If you've read this far, you're already asking "OMG, why?!". Here's why: I want to iterate on building a small Kubernetes cluster with a variety of tools and configurations, and it's only for prototyping.
Some folks do this with a small array of Raspberry Pi hardware, and that's fine, but I don't love blinking LEDs that much and I'm already sitting on plenty of gear: the smallest machine I use on a daily basis has 16GB of RAM, and the largest is totally ridiculous.
You may need to install some prerequisite applications. Give the following a shot:
sudo ./INSTALL-PREREQS.sh
The script only supports Debian and Red Hat based systems today, and very loosely at that. Let me know how we can make that better.
Open monokube.env and update the variables as necessary. There are
a number of IMAGE_URL variables: uncomment the one you want and leave
the rest commented. Feel free to add your own, too!
Run ./monokube-build.sh If you don't have all the required binaries
it will stop and ask you to correct that. It'd be great to have a
list of required packages for Debian and RHEL variants here - pull
requests accepted.
Once the command completes, you'll have a ./cluster directory with a
.ssh directory and key inside of it. There are also metadata
directories for each of the newly created nodes. We use cloud-init's
local datastore option to configure the machines on boot, so you
should be able to log in with the username monokube and the
./cluster/.ssh/id_rsa key. Too bad you don't know the machine IP
addresses yet.
Run ./monokube-update-metadata.sh and once complete you should
find a file named ./cluster/inventory. I'm not going to force you
to use Ansible to configure the hosts, but I'm using Ansible so the
project generates an Ansible inventory file. Maybe that becomes
a configuration option in the future. For now, if you'd like to
configure ansible run cd ./ansible && ./setup.sh. Once done, run
./venv/bin/activate followed by ansible all -a w to test a command.
There's a README.md in the ansible directory with more information.
If you want to keep tabs on your hosts while you work, check out:
./metrics/start-influxdata-sandbox.sh and make sure to install
telegraf on the nodes. (It's included in the
./ansible/playbooks/bootstrap.yml playbook)
Now do what you need to do to test a Kubernetes installation.
Once you're done, you can run ./monokube-destroy.sh.
Warning: I've only tried the Ubuntu images thus far. Report bugs if
you find issues with the others. We can also attach a text file as a
serial console to the hosts with something like
--serial file,path=/tmp/${node_name}.log and that might help debug
issues.
So how should I install Kubernetes?
I don't know yet, check back in a bit.
This feels more like a general purpose cluster maker
Yeah, kinda. Maybe there's an abstraction waiting to be extracted.
Wow, that's a whole lot of bash
That's not really a question, but I get where you're coming from. I believe in hacks before systems, and if this project survives maybe it'll grow to a "real" programming language.