nixpkgs/doc/builders/images/dockertools.section.md
Christian Kemper f6ae4479ea dockerTools: allowing architecture to be specified
... for buildImage, buildLayeredImage and streamLayeredImage,
adding docs and tests.
2022-12-08 20:29:09 +01:00

21 KiB

pkgs.dockerTools

pkgs.dockerTools is a set of functions for creating and manipulating Docker images according to the Docker Image Specification v1.2.0. Docker itself is not used to perform any of the operations done by these functions.

buildImage

This function is analogous to the docker build command, in that it can be used to build a Docker-compatible repository tarball containing a single image with one or multiple layers. As such, the result is suitable for being loaded in Docker with docker load.

The parameters of buildImage with relative example values are described below:

[]{#ex-dockerTools-buildImage} []{#ex-dockerTools-buildImage-runAsRoot}

buildImage {
  name = "redis";
  tag = "latest";

  fromImage = someBaseImage;
  fromImageName = null;
  fromImageTag = "latest";

  copyToRoot = pkgs.buildEnv {
    name = "image-root";
    paths = [ pkgs.redis ];
    pathsToLink = [ "/bin" ];
  };

  runAsRoot = ''
    #!${pkgs.runtimeShell}
    mkdir -p /data
  '';

  config = {
    Cmd = [ "/bin/redis-server" ];
    WorkingDir = "/data";
    Volumes = { "/data" = { }; };
  };

  diskSize = 1024;
  buildVMMemorySize = 512;
}

The above example will build a Docker image redis/latest from the given base image. Loading and running this image in Docker results in redis-server being started automatically.

  • name specifies the name of the resulting image. This is the only required argument for buildImage.

  • tag specifies the tag of the resulting image. By default it's null, which indicates that the nix output hash will be used as tag.

  • fromImage is the repository tarball containing the base image. It must be a valid Docker image, such as exported by docker save. By default it's null, which can be seen as equivalent to FROM scratch of a Dockerfile.

  • fromImageName can be used to further specify the base image within the repository, in case it contains multiple images. By default it's null, in which case buildImage will peek the first image available in the repository.

  • fromImageTag can be used to further specify the tag of the base image within the repository, in case an image contains multiple tags. By default it's null, in which case buildImage will peek the first tag available for the base image.

  • copyToRoot is a derivation that will be copied in the new layer of the resulting image. This can be similarly seen as ADD contents/ / in a Dockerfile. By default it's null.

  • runAsRoot is a bash script that will run as root in an environment that overlays the existing layers of the base image with the new resulting layer, including the previously copied contents derivation. This can be similarly seen as RUN ... in a Dockerfile.

NOTE: Using this parameter requires the kvm device to be available.

  • config is used to specify the configuration of the containers that will be started off the built image in Docker. The available options are listed in the Docker Image Specification v1.2.0.

  • architecture is optional and used to specify the image architecture, this is useful for multi-architecture builds that don't need cross compiling. If not specified it will default to hostPlatform.

  • diskSize is used to specify the disk size of the VM used to build the image in megabytes. By default it's 1024 MiB.

  • buildVMMemorySize is used to specify the memory size of the VM to build the image in megabytes. By default it's 512 MiB.

After the new layer has been created, its closure (to which contents, config and runAsRoot contribute) will be copied in the layer itself. Only new dependencies that are not already in the existing layers will be copied.

At the end of the process, only one new single layer will be produced and added to the resulting image.

The resulting repository will only list the single image image/tag. In the case of the buildImage example, it would be redis/latest.

It is possible to inspect the arguments with which an image was built using its buildArgs attribute.

NOTE: If you see errors similar to getProtocolByName: does not exist (no such protocol name: tcp) you may need to add pkgs.iana-etc to contents.

NOTE: If you see errors similar to Error_Protocol ("certificate has unknown CA",True,UnknownCa) you may need to add pkgs.cacert to contents.

By default buildImage will use a static date of one second past the UNIX Epoch. This allows buildImage to produce binary reproducible images. When listing images with docker images, the newly created images will be listed like this:

$ docker images
REPOSITORY   TAG      IMAGE ID       CREATED        SIZE
hello        latest   08c791c7846e   48 years ago   25.2MB

You can break binary reproducibility but have a sorted, meaningful CREATED column by setting created to now.

pkgs.dockerTools.buildImage {
  name = "hello";
  tag = "latest";
  created = "now";
  copyToRoot = pkgs.buildEnv {
    name = "image-root";
    paths = [ pkgs.hello ];
    pathsToLink = [ "/bin" ];
  };

  config.Cmd = [ "/bin/hello" ];
}

Now the Docker CLI will display a reasonable date and sort the images as expected:

$ docker images
REPOSITORY   TAG      IMAGE ID       CREATED              SIZE
hello        latest   de2bf4786de6   About a minute ago   25.2MB

However, the produced images will not be binary reproducible.

buildLayeredImage

Create a Docker image with many of the store paths being on their own layer to improve sharing between images. The image is realized into the Nix store as a gzipped tarball. Depending on the intended usage, many users might prefer to use streamLayeredImage instead, which this function uses internally.

name

The name of the resulting image.

tag optional

Tag of the generated image.

Default: the output path's hash

fromImage optional

The repository tarball containing the base image. It must be a valid Docker image, such as one exported by docker save.

Default: null, which can be seen as equivalent to FROM scratch of a Dockerfile.

contents optional

Top-level paths in the container. Either a single derivation, or a list of derivations.

Default: []

config optional

architecture is optional and used to specify the image architecture, this is useful for multi-architecture builds that don't need cross compiling. If not specified it will default to hostPlatform.

Run-time configuration of the container. A full list of the options are available at in the Docker Image Specification v1.2.0.

Default: {}

created optional

Date and time the layers were created. Follows the same now exception supported by buildImage.

Default: 1970-01-01T00:00:01Z

maxLayers optional

Maximum number of layers to create.

Default: 100

Maximum: 125

extraCommands optional

Shell commands to run while building the final layer, without access to most of the layer contents. Changes to this layer are "on top" of all the other layers, so can create additional directories and files.

fakeRootCommands optional

Shell commands to run while creating the archive for the final layer in a fakeroot environment. Unlike extraCommands, you can run chown to change the owners of the files in the archive, changing fakeroot's state instead of the real filesystem. The latter would require privileges that the build user does not have. Static binaries do not interact with the fakeroot environment. By default all files in the archive will be owned by root.

enableFakechroot optional

Whether to run in fakeRootCommands in fakechroot, making programs behave as though / is the root of the image being created, while files in the Nix store are available as usual. This allows scripts that perform installation in / to work as expected. Considering that fakechroot is implemented via the same mechanism as fakeroot, the same caveats apply.

Default: false

Behavior of contents in the final image

Each path directly listed in contents will have a symlink in the root of the image.

For example:

pkgs.dockerTools.buildLayeredImage {
  name = "hello";
  contents = [ pkgs.hello ];
}

will create symlinks for all the paths in the hello package:

/bin/hello -> /nix/store/h1zb1padqbbb7jicsvkmrym3r6snphxg-hello-2.10/bin/hello
/share/info/hello.info -> /nix/store/h1zb1padqbbb7jicsvkmrym3r6snphxg-hello-2.10/share/info/hello.info
/share/locale/bg/LC_MESSAGES/hello.mo -> /nix/store/h1zb1padqbbb7jicsvkmrym3r6snphxg-hello-2.10/share/locale/bg/LC_MESSAGES/hello.mo

Automatic inclusion of config references

The closure of config is automatically included in the closure of the final image.

This allows you to make very simple Docker images with very little code. This container will start up and run hello:

pkgs.dockerTools.buildLayeredImage {
  name = "hello";
  config.Cmd = [ "${pkgs.hello}/bin/hello" ];
}

Adjusting maxLayers

Increasing the maxLayers increases the number of layers which have a chance to be shared between different images.

Modern Docker installations support up to 128 layers, but older versions support as few as 42.

If the produced image will not be extended by other Docker builds, it is safe to set maxLayers to 128. However, it will be impossible to extend the image further.

The first (maxLayers-2) most "popular" paths will have their own individual layers, then layer #maxLayers-1 will contain all the remaining "unpopular" paths, and finally layer #maxLayers will contain the Image configuration.

Docker's Layers are not inherently ordered, they are content-addressable and are not explicitly layered until they are composed in to an Image.

streamLayeredImage

Builds a script which, when run, will stream an uncompressed tarball of a Docker image to stdout. The arguments to this function are as for buildLayeredImage. This method of constructing an image does not realize the image into the Nix store, so it saves on IO and disk/cache space, particularly with large images.

The image produced by running the output script can be piped directly into docker load, to load it into the local docker daemon:

$(nix-build) | docker load

Alternatively, the image be piped via gzip into skopeo, e.g., to copy it into a registry:

$(nix-build) | gzip --fast | skopeo copy docker-archive:/dev/stdin docker://some_docker_registry/myimage:tag

pullImage

This function is analogous to the docker pull command, in that it can be used to pull a Docker image from a Docker registry. By default Docker Hub is used to pull images.

Its parameters are described in the example below:

pullImage {
  imageName = "nixos/nix";
  imageDigest =
    "sha256:473a2b527958665554806aea24d0131bacec46d23af09fef4598eeab331850fa";
  finalImageName = "nix";
  finalImageTag = "2.11.1";
  sha256 = "sha256-qvhj+Hlmviz+KEBVmsyPIzTB3QlVAFzwAY1zDPIBGxc=";
  os = "linux";
  arch = "x86_64";
}
  • imageName specifies the name of the image to be downloaded, which can also include the registry namespace (e.g. nixos). This argument is required.

  • imageDigest specifies the digest of the image to be downloaded. This argument is required.

  • finalImageName, if specified, this is the name of the image to be created. Note it is never used to fetch the image since we prefer to rely on the immutable digest ID. By default it's equal to imageName.

  • finalImageTag, if specified, this is the tag of the image to be created. Note it is never used to fetch the image since we prefer to rely on the immutable digest ID. By default it's latest.

  • sha256 is the checksum of the whole fetched image. This argument is required.

  • os, if specified, is the operating system of the fetched image. By default it's linux.

  • arch, if specified, is the cpu architecture of the fetched image. By default it's x86_64.

nix-prefetch-docker command can be used to get required image parameters:

$ nix run nixpkgs.nix-prefetch-docker -c nix-prefetch-docker --image-name mysql --image-tag 5

Since a given imageName may transparently refer to a manifest list of images which support multiple architectures and/or operating systems, you can supply the --os and --arch arguments to specify exactly which image you want. By default it will match the OS and architecture of the host the command is run on.

$ nix-prefetch-docker --image-name mysql --image-tag 5 --arch x86_64 --os linux

Desired image name and tag can be set using --final-image-name and --final-image-tag arguments:

$ nix-prefetch-docker --image-name mysql --image-tag 5 --final-image-name eu.gcr.io/my-project/mysql --final-image-tag prod

exportImage

This function is analogous to the docker export command, in that it can be used to flatten a Docker image that contains multiple layers. It is in fact the result of the merge of all the layers of the image. As such, the result is suitable for being imported in Docker with docker import.

NOTE: Using this function requires the kvm device to be available.

The parameters of exportImage are the following:

exportImage {
  fromImage = someLayeredImage;
  fromImageName = null;
  fromImageTag = null;

  name = someLayeredImage.name;
}

The parameters relative to the base image have the same synopsis as described in buildImage, except that fromImage is the only required argument in this case.

The name argument is the name of the derivation output, which defaults to fromImage.name.

Environment Helpers

Some packages expect certain files to be available globally. When building an image from scratch (i.e. without fromImage), these files are missing. pkgs.dockerTools provides some helpers to set up an environment with the necessary files. You can include them in copyToRoot like this:

buildImage {
  name = "environment-example";
  copyToRoot = with pkgs.dockerTools; [
    usrBinEnv
    binSh
    caCertificates
    fakeNss
  ];
}

usrBinEnv

This provides the env utility at /usr/bin/env.

binSh

This provides bashInteractive at /bin/sh.

caCertificates

This sets up /etc/ssl/certs/ca-certificates.crt.

fakeNss

Provides /etc/passwd and /etc/group that contain root and nobody. Useful when packaging binaries that insist on using nss to look up username/groups (like nginx).

shadowSetup

This constant string is a helper for setting up the base files for managing users and groups, only if such files don't exist already. It is suitable for being used in a buildImage runAsRoot script for cases like in the example below:

buildImage {
  name = "shadow-basic";

  runAsRoot = ''
    #!${pkgs.runtimeShell}
    ${pkgs.dockerTools.shadowSetup}
    groupadd -r redis
    useradd -r -g redis redis
    mkdir /data
    chown redis:redis /data
  '';
}

Creating base files like /etc/passwd or /etc/login.defs is necessary for shadow-utils to manipulate users and groups.

fakeNss

If your primary goal is providing a basic skeleton for user lookups to work, and/or a lesser privileged user, adding pkgs.fakeNss to the container image root might be the better choice than a custom script running useradd and friends.

It provides a /etc/passwd and /etc/group, containing root and nobody users and groups.

It also provides a /etc/nsswitch.conf, configuring NSS host resolution to first check /etc/hosts, before checking DNS, as the default in the absence of a config file (dns [!UNAVAIL=return] files) is quite unexpected.

You can pair it with binSh, which provides bin/sh as a symlink to bashInteractive (as /bin/sh is configured as a shell).

buildImage {
  name = "shadow-basic";

  copyToRoot = pkgs.buildEnv {
    name = "image-root";
    paths = [ binSh pkgs.fakeNss ];
    pathsToLink = [ "/bin" "/etc" "/var" ];
  };
}

buildNixShellImage

Create a Docker image that sets up an environment similar to that of running nix-shell on a derivation. When run in Docker, this environment somewhat resembles the Nix sandbox typically used by nix-build, with a major difference being that access to the internet is allowed. It additionally also behaves like an interactive nix-shell, running things like shellHook and setting an interactive prompt. If the derivation is fully buildable (i.e. nix-build can be used on it), running buildDerivation inside such a Docker image will build the derivation, with all its outputs being available in the correct /nix/store paths, pointed to by the respective environment variables like $out, etc.

::: {.warning} The behavior doesn't match nix-shell or nix-build exactly and this function is known not to work correctly for e.g. fixed-output derivations, content-addressed derivations, impure derivations and other special types of derivations. :::

Arguments

drv

The derivation on which to base the Docker image.

Adding packages to the Docker image is possible by e.g. extending the list of nativeBuildInputs of this derivation like

buildNixShellImage {
  drv = someDrv.overrideAttrs (old: {
    nativeBuildInputs = old.nativeBuildInputs or [] ++ [
      somethingExtra
    ];
  });
  # ...
}

Similarly, you can extend the image initialization script by extending shellHook

name optional

The name of the resulting image.

Default: drv.name + "-env"

tag optional

Tag of the generated image.

Default: the resulting image derivation output path's hash

uid/gid optional

The user/group ID to run the container as. This is like a nixbld build user.

Default: 1000/1000

homeDirectory optional

The home directory of the user the container is running as

Default: /build

shell optional

The path to the bash binary to use as the shell. This shell is started when running the image.

Default: pkgs.bashInteractive + "/bin/bash"

command optional

Run this command in the environment of the derivation, in an interactive shell. See the --command option in the nix-shell documentation.

Default: (none)

run optional

Same as command, but runs the command in a non-interactive shell instead. See the --run option in the nix-shell documentation.

Default: (none)

Example

The following shows how to build the pkgs.hello package inside a Docker container built with buildNixShellImage.

with import <nixpkgs> {};
dockerTools.buildNixShellImage {
  drv = hello;
}

Build the derivation:

nix-build hello.nix
these 8 derivations will be built:
  /nix/store/xmw3a5ln29rdalavcxk1w3m4zb2n7kk6-nix-shell-rc.drv
...
Creating layer 56 from paths: ['/nix/store/crpnj8ssz0va2q0p5ibv9i6k6n52gcya-stdenv-linux']
Creating layer 57 with customisation...
Adding manifests...
Done.
/nix/store/cpyn1lc897ghx0rhr2xy49jvyn52bazv-hello-2.12-env.tar.gz

Load the image:

docker load -i result
0d9f4c4cd109: Loading layer [==================================================>]   2.56MB/2.56MB
...
ab1d897c0697: Loading layer [==================================================>]  10.24kB/10.24kB
Loaded image: hello-2.12-env:pgj9h98nal555415faa43vsydg161bdz

Run the container:

docker run -it hello-2.12-env:pgj9h98nal555415faa43vsydg161bdz
[nix-shell:/build]$

In the running container, run the build:

buildDerivation
unpacking sources
unpacking source archive /nix/store/8nqv6kshb3vs5q5bs2k600xpj5bkavkc-hello-2.12.tar.gz
...
patching script interpreter paths in /nix/store/z5wwy5nagzy15gag42vv61c2agdpz2f2-hello-2.12
checking for references to /build/ in /nix/store/z5wwy5nagzy15gag42vv61c2agdpz2f2-hello-2.12...

Check the build result:

$out/bin/hello
Hello, world!