Branch	Status
master

epoxy-images

Support for building Linux kernels, rootfs images, and ROMs for ePoxy

An ePoxy managed system depends on several image types:

• generic Linux images that provide a minimal network boot environment.
• stage1 images that embed node network configuration and are either flashed to NICs, or burned to CDs.
• stage3 Linux ROM update images, that (re)flash iPXE ROMs to NICs.

Build Automation

The epoxy-images repo is connected to Google Cloud Build.

• mlab-sandbox - push to a branch matching sandbox-* builds cloudbuild.yaml & cloudbuild-stage1.yaml.
• mlab-staging - push to master builds both cloudbuild.yaml and cloudbuild-stage1.yaml
• mlab-oti - tags matching v[0-9]+.[0-9]+.[0-9]+ builds cloudbuild.yaml & cloudbuild-stage1.yaml

Building images

See cloudbuild-stage1.yaml for current steps for stage1 images.

You can also run the build locally using docker.

docker build -t epoxy-images-builder  .

docker run --privileged -e PROJECT=mlab-sandbox -e ARTIFACTS=/workspace/output \
  -v $PWD:/workspace -it epoxy-images-builder /workspace/builder.sh stage1_minimal

docker run --privileged -e PROJECT=mlab-sandbox -e ARTIFACTS=/workspace/output \
  -v $PWD:/workspace -it epoxy-images-builder /workspace/builder.sh stage1_mlxrom

docker run --privileged -e PROJECT=mlab-sandbox -e ARTIFACTS=/workspace/output \
  -v $PWD:/workspace -it epoxy-images-builder /workspace/builder.sh stage1_isos

Using an ISO, you should be able to boot the image using VirtualBox or a similar tool. If your ssh key is in configs/stage2/authorized_keys, and the VM is configured to attach to a Host-only network on the 192.168.0.0/24 subnet, then you can ssh to the machine at:

ssh [email protected]

Deploying images

The M-Lab deployment of the ePoxy server reads images from GCS. The cloudbuild steps deploy images to similarly named folders:

• output/stage1_mlxrom/* -> gs://epoxy-mlab-sandbox/stage1_mlxrom/
• output/stage1_isos/* -> gs://epoxy-mlab-sandbox/stage1_isos/

BIOS & UEFI Support

The simpleiso command creates ISO images that are capable of booting from either BIOS or UEFI systems. BIOS systems use isolinux while UEFI systems use grub. These images should also work with USB media.

Testing USB images

VirtualBox natively supports boot from ISO images & supports BIOS or UEFI boot environments. To support VM boot from USB images we must create a virtualbox disk image from the raw USB disk image.

VBoxManage convertdd boot.fat16.gpt.img boot.vdi --format VDI

Then select that image in the VM configuration.

Upgrading Kubernetes components

Upgrading Kubernetes components on platform nodes is a separate process from upgrading them in the API cluster. Upgrading Kubernetes on platform nodes should always occur after the API cluster has been upgraded for a given project. The script ./setup_stage3_ubuntu.sh has logic which is designed to enforce this requirement, but it is still worth mentioning here.

Upgrading Kubernetes components on platform nodes should be as simple as copying the values for the identially named config variables from the k8s-support repository to the ones in ./config.sh in this repository:

• K8S_VERSION
• K8S_CNI_VERSION
• K8S_CRICTL_VERSION

Once the version strings are updated, and match those in the k8s-support repository, just follow the usual deployment path for epoxy-images i.e., push to sandbox, create PR, merge to master, tag repository. The Cloud Builds for this repository will generate new boot images with the updated Kubernetes components. In mlab-sandbox and mlab-staging, the newly built images will be automatically deployed to a node upon reboot. However, in production (mlab-oti) they will not be automatically deployed without further action. See the following section for more details.

Configure ePoxy to use a newer image version

In order to deploy the new boot images to production you will need to modify the ImagesVersion property of every ePoxy Host GCD entity to match the tag name of the production release for this repository. This can be done using the epoxy_admin tool. If you don't already have it installed, then install it with:

$ go get github.com/m-lab/epoxy/cmd/epoxy_admin

Once installed, you can update the ePoxy Host GCD entities in the mlab-oti project with a command like the following. NOTE: do not run this command against the mlab-sandbox or mlab-staging projects, as ImagesVersion is a static value in those projects and should always be "latest":

$ epoxy_admin update --project mlab-oti --images-version <tag> --hostname "^mlab[1-3]"

Trigger a rolling reboot

None of the nodes in any project will be running the updated images until they are rebooted. You can trigger a rolling reboot of all nodes in a cluster with a small shell command like the following:

$ for node in $(kubectl --context <project> get nodes | grep '^mlab[1-4]' | awk '{print $1}'); do \
    ssh $node 'sudo touch /var/run/mlab-reboot'; \
  done

The former command assumes you have ssh access to every platform node. It leverages the Kured DaemonSet running on the platform by creating the "reboot sentinel" file (/var/run/mlab-reboot) on every node, which tells Kured that a reboot is required. From there, Kured handles rebooting all flagged nodes in a safe way (one node a time).

You can check the progress and/or completion of the upgrade by looking at the kubelet version for a node as reported by kubectl:

$ kubectl --context <project> get nodes