If you'd like to build your own image instead of relying on an Amazon-provided image, follow these steps. You can skip to the setup guide for Kubernetes or the setup guide for Amazon ECS to use an existing image in Amazon EC2. (We're still working on other use cases!)
The build process artifacts and resulting images can consume in excess of 80GB in the local directory.
The build process is also fairly demanding on your CPU, since we build all included software from scratch. (The first time. Package builds are cached, and only changes are built afterward.) The build scales well to 32+ cores. The first time you build, the fastest machines can take about 12 minutes while slower machines with only a couple cores can take 3-4 hours.
The build system requires certain operating system packages to be installed.
Ensure the following OS packages are installed:
apt install build-essential openssl libssl-dev pkg-config liblz4-tool
yum install make automake gcc openssl openssl-devel pkg-config lz4 perl-FindBin perl-lib
The build system is based on the Rust language. We recommend you install the latest stable Rust using rustup, either from the official site or your development host's package manager. Rust 1.51.0 or higher is required.
To organize build tasks, we use cargo-make. To get it, run:
cargo install cargo-make
Bottlerocket uses Docker to orchestrate package and image builds.
We recommend Docker 20.10.10 or later.
Builds rely on Docker's integrated BuildKit support, which has received many fixes and improvements in newer versions.
The default seccomp policy of older versions of Docker do not support the clone3
syscall in recent versions of Fedora or Ubuntu, on which the Bottlerocket SDK is based.
You'll need to have Docker installed and running, with your user account added to the docker
group.
Docker's post-installation steps for Linux will walk you through that.
Note: If you're on a newer Linux distribution using the unified cgroup hierarchy with cgroups v2, you may need to disable it to work with current versions of runc. You'll know this is the case if you see an error like
docker: Error response from daemon: OCI runtime create failed: this version of runc doesn't work on cgroups v2: unknown.
Set the kernel parametersystemd.unified_cgroup_hierarchy=0
in your boot configuration (e.g. GRUB) and reboot.
To build an image, run:
cargo make
This will build an image for the default variant (a recent aws-k8s-*
, see the BUILDSYS_VARIANT
variable in Makefile.toml to find the current default variant).
All packages will be built in turn, and then compiled into an img
file in the build/images/
directory.
The version number in Release.toml will be used in naming the file, and will be used inside the image as the release version. If you're planning on publishing your build, you may want to change the version.
To build an image for a different variant, run:
cargo make -e BUILDSYS_VARIANT=my-variant-here
To build an image for a different architecture, run:
cargo make -e BUILDSYS_ARCH=my-arch-here
If you want to limit the build concurrency, set BUILDSYS_JOBS
(the default is 8
):
cargo make -e BUILDSYS_JOBS=4
(You can use variant and arch arguments together, too.)
Most packages will include license files extracted from upstream source archives.
However, in some rare cases there are multiple licenses that could apply to a package.
Bottlerocket's build system uses the Licenses.toml
file in conjunction with the licenses
directory to configure the licenses used for such special packages.
Here is an example of a simple Licenses.toml
configuration file:
[package]
spdx-id = "SPDX-ID"
licenses = [
{ path = "the-license.txt" }
]
In the previous example, it is expected that the file the-license.txt
is present in licenses
.
You can retrieve the licenses from a remote endpoint, or the local filesystem if you specify the license-url
field:
[package]
spdx-id = "SPDX-ID AND SPDX-ID-2" # Package with multiple licenses
licenses = [
# This file is copied from a file system, and will be saved as `path`
{ license-url = "file:///path/to/spdx-id-license.txt", path = "spdx-id-license.txt" },
# This file is fetched from an https endpoint, and will be saved as `path`
{ license-url = "https://localhost/spdx-id-license-v2.txt", path = "spdx-id-license-2.txt" }
]
If you want to build any of the NVIDIA variants, you can follow these steps to prepare a Licenses.toml
file using the License for customer use of NVIDIA software:
- Create a
Licenses.toml
file in your Bottlerocket root directory, with the following content:
[nvidia]
spdx-id = "LicensesRef-NVIDIA-Customer-Use"
licenses = [
{ path = "LICENSE", license-url = "https://www.nvidia.com/en-us/drivers/nvidia-license/" }
]
- Fetch the licenses with this command:
cargo make -e BUILDSYS_UPSTREAM_LICENSE_FETCH=true fetch-licenses
- Build your image, setting the
BUILDSYS_UPSTREAM_SOURCE_FALLBACK
flag totrue
, if you haven't cached the driver's sources:
K8S_VERSION=1.24
cargo make \
-e BUILDSYS_VARIANT=aws-k8s-${K8S_VERSION}-nvidia \
-e BUILDSYS_UPSTREAM_SOURCE_FALLBACK="true"
To use the image in Amazon EC2, we need to register the image as an AMI.
To do this, you'll need to have your AWS account credentials setup on your system.
There are lots of ways to do this; one method is using the aws
CLI via its configure
command with your user's access and secret keys.
If you're using an EC2 instance, the EC2 instance's IAM role will be used automatically if available.
For a simple start, pick an EC2 region, then run:
cargo make -e PUBLISH_REGIONS=your-region-here ami
Note that the task ("ami") must come after the arguments to cargo make
that are specified with -e
.
Your new AMI ID will be printed after it's registered.
If you built your image for a different architecture or variant, just use the same arguments here:
cargo make -e PUBLISH_REGIONS=your-region-here -e BUILDSYS_VARIANT=my-variant-here ami
(There's a lot more detail on building and managing AMIs in the PUBLISHING guide.)
See any of the setup guides tailored to the various execution environments for information on running Bottlerocket images:
- Setup guide for Kubernetes
- Setup guide for Amazon ECS
- Setup guide for VMware
- Setup guide for QEMU/KVM
See the PUBLISHING guide for information on deploying Bottlerocket images and repositories.
To further extend Bottlerocket, you may want to build extra kernel modules. The specifics of building an out-of-tree module will vary by project, but the first step is to download the "kmod kit" that contains the kernel headers and toolchain you'll need to use.
kmod kits are included in the official Bottlerocket repos starting with Bottlerocket v1.0.6. Let's say you want to download the kit for building x86_64 modules for v1.11.0 and variant aws-k8s-1.24.
First, you need tuftool:
cargo install tuftool
Next, you need the Bottlerocket root role, which is used by tuftool to verify the kmod kit. This will download and verify the root role itself:
curl -O "https://cache.bottlerocket.aws/root.json"
sha512sum -c <<<"b81af4d8eb86743539fbc4709d33ada7b118d9f929f0c2f6c04e1d41f46241ed80423666d169079d736ab79965b4dd25a5a6db5f01578b397496d49ce11a3aa2 root.json"
Next, set your desired parameters, and download the kmod kit:
ARCH=x86_64
VERSION=v1.11.0
VARIANT=aws-k8s-1.24
OUTDIR="${VARIANT}-${VERSION}"
tuftool download "${OUTDIR}" --target-name ${VARIANT}-${ARCH}-kmod-kit-${VERSION}.tar.xz \
--root ./root.json \
--metadata-url "https://updates.bottlerocket.aws/2020-07-07/${VARIANT}/${ARCH}/" \
--targets-url "https://updates.bottlerocket.aws/targets/"
To use the kmod kit, extract it, and update your PATH to use its toolchain:
tar xf "${VARIANT}-${ARCH}-kmod-kit-${VERSION}.tar.xz"
export CROSS_COMPILE="${ARCH}-bottlerocket-linux-musl-"
export KERNELDIR="${PWD}/${VARIANT}-${ARCH}-kmod-kit-${VERSION}/kernel-devel"
export PATH="${PWD}/${VARIANT}-${ARCH}-kmod-kit-${VERSION}/toolchain/usr/bin:${PATH}"
Now you can compile modules against the kernel headers in ${KERNELDIR}
.