Installing to a local machine

Installing Charmed Kubernetes on a single machine is possible for the purposes of testing and development.

However, be aware that the full deployment of Charmed Kubernetes has system requirements which may exceed a standard laptop or desktop machine. It is only recommended for a machine with 32GB RAM and 250GB of SSD storage.

<span class="p-notification__title">Note:</span>
<p class="p-notification__message">If you don't meet these requirements or want a lightweight way to develop on pure Kubernetes, we recommend  <a href="https://microk8s.io/">MicroK8s</a></p>

In order to run locally, you will need a local cloud. This can be achieved by using lightweight containers managed by LXD. LXD version 3.0 or better is required.

1. Set up LXD

If LXD has not previously been installed

LXD 3.0 or above should be installed from a snap and configured for Charmed Kubernetes

Install LXD

sudo snap install lxd

Run the LXD init script

/snap/bin/lxd init

The init script itself may vary depending on the version of LXD. The important configuration options for the installer are:

• Networking: Do NOT enable ipv6 networking on the bridge interface
• Storage Pool: Use the ‘dir’ storage type

You can now move on to the next step

If LXD is already installed

If you installed LXD from a snap, you can skip this step (but if necessary, you may need to alter the default profile). If your system had LXD pre-installed, or you have installed it from the archive (i.e. with apt install), you will need to migrate to the snap version.

If you aren’t sure whether LXD is installed, you can check installed snaps with:

snap list | grep lxd

and installed deb packages with:

dpkg -s lxd |  grep Status

If you do have the deb version of LXD installed, you should migrate to the snap version after it has been installed. The snap includes a script to do this for you:

sudo snap install lxd
sudo /snap/bin/lxd.migrate

This will move all container specific data to the snap version and clean up the unused Debian packages, which may take a few minutes.

If LXD was installed, but never used, there will be no data in the default profile, so you should now initialise LXD:

sudo lxd init

Currently, Charmed Kubernetes only supports dir as a storage option and does not support ipv6, which should be set to none from the init script. Additional profiles will be added automatically to LXD to support the requirements of Charmed Kubernetes.

2. Install Juju

Juju should be installed from a snap:

sudo snap install juju --classic

Juju comes preconfigured to work with LXD. A cloud created by using LXD containers on the local machine is known as localhost to Juju. To begin, you need to create a Juju controller for this cloud:

juju bootstrap localhost

Juju creates a default model, but it is useful to create a new model for each project:

juju add-model k8s

3. Deploy Charmed Kubernetes

All that remains is to deploy Charmed Kubernetes. A simple install can be achieved with one command:

juju deploy charmed-kubernetes

This will install the latest stable version of Charmed Kubernetes with the default components and configuration. If you wish to customise this install (which may be helpful if you are close to the system requirements), please see the main install page.

Next Steps

Now you have a cluster up and running, check out the Operations guide for how to use it!

Troubleshooting

I get an error message when running lxc or lxd init

The most common cause of this message:

Error: Get http://unix.socket/1.0: dial unix /var/snap/lxd/common/lxd/unix.socket: connect: permission denied

…is that either you have not run lxd init, or you are logged in as a user who is not part of the lxd group (the user installing the snap is automatically added).

To add the current user to the relevant group:

sudo usermod -a -G lxd $USER

You may need to start a new shell (or logout and login) for this to take effect:

newgrp lxd

Confirm CNIs do not need specific kernel parameters unsupported by the lxd-profile

If the CNI pods fail to start, see notes on the specific CNI page.

CNIs like Cilium and Calico need access to /sys/fs/bpf, but that mountpoint is not supported by the juju’s validation check for the charm specific lxd-profile.yaml. See CNI Overview for more details.

Services fail to start with errors related to missing files in the /proc filesystem

For example, systemctl status snap.kube-proxy.daemon may report the following:

Error: open /proc/sys/net/netfilter/nf_conntrack_max: no such file or directory

This is most commonly caused when lxd-profile.yaml is not applied. Verify the profile in use by the kubernetes-worker charm:

lxc profile list
lxc profile show juju-[model]-kubernetes-worker-[revision]

Identify any missing fields from the above lxd-profile.yaml file and add them as needed with:

lxc profile edit juju-[model]-kubernetes-worker-[revision]

You may need to remove and re-add the affected unit for the changes to take effect:

juju remove-unit kubernetes-worker/[n]
juju add-unit kubernetes-worker

Kubelet fails to start with errors related to inotify_add_watch

For example, systemctl status snap.kubelet.daemon.service may report the following error:

kubelet.go:1414] "Failed to start cAdvisor" err="inotify_add_watch /sys/fs/cgroup/cpu,cpuacct: no space left on device"

This problem usually is related to the kernel parameters, fs.inotify.max_user_instances and fs.inotify.max_user_watches.

At first, you should increase their values on the machine that is hosting the Charmed Kubernetes installation:

sysctl -w fs.inotify.max_user_instances=8192
sysctl -w fs.inotify.max_user_watches=1048576

Then the new values should be applied to the worker units:

juju config kubernetes-worker sysctl="{ fs.inotify.max_user_instances=8192 }"
juju config kubernetes-worker sysctl="{ fs.inotify.max_user_watches=1048576 }"

Calico is blocked with warning about ignore-loose-rpf

Calico may be blocked with status: ignore-loose-rpf config is in conflict with rp_filter value.

If the kernel net.ipv4.conf.all.rp_filter value is set to 2, Calico will complain, because it expects the kernel to have strict reverse path forwarding set (ie. value be 0 or 1) for security. In LXD containers, it’s not possible to manipulate the value; it’s dependent on the host. In this situation we can set the charm config ignore-loose-rpf=true.

juju config calico ignore-loose-rpf=true