Easier ways to download bottlerocket image for bare metal

[StefanTheWiz]

Hi.

According to this guide on provisioning for bare metal, to download the OS image, I'd have to install rust then tuftool and a bunch of other stuff to get these working.

Is there a reason why this involves so many steps?
Could it be simplified down to:

wget "https://updates.bottlerocket.aws/targets/bottlerocket-${VARIANT}-${ARCH}-${VERSION}.img.lz4"

For context: I'm trying to automate some of the steps in the bare metal guide.

[stockholmux]

@StefanTheWiz When you say "I'd have to install rust then tuftool and a bunch of other stuff" - what is the "other stuff" you're referring to?

[StefanTheWiz]

apt install -y --no-install-recommends \
    ca-certificates \
    build-essential \
    libssl-dev \
    curl \
    pkg-config

[StefanTheWiz]

I'm using hashicorp packer to automate building the OS image.
I'm starting with an ubuntu VM in rescue mode so I can write the final image over the block device that will snapshotted.
However, because the VM is in rescue mode, it has limited storage and I start running out of space when installing tuftool.

If the download process was as simple as a link or a bash script, this would reduce my build time down to under 2 mins and it would have saved me about a day's work trying to get this stuff to build and install.

[zmrow]

@StefanTheWiz Could you elaborate a bit more on what you're trying to do? It sounds like you're trying to build an image, rather than download it?

Building Bottlerocket requires the tools you mentioned (build-essential, curl, etc), but tuftool should not.

[StefanTheWiz]

I think these came up as part of installing rust if I remember correctly

[zmrow]

@StefanTheWiz would it be sufficient to provide binary releases of tuftool on Github? That would avoid needing Rust or the tools you mentioned.

[StefanTheWiz]

yes, I think so.

[tommie]

I agree that tuftool is a really strange requirement for "just downloading" the image. It's very complicated.

I made this script for using wget and jq: https://gist.github.com/tommie/f4528a7da1af9d8228c385e175462faa

I'd probably also be fine with a tuftool binary; something I can easily wget to then fetch the image.

[tommie]

I should also add: why do I have to pass three URLs/paths to tuftool? Can't the metadata/targets URL be stored in root.json?

[cbgbt]

I've created awslabs/tough#662 to track creating binary releases of tuftoolon GitHub.

[stockholmux]

FYI - I added a Dockerfile for tuftool which simplifies getting the dependencies and building the tool.