for the build arg identified by <build-arg-key>
, when building the target containing the mentioned artifact. See also BUILD for more details about the build arg options.
Note that build args and the artifact references they apply to need to be surrounded by parenthesis:
--keep-ts
Instructs Earthly to not overwrite the file creation timestamps with a constant.
--keep-own
Instructs Earthly to keep file ownership information. This applies only to the artifact form and has no effect otherwise.
--chmod <octal-format>
Instructs Earthly to change the file permissions of the copied files. The <chmod>
needs to be in octal format, e.g. --chmod 0755
or --chmod 755
.
Note that you must include the flag in the corresponding SAVE ARTIFACT --keep-own ...
command, if using artifact form.
--if-exists
Only copy source if it exists; if it does not exist, earthly will simply ignore the COPY command and won't treat any missing sources as failures.
--symlink-no-follow
Allows copying a symbolic link from another target; it has no effect when copying files from the host. The option must be used in both the COPY
and SAVE ARTIFACT
commands; for example:
--from
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
... would be equivalent to final-target
in the following Earthfile
--platform <platform>
In artifact form, it specifies the platform to build the artifact on.
For more information see the multi-platform guide.
--allow-privileged
Same as FROM --allow-privileged
.
--build-arg <key>=<value>
(deprecated)
The option --build-arg
is deprecated. Use --<build-arg-key>=<build-arg-value>
instead.
Examples
Assuming the following directory tree, of a folder named test
:
Here is how the following copy commands will behave:
One can also copy from other Earthfile targets:
Parentheses are required when passing build-args:
For detailed examples demonstrating how other scenarios may function, please see our test suite.
ARG
Synopsis
ARG [--required] <name>[=<default-value>]
(constant form)ARG [--required] <name>=$(<default-value-expr>)
(dynamic form)
Description
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 recipe of the current target or command and only from the point it is declared onward. For more information regarding builtin args, see the builtin args page.
In its constant form, the arg takes a default value defined as a constant string. If the <default-value>
is not provided, then the default value is an empty string. In its dynamic form, the arg takes a default value defined as an expression. The expression is evaluated at run time and its result is used as the default value. The expression is interpreted via the default shell (/bin/sh -c
) within the build environment.
The value of an arg can be overridden either from the earthly
command
or from a command from another target, when implicitly or explicitly invoking the target containing the ARG
for example
For more information on how to use build args see the build arguments and secrets guide. A number of builtin args are available and are pre-filled by Earthly. For more information see builtin args.
Shadowing Variables
By default, ARG
scoping isn't intuitive. When an ARG
statement is parsed, earthly will look up any previous declaration of the ARG
and use the previous value, ignoring any new default. So for example:
would print bar
.
The experimental --arg-scope-and-set
feature flag changes this behavior. With VERSION --arg-scope-and-set 0.7
local ARG
s may shadow global ARG
s, and redeclaring an ARG
in the same scope will cause an error. This means that the above example would instead print bacon
.
Options
--required
A required ARG
must be provided at build time and can never have a default value. Required args can help eliminate cases where the user has unexpectedly set an ARG
to ""
.
--global
--global
A global ARG
is an arg that is made available to all targets in the Earthfile. This is useful for setting a default value for an arg that is used in many targets.
Global args may only be declared in base targets.
SAVE ARTIFACT
Synopsis
SAVE ARTIFACT [--keep-ts] [--keep-own] [--if-exists] [--force] <src> [<artifact-dest-path>] [AS LOCAL <local-path>]
Description
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. Note that local artifacts are only produced by targets that are run directly with earthly
, or when invoked using BUILD
.
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.
Hint
In order to inspect the contents of an artifacts environment, you can run
This command dumps the contents of the artifact environment of the target +<target>
into a local directory called output
, which can be inspected directly.
Important
Note that there is a distinction between a directory artifact and file artifact when it comes to local output. When saving an artifact locally, a directory artifact will replace the destination entirely, while a file (or set of files) artifact will be copied into the destination directory.
Important
As of VERSION 0.6
, local artifacts are only saved if they are connected to the initial target through a chain of BUILD
commands.
Options
--keep-ts
Instructs Earthly to not overwrite the file creation timestamps with a constant.
--keep-own
Instructs Earthly to keep file ownership information.
--if-exists
Only save artifacts if they exists; if not, earthly will simply ignore the SAVE ARTIFACT command and won't treat any missing sources as failures.
--symlink-no-follow
Save the symbolic link rather than the contents of the symbolically linked file. Note that the same flag must also be used in the corresponding COPY
command. For example:
--force
Force save operations which may be unsafe, such as writing to (or overwriting) a file or directory on the host filesystem located outside of the context of the directory containing the Earthfile.
Examples
Assuming the following directory tree, of a folder named test
:
Here is how the following SAVE ARTIFACT ... AS LOCAL
commands will behave:
For detailed examples demonstrating how other scenarios may function, please see our test suite.
SAVE IMAGE
Synopsis
SAVE IMAGE [--cache-from=<cache-image>] [--push] <image-name>...
(output form)SAVE IMAGE --cache-hint
(cache hint form)
Description
In the output form, the command SAVE IMAGE
marks the current build environment as the image of the target and assigns one or more output image names.
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 remote caching guide.
Assigning multiple image names
The SAVE IMAGE
command allows you to assign more than one image name:
Or
Important
As of VERSION 0.6
, the --referenced-save-only
feature flag is enabled by default. Images are only saved if they are connected to the initial target through a chain of BUILD
commands.
Options
--push
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
--cache-from=<cache-image>
Adds additional cache sources to be used when --use-inline-cache
is enabled. For more information see the remote caching guide.
--cache-hint
Instructs Earthly that the current target should be included as part of the explicit cache. For more information see the remote caching guide.
--no-manifest-list
Instructs Earthly to not create a manifest list for the image. This may be useful on platforms that do not support multi-platform images (for example, AWS Lambda), and the image produced needs to be of a different platform than the default one.
BUILD
Synopsis
BUILD [options...] <target-ref> [--<build-arg-name>=<build-arg-value>...]
Description
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. The invocation will mark any images, or artifacts saved by the referenced target for local output (assuming local output is enabled), and any push commands issued by the referenced target for pushing (assuming pushing is enabled).
Multiple BUILD
commands issued one after the other will be executed in parallel if the referenced targets don't depend on each other.
What is being output and pushed
In Earthly v0.6+, what is being output and pushed is determined either by the main target being invoked on the command-line directly, or by targets directly connected to it via a chain of BUILD
calls. Other ways to reference a target, such as FROM
, COPY
, WITH DOCKER --load
etc, do not contribute to the final set of outputs or pushes.
If you are referencing a target via some other command, such as COPY
and you would like for the outputs or pushes to be included, you can issue an equivalent BUILD
command in addition to the COPY
. For example
Should be amended with the following additional BUILD
call:
This, however, assumes that the target +my-target
is itself connected via a BUILD
chain to the main target being built. If that is not the case, additional BUILD
commands should be issued higher up the hierarchy.
Options
--<build-arg-key>=<build-arg-value>
Sets a value override of <build-arg-value>
for the build arg identified by <build-arg-key>
.
The override value of a build arg may be a constant string
or an expression involving other build args
or a dynamic expression, based on the output of a command executed in the context of the build environment.
Dynamic expressions are delimited by $(...)
.
--platform <platform>
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
For more information see the multi-platform guide.
--auto-skip
(coming soon)
Instructs Earthly to skip the build of the target if the target's dependencies have not changed from a previous successful build. For more information on how to use this feature, see the auto-skip section of the caching in Earthfiles guide.
--allow-privileged
Same as FROM --allow-privileged
.
--build-arg <build-arg-key>=<build-arg-value>
(deprecated)
This option is deprecated. Please use --<build-arg-key>=<build-arg-value>
instead.
VERSION
Synopsis
VERSION [options...] <version-number>
Description
The command VERSION
identifies which set of features to enable in Earthly while handling the corresponding Earthfile. The VERSION
command is mandatory starting with Earthly 0.7. The VERSION
command must be the first command in the Earthfile.
Options
Individual features may be enabled by setting the corresponding feature flag. New features start off as experimental, which is why they are disabled by default. Once a feature reaches maturity, it will be enabled by default under a new version number.
Please note that using individual feature flags directly does not guarantee the forwards-backwards compatibility of Earthfiles across versions. Using individual feature flags is experimental and is not recommended for production use.
All features are described in the version-specific features reference.
PROJECT
Synopsis
PROJECT <org-name>/<project-name>
Description
The command PROJECT
marks the current Earthfile as being part of the project belonging to the Earthly organization <org-name>
and the project <project-name>
. The project is used by Earthly to retrieve cloud-based secrets belonging to the project, as well as to associate any Earthly CI pipelines, and any build logs to the project when logged in.
The PROJECT
command can only be used in the base
recipe and it applies to the entire Earthfile. The PROJECT
command can never contain any ARG
s that need expanding.
GIT CLONE
Synopsis
GIT CLONE [--branch <git-ref>] [--keep-ts] <git-url> <dest-path>
Description
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.
Note that the repository is cloned via a shallow-clone opperation (i.e. a single-depth clone).
If you need to perform a full-depth clone of a repository, you can use the following pattern:
As of Earthly v0.7.21, git credentials are no longer stored in the .git/config
file; this includes the username. This means any ssh-based or https-based fetches or pushes will no longer work unless you restore the configured url, which can be done with:
See the "GIT CLONE vs RUN git clone" section under the best practices guide for more details.
Options
--branch <git-ref>
Points the HEAD
to the git reference specified by <git-ref>
. If this option is not specified, then the remote HEAD
is used instead.
--keep-ts
Instructs Earthly to not overwrite the file creation timestamps with a constant.
FROM DOCKERFILE
Synopsis
FROM DOCKERFILE [options...] <context-path>
Description
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. By default, 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
. Additionally, when using a <context-path>
from the host system, a .dockerignore
in the directory root will be used to exclude files (unless .earthlyignore
or .earthignore
are present). Use VERSION --use-docker-ignore 0.7
to enable.
Options
-f <dockerfile-path>
Specify an alternative Dockerfile to use. The <dockerfile-path>
can be either a path on the host system, relative to the current Earthfile, or an artifact reference pointing to a Dockerfile.
It is possible to split the Dockerfile
and the build context across two separate artifact references:
Note that +mybuildcontext/mydata
on its own would copy the directory and its contents; where as +mybuildcontext/mydata/*
is required to copy all of the contents from within the mydata
directory ( without copying the wrapping mydata
directory).
If both the Dockerfile
and build context are inside the same target, one must reference the same target twice, e.g. FROM DOCKERFILE -f +target/dir/Dockerfile +target/dir
.
--build-arg <key>=<value>
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.
--target <target-name>
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.
--platform <platform>
Specifies the platform to build on.
For more information see the multi-platform guide.
WITH DOCKER
Synopsis
Description
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.
If multiple targets are referenced via --load
, the images are built in parallel. Similarly, multiple images referenced with --pull
will be downloaded in parallel.
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:
For more examples, see the Docker in Earthly guide and the Integration testing guide.
For information on using WITH DOCKER
with podman see the Podman guide
Note
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-3.18-docker-23.0.6-r4
and earthly/dind:ubuntu-23.04-docker-24.0.5-1
to be used together with WITH DOCKER
.
Note
Note that the cleanup phase (after the RUN
command has finished), does not occur when using a LOCALLY
target, users should use RUN docker run --rm ...
to have docker remove the image after execution.
Options
--pull <image-name>
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.
Note
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.
--load <image-name>=<target-ref>
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
.
<target-ref>
may be a simple target reference (+some-target
), or a target reference with a build arg (+some-target --SOME_BUILD_ARG=value)
.
This option may be repeated in order to provide multiple images to be loaded.
The WITH DOCKER --load
option does not mark any saved images or artifacts of the referenced target for local output, nor does it mark any push commands of the referenced target for pushing. For that, please use BUILD
.
--compose <compose-file>
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.
--service <compose-service>
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.
--platform <platform>
Specifies the platform for any referenced --load
and --pull
images.
For more information see the multi-platform guide.
--allow-privileged
Same as FROM --allow-privileged
.
--build-arg <key>=<value>
(deprecated)
This option is deprecated. Please use --load <image-name>=(<target-ref> --<build-arg-key>=<build-arg-value>)
instead.
IF
Synopsis
Description
The IF
clause can perform varying commands depending on the outcome of one or more conditions. The expression passed as part of <condition>
is evaluated by running it in the build environment. If the exit code of the expression is zero, then the block of that condition is executed. Otherwise, the control continues to the next ELSE IF
condition (if any), or if no condition returns a non-zero exit code, the control continues to executing the <else-block>
, if one is provided.
A very common pattern is to use the POSIX shell [ ... ]
conditions. For example the following marks port 8080
as exposed if the file ./foo
exists.
Note
Performing a condition requires that a FROM
(or a from-like command, such as LOCALLY
) has been issued before the condition itself.
For example, the following is NOT a valid Earthfile.
The reason this is invalid is because the IF
condition is actually running the /usr/bin/[
executable to test if the condition is true or false, and therefore requires that a valid build environment has been initialized.
Here is how this might be fixed.
By initializing the build environment with FROM busybox
, the IF
condition can execute on top of the busybox
image.
Important
Changes to the filesystem in any of the conditions are not preserved. If a file is created as part of a condition, then that file will not be present in the build environment for any subsequent commands.
Options
--privileged
Same as RUN --privileged
.
--ssh
Same as RUN --ssh
.
--no-cache
Same as RUN --no-cache
.
--mount <mount-spec>
Same as RUN --mount <mount-spec>
.
--secret <env-var>=<secret-id>
Same as RUN --secret <env-var>=<secret-id>
.
FOR
Synopsis
Description
The FOR
clause can iterate over the items resulting from the expression <expression>
. On each iteration, the value of <variable-name>
is set to the current item in the iteration and the block of commands <for-block>
is executed in the context of that variable set as a build arg.
The expression may be either a constant list of items (e.g. foo bar buz
), or the output of a command (e.g. $(echo foo bar buz)
), or a parameterized list of items (e.g. foo $BARBUZ
). The result of the expression is then tokenized using the list of separators provided via the --sep
option. If unspecified, the separator list defaults to [tab]
, [new line]
and [space]
(\t\n
).
Important
Changes to the filesystem in expressions are not preserved. If a file is created as part of a FOR
expression, then that file will not be present in the build environment for any subsequent commands.
Examples
As an example, FOR
may be used to iterate over a list of files for compilation
As another example, FOR
may be used to iterate over a set of directories in a monorepo and invoking targets within them.
Options
--sep <separator-list>
The list of separators to use when tokenizing the output of the expression. If unspecified, the separator list defaults to [tab]
, [new line]
and [space]
(\t\n
).
--privileged
Same as RUN --privileged
.
--ssh
Same as RUN --ssh
.
--no-cache
Same as RUN --no-cache
.
--mount <mount-spec>
Same as RUN --mount <mount-spec>
.
--secret <env-var>=<secret-id>
Same as RUN --secret <env-var>=<secret-id>
.
WAIT
Synopsis
Description
The WAIT
clause executes the encapsulated commands and waits for them to complete. This includes pushing and outputting local artifacts -- a feature which can be used to control the order of interactions with the outside world.
Even though the WAIT
clause limits parallelism by forcing everything within it to finish executing before continuing, the commands within a WAIT
block execute in parallel.
Examples
As an example, a WAIT
block can be used to build and push to a remote registry (in parallel), then, after that execute a script which requires those images to exist in the remote registry:
One can also use a WAIT
block to control the order in which a SAVE ARTIFACT ... AS LOCAL
command is executed:
LET (experimental)
Note
The LET
command is currently incomplete and has experimental status. To use this feature, it must be enabled via VERSION --arg-scope-and-set 0.7
.
Synopsis
LET <name>=<value>
Description
The command LET
declares a variable with the name <name>
and with a value <value>
. This command works similarly to ARG
except that it cannot be overridden from the CLI.
LET
variables are allowed to shadow ARG
variables, which allows you to promote an ARG
to a local variable so that it may be used with SET
.
Example
SET (experimental)
Note
The SET
command is currently incomplete and has experimental status. To use this feature, it must be enabled via VERSION --arg-scope-and-set 0.7
.
Synopsis
SET <name>=<value>
Description
The command SET
may be used to change the value of a previously declared variable, so long as the variable was declared with LET
.
ARG
variables may not be changed by SET
, since ARG
is intended to accept overrides from the CLI. If you want to change the value of an ARG
variable, redeclare it with LET someVar = "$someVar"
first.
See the LET
docs for more info.
TRY (experimental)
Note
The TRY
command is currently incomplete and has experimental status. To use this feature, it must be enabled via VERSION --try 0.7
.
Synopsis
Description
The TRY
clause executes commands within the <try-block>
, while ensuring that the <finally-block>
is always executed, even if the <try-block>
fails.
This clause is still under active development. For now, only a single RUN
command is permitted within the <try-block>
, and only one or more SAVE ARTIFACT
commands are permitted in the <finally-block>
. The clause is thus useful for outputting coverage information in unit testing, outputting screenshots in UI integration tests, or outputting junit.xml
, or similar.
Example
CACHE
Synopsis
Description
The CACHE
command creates a cache mountpoint at <mountpoint>
in the build environment. The cache mountpoint is a directory which is shared between the instances of the same build target. The contents of the cache mountpoint are preserved between builds, and can be used to share data across builds.
At the end of the target, the contents of the cache mountpoint are persisted as an additional layer in the image. This means that the contents are available to subsequent targets in the same build using FROM
, or to any saved images SAVE IMAGE
.
Options
--sharing <sharing-mode>
The sharing mode for the cache mount, from one of the following:
locked
(default) - the cache mount is locked for the duration of the execution, other concurrent builds will wait for the lock to be released.shared
- the cache mount is shared between all concurrent builds.private
- if another concurrent build attempts to use the cache, a new (empty) cache will be created for the concurrent build.
--chmod <octal-format>
The permission of the mounted folder, in octal format (the same format the chmod unix command line expects). Default --chmod 0644
--id <cache-id>
The cache ID for a global cache volume to be used across other targets or Earthfiles. Use VERSION --global-cache 0.7
to enable.
LOCALLY
Synopsis
LOCALLY
Description
The LOCALLY
command can be used in place of a FROM
command, which will cause earthly to execute all commands under the target directly on the host system, rather than inside a container. Commands within a LOCALLY
target will never be cached. This feature should be used with caution as locally run commands have no guarantee they will behave the same on different systems.
LOCALLY
defined targets only support a subset of commands (along with a subset of their flags): RUN
, RUN --push
, SAVE ARTIFACT
, and COPY
.
RUN
commands have access to the environment variables which are exposed to the earthly
command; however, the commands are executed within a working directory which is set to the location of the referenced Earthfile and not where the earthly
command is run from.
For example, the following Earthfile will display the current user, hostname, and directory where the Earthfile is stored:
Note
In Earthly, outputting images and artifacts locally takes place only at the end of a successful build. In order to use such images or artifacts in LOCALLY
targets, they need to be referenced correctly.
For images, use the --load
option under WITH DOCKER
:
Do NOT use BUILD
for using images in LOCALLY
targets:
For artifacts, use COPY
, the same way you would in a regular target:
Do NOT use SAVE ARTIFACT ... AS LOCAL
and BUILD
for referencing artifacts in LOCALLY
targets:
COMMAND
Synopsis
COMMAND
Description
UDCs have been renamed to Functions
Functions used to be called UDCs (User Defined Commands). Earthly 0.7 still uses COMMAND
for declaring functions, but the keyword is deprecated and will be replaced by FUNCTION
in Earthly 0.8.
The command COMMAND
marks the beginning of a function definition. Functions are reusable sets of instructions that can be inserted in targets or other functions. In order to reference and execute a function, you may use the command DO
.
Unlike performing a BUILD +target
, functions inherit the build context and the build environment from the caller.
Functions create their own ARG
scope, which is distinct from the caller. Any ARG
that needs to be passed from the caller needs to be passed explicitly via DO +MY_FUNCTION --<build-arg-key>=<build-arg-value>
.
Global imports and global args are inherited from the base
target of the same Earthfile where the command is defined in (this may be distinct from the base
target of the caller).
For more information see the Functions Guide.
DO
Synopsis
DO [--allow-privileged] <function-ref> [--<build-arg-key>=<build-arg-value>...]
Description
The command DO
expands and executes the series of commands contained within a function referenced by <function-ref>
.
Unlike performing a BUILD +target
, functions inherit the build context and the build environment from the caller.
Functions create their own ARG
scope, which is distinct from the caller. Any ARG
that needs to be passed from the caller needs to be passed explicitly via DO +MY_FUNCTION --<build-arg-key>=<build-arg-value>
.
For more information see the Functions Guide.
Options
--allow-privileged
Same as FROM --allow-privileged
.
IMPORT
Synopsis
IMPORT [--allow-privileged] <earthfile-ref> [AS <alias>]
Description
The command IMPORT
aliases an Earthfile reference (<earthfile-ref>
) that can be used in subsequent target, artifact or command references.
If not provided, the <alias>
is inferred automatically as the last element of the path provided in <earthfile-ref>
. For example, if <earthfile-ref>
is github.com/foo/bar/buz:v1.2.3
, then the alias is inferred as buz
.
The <earthfile-ref>
can be a reference to any directory other than .
. If the reference ends in ..
, then mentioning AS <alias>
is mandatory.
If an IMPORT
is defined in the base
target of the Earthfile, then it becomes a global IMPORT
and it is made available to every other target or command in that file, regardless of their base images used.
For more information see the importing guide.
Options
--allow-privileged
Similar to FROM --allow-privileged
, extend the ability to request privileged capabilities to all invocations of the imported alias.
CMD (same as Dockerfile CMD)
Synopsis
CMD ["executable", "arg1", "arg2"]
(exec form)CMD ["arg1, "arg2"]
(as default arguments to the entrypoint)CMD command arg1 arg2
(shell form)
Description
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 (same as Dockerfile LABEL)
Synopsis
LABEL <key>=<value> <key>=<value> ...
Description
The LABEL
command adds label metadata to an image. It works the same way as the Dockerfile LABEL
command.
EXPOSE (same as Dockerfile EXPOSE)
Synopsis
EXPOSE <port> <port> ...
EXPOSE <port>/<protocol> <port>/<protocol> ...
Description
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 (same as Dockerfile ENV)
Synopsis
ENV <key> <value>
ENV <key>=<value>
Description
The ENV
command sets the environment variable <key>
to the value <value>
. It works the same way as the Dockerfile ENV
command.
Note
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 (same as Dockerfile ENTRYPOINT)
Synopsis
ENTRYPOINT ["executable", "arg1", "arg2"]
(exec form)ENTRYPOINT command arg1 arg2
(shell form)
Description
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 (same as Dockerfile VOLUME)
Synopsis
VOLUME <path-to-target-mount> <path-to-target-mount> ...
VOLUME ["<path-to-target-mount>", <path-to-target-mount> ...]
Description
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 (same as Dockerfile USER)
Synopsis
USER <user>[:<group>]
USER <UID>[:<GID>]
Description
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 (same as Dockerfile WORKDIR)
Synopsis
WORKDIR <path-to-dir>
Description
The WORKDIR
command sets the working directory for following commands 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 (same as Dockerfile HEALTHCHECK)
Synopsis
HEALTHCHECK NONE
(disable health checking)HEALTHCHECK [--interval=DURATION] [--timeout=DURATION] [--start-period=DURATION] [--retries=N] [--start-interval=DURATION] CMD command arg1 arg2
(check container health by running command inside the container)
Description
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.
Options
--interval=DURATION
Sets the time interval between health checks. Defaults to 30s
.
--timeout=DURATION
Sets the timeout for a single run before it is considered as failed. Defaults to 30s
.
--start-period=DURATION
Sets an initialization time period in which failures are not counted towards the maximum number of retries. Defaults to 0s
.
--retries=N
Sets the number of retries before a container is considered unhealthy
. Defaults to 3
.
--start-interval=DURATION
Sets the time interval between health checks during the start period. Defaults to 5s
.
HOST
Synopsis
HOST <hostname> <ip>
Description
The HOST
command creates a hostname entry (under /etc/hosts
) that causes <hostname>
to resolve to the specified <ip>
address.
PIPELINE (deprecated)
Note
The PIPELINE
command is in beta status and is only useful for Earthly CI.
Synopsis
PIPELINE [--push]
Description
The PIPELINE
command is used to declare that the current target is an Earthly CI pipeline. The PIPELINE
command must be the first command in the target.
To use a PIPELINE
, you must also declare a PROJECT
in the Earthfile. This PROJECT
must match the name of the CI's project that references the git repository
A pipeline is a target that is executed by Earthly CI when a certain trigger is activated. Triggers can be declared via the TRIGGER
command. Pipeline targets allow only the commands TRIGGER
, ARG
and BUILD
. Other commands may be used indirectly in other targets that can be then referenced by BUILD
.
Pipeline targets are always executed with no outputs, in strict mode.
Note
Pipelines and their definitions, including their triggers must be merged into the primary branch (which, unless overridden, is the default branch on GitHub) in order for the triggers to take effect.
Example
The following example shows a simple pipeline called my-pipeline
, which is triggered on either a push to the main
branch, or a pull request against the main
branch. The pipeline executes the target my-build
, which simply prints Hello world
.
Options
--push
Indicates that the targets referenced by this pipeline will be called in push-mode. SAVE IMAGE --push
commands will trigger pushes to the remote registry, and RUN --push
commands will execute.
TRIGGER (deprecated)
Note
The TRIGGER
command is in beta status and is only useful for Earthly CI.
Synopsis
TRIGGER manual
(manual form)TRIGGER pr <pr-branch>
(PR form)TRIGGER push <push-branch>
(push form)
Description
The TRIGGER
command is only allowed in the context of a pipeline target (declared via PIPELINE
), and is used to configure the way in which the pipeline is triggered. Multiple triggers are allowed for a single pipeline.
In the manual form, the pipeline is triggered manually via the Earthly CI UI or via earthly
on the command-line.
In the PR form, the pipeline is triggered when a pull request is opened against the branch <pr-branch>
.
In the push form, the pipeline is triggered when a push is made to the branch <push-branch>
.
Note
Pipelines and their definitions, including their triggers must be merged into the primary branch (which, unless overridden, is the default branch on GitHub) in order for the triggers to take effect.
SHELL (not supported)
The classical SHELL
Dockerfile command is not yet supported. Use the exec form of RUN
, ENTRYPOINT
and CMD
instead and prepend a different shell.
ADD (not supported)
The classical ADD
Dockerfile command is not yet supported. Use COPY instead.
ONBUILD (not supported)
The classical ONBUILD
Dockerfile command is not supported.
STOPSIGNAL (not supported)
The classical STOPSIGNAL
Dockerfile command is not yet supported.
Last updated