Update July 2024: Ubuntu 22.04 Minimal image with Generic Kernel
Virtual Machine images provided by the Jetstream team are fully fledged and therefore quite large.
Julien Chastang proposed the shift to a minimalist distribution to save disk space, resources and shorten deployment time, see the relevant issue discussion on the Jetstream Gitlab organization.
We are mostly interested in replacing Ubuntu as the default image for our Kubespray developments, so we are looking only at OS supported by Kubespray
I previously tested the container-specific Flatcar image, now testing Ubuntu 22.04 Minimal.
Ubuntu 22.04 Minimal cloud image
Best features:
- Minimalist, image is less than
200 MB - It’s Ubuntu, everyone is familiar with it
- Wide support (for example GPU virtualization used in Jetstream should work)
- Automatic updates
Downloaded from https://cloud-images.ubuntu.com/minimal/releases/jammy/release/
The name of the image is Ubuntu2204Minimal, it is set as a Community image (script to load image to Openstack:
openstack image list --community | grep Minimal
| 0fa9f3b4-d29f-4f68-a8a7-16bf44ffae69 | Ubuntu2204Minimal | active |Ubuntu minimal version with Generic kernel
By default Ubuntu Minimal ships with the kvm kernel which is a smaller kernel, it does not offer some modules, for example nfsd is not available, this prevents running NFS servers on the node or even in containers inside the node.
To circumvent this issue, I created a new version of the Minimal image with the only modification being the generic kernel instead of the kvm one.
The name of the image is Ubuntu2204MinimalGenKernel, ID 671f03f1-4d86-4363-b03c-f5d54818693a
See https://github.com/zonca/jupyterhub-deploy-kubernetes-jetstream/issues/80
This image a bit larger image:
Filesystem Size Used Avail Use% Mounted on
tmpfs 297M 664K 297M 1% /run
/dev/sda1 20G 2.7G 17G 14% /
tmpfs 1.5G 0 1.5G 0% /dev/shm
tmpfs 5.0M 0 5.0M 0% /run/lock
/dev/sda15 105M 6.1M 99M 6% /boot/efi
tmpfs 297M 4.0K 297M 1% /run/user/1000GPU support
While Ubuntu potentially supports GPUs, the default image from Ubuntu does not. We need some specific customization on the image.
Therefore we would need to create a GPU-specific minimal version that also includes the GPU drivers.
Exosphere
The instance boots correctly on Exosphere, I can ssh into the instance as exouser.
It looks like Exosphere installs another 1.4GB of packages, so disk usage is higher compared to Horizon, anyway still tiny compared to a full-featured Ubuntu.
df -h
Filesystem Size Used Avail Use% Mounted on
/dev/root 20G 2.5G 17G 13% /
tmpfs 1.5G 0 1.5G 0% /dev/shm
tmpfs 600M 672K 599M 1% /run
tmpfs 5.0M 0 5.0M 0% /run/lock
/dev/sda15 105M 6.1M 99M 6% /boot/efi
tmpfs 300M 4.0K 300M 1% /run/user/1000Horizon
All normal, boots fine, usual login with the ubuntu user and the Openstack keypair.
Occupies minimal space:
Filesystem Size Used Avail Use% Mounted on
/dev/root 20G 1.1G 19G 6% /
tmpfs 1.5G 0 1.5G 0% /dev/shm
tmpfs 600M 460K 599M 1% /run
tmpfs 5.0M 0 5.0M 0% /run/lock
/dev/sda15 105M 6.1M 99M 6% /boot/efi
tmpfs 300M 4.0K 300M 1% /run/user/1000No warnings in dmesg.
Kubespray
The real objective of this test is to deploy Kubernetes on Jetstream with this smaller image, following the latest tutorial with minor modifications.
We can replace the official Ubuntu with the minimal image on Terraform by modifying in cluster.tfvars:
image = "Ubuntu2204Minimal"With the minimal image, Kubespray gives the error:
modprobe: FATAL: Module ip_vs_sh not found in directory /lib/modules/5.15.0-1058-kvm\n"the simple workaround is to set:
kube_proxy_mode: iptablesinstead of
kube_proxy_mode: ipvsin inventory/kubejetstream/group_vars/k8s_cluster/k8s-cluster.yml
CPU Nodes are fine, but GPU nodes as expected do not work.
k get nodes -o wide
NAME STATUS ROLES AGE VERSION INTERNAL-IP EXTERNAL-IP OS-IMAGE KERNEL-VERSION CONTAINER-RUNTIME
kubejetstream-1 Ready control-plane 20m v1.25.6 10.1.90.167 149.165.174.34 Ubuntu 22.04.4 LTS 5.15.0-1058-kvm containerd://1.6.15
kubejetstream-k8s-node-nf-cpu-1 Ready <none> 18m v1.25.6 10.1.90.186 <none> Ubuntu 22.04.4 LTS 5.15.0-1058-kvm containerd://1.6.15
kubejetstream-k8s-node-nf-cpu-2 Ready <none> 18m v1.25.6 10.1.90.5 <none> Ubuntu 22.04.4 LTS 5.15.0-1058-kvm containerd://1.6.15
kubejetstream-k8s-node-nf-gpu-1 NotReady <none> 18m v1.25.6 10.1.90.230 <none> Ubuntu 22.04.4 LTS 5.15.0-1058-kvm containerd://1.6.15Did not feel it was necessary to test JupyterHub, I’m confident it will work.
Let’s review the disk space occupied on a m3.medium worker with Ubuntu Minimal before running Ansible:
ubuntu@kubejetstream-k8s-node-1:~$ df -h
Filesystem Size Used Avail Use% Mounted on
/dev/root 58G 831M 58G 2% /
tmpfs 15G 0 15G 0% /dev/shm
tmpfs 5.9G 484K 5.9G 1% /run
tmpfs 5.0M 0 5.0M 0% /run/lock
/dev/sda15 105M 6.1M 99M 6% /boot/efi
tmpfs 3.0G 4.0K 3.0G 1% /run/user/1000and after Kubernetes has being installed (from 800 MB to 6.8 GB):
Filesystem Size Used Avail Use% Mounted on
/dev/root 58G 6.8G 52G 12% /
tmpfs 15G 0 15G 0% /dev/shm
tmpfs 5.9G 2.1M 5.9G 1% /run
tmpfs 5.0M 0 5.0M 0% /run/lock
/dev/sda15 105M 6.1M 99M 6% /boot/efi
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/1a777fcf86f2e982969b83f78302bb19427470995262032dabe0384af26d202b/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/1331d819ef46d1d4ee94cd1faa58c0965e6c63b25d5fd69bb86e5b1769c9041e/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/6c6a245269c66d8a7ba69d31e590e9e807d17d19acfe9d2a251daa6b73127b78/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/2ed3c7de52d4c63e3a4df8e00486f4fcde91634891708e88aaf506256c2637c7/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/03b1e64d75fd951f7aefa050746717baa3811636636254132d4d11758c1a239c/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/bda38bc2b547cd4bdeca1aa7653fdf6483da2fbb8b69abbda2a35a3d93b982a8/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/3288593c52f5a21fec4be1222326d22f65612c61135e289addab1fe11b423334/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/e9fd081509ca17913118ae4c69a80dc7cd50ea3a14469a90bc7e63ff5ff36c6f/shm
tmpfs 3.0G 4.0K 3.0G 1% /run/user/1000Here is instead the comparison with the default Jetstream Ubuntu 22 image:
Filesystem Size Used Avail Use% Mounted on
udev 15G 0 15G 0% /dev
tmpfs 3.0G 1.2M 3.0G 1% /run
/dev/sda1 58G 8.8G 50G 16% /
tmpfs 15G 0 15G 0% /dev/shm
tmpfs 5.0M 0 5.0M 0% /run/lock
tmpfs 15G 0 15G 0% /sys/fs/cgroup
/dev/loop0 64M 64M 0 100% /snap/core20/2264
/dev/loop1 92M 92M 0 100% /snap/lxd/24061
/dev/sda15 105M 6.1M 99M 6% /boot/efi
/dev/loop2 39M 39M 0 100% /snap/snapd/21465
tmpfs 3.0G 4.0K 3.0G 1% /run/user/1000
xxxxxxxxxxxx 9.8T 493G 9.3T 5% /softwareAfter having installed Kubernetes (from 8.8 GB to 13):
Size Used Avail Use% Mounted on
udev 15G 0 15G 0% /dev
tmpfs 3.0G 3.0M 3.0G 1% /run
/dev/sda1 58G 13G 46G 23% /
tmpfs 15G 0 15G 0% /dev/shm
tmpfs 5.0M 0 5.0M 0% /run/lock
tmpfs 15G 0 15G 0% /sys/fs/cgroup
/dev/loop0 64M 64M 0 100% /snap/core20/2264
/dev/loop1 92M 92M 0 100% /snap/lxd/24061
/dev/sda15 105M 6.1M 99M 6% /boot/efi
/dev/loop2 39M 39M 0 100% /snap/snapd/21465
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/c0fc748f6469ded859983f821e8f167691b52a529bdc5e9a1269646ab1d1466f/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/5c853382c912e5cddd3dc6c726bce09fdbec753a89b60b6cf334214c34216cbc/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/1d5f384f212ecf62aafa44e8cd1dbabf5bc0fa28aeb10759da9e372494d82661/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/7d8ab2baa7b00523538b0921b0fc6c9d6aee1d121db8752fdb38bdce877784b7/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/1669d88641a2eafd7dc3c841e9f71ef6014d61bd986bc527e899b2b972837279/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/7d0ded66f7e811fdb0fee58dba737494aca1e36510f8291d7ebca5f5a3ad9824/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/4ec65ad53b8ab3391a04cc2442fa3278b114fbe4791ca1b24cc7db213be8f896/shm
shm 64M 0 64M 0% /run/containerd/io.containerd.grpc.v1.cri/sandboxes/2bb33f80ec8afc1639dd57094dbbd9c59e97640cb10df557f68d3bd8582a10f1/shm
tmpfs 3.0G 4.0K 3.0G 1% /run/user/1000
xxxxxxxxxxxx 9.8T 493G 9.3T 5% /software