SELECT * FROM entries ORDER BY created_at DESC;I installed the cluster in its permanent location.
It now runs two apps and I've set up nginx ingress
controller
for an easy access over a domain name. Following the kube101
series,
it was straightforward to set up cert-manager and
obtain TLS with Let's Encrypt. I have some docs
about this process in the tls directory. I can now access grafana over
public internet. In future, I will explore how to limit the access based on an
IP and/or how to use VPS to mitigate some security holes I might be opening by
exposing services such as grafana.
Last week I stumbled onto a website which sells refurbished servers. I was hooked. However, I figured that paying a few hundred pounds was not a smart move as I had no idea what I was doing. Instead, I gathered some electronic scrap I found around my parents' house: my old laptop, pi, some old computers, cables tossed around in various dubious dark places, an old switch.
I experimented with Ubuntu server
20.04. I was pleasantly surprised with
the install wizard. It asked me for my Github username and then imported my SSH
keys. The wizard also suggested some default packages which helped me explore
my options. I installed microk8s and connected three
machines in a cluster. Straightforward. However, it drained my machines to
their bones. 40% CPU usage idle. It was the high availability
addon, which replicated all data
across 3 master nodes. I decided I would come back to this later and built an
ugly case which held all the machines.
For the effect, I included an old monitor.
I would be moving the cluster to another location, the basement, in a week. Hopefully by then I will have figured out a less eye damaging cable work.
Now that the part scavenging and the drilling (to mount the parts) was over, I
dove into k8s. The high CPU usage I mentioned annoyed me. I experimented with
k3s as an alternative to microk8s. After a few hours I still I could not
get machines to connect into a cluster. With microk8s it was smooth. I
assumed there would be more issues with k3s even if I solved this one, so I
reverted back and decided to turn off the HA cluster feature I mentioned. The
compromise is that if the master node went down, I would lose data. I wasn't
sure what data specifically, but I suspected that everything which runs on the
master and all resource configs information such as pods, jobs, volumes, etc.
k8s seemed overwhelming at first and I was frankly lost with the
kubectl
CLI. I watched some episodes of the Jeff Geerling's 5 star Kube101
series. 2 hours in and I learnt that it was
hip to alias kubectl to k. Equipped with this primal knowledge I ported my
app, suckless.hn, to the cluster as a cron
job. It took about 6
hours but having to write the config files, debug issues and setup the
environment was illuminating. Having some kind of feel for the aesthetic
zeitgeist k8s were born in, as of now I have cleared some uncertainty about
this technology going further.
Then I set up grafana, a data visualization tool.
Because who doesn't like pretty graphs. It feeds off with data from
prometheus,
and connects out of the box. All of that can be enabled with one command (see
MICROK8S.md).
All this was a lot of work and many silly mistakes. On the plus side, I hope to stop paying the AWS EC2 troll toll soon.
Next, I have to look into exposing the cluster access to public internet. I access the machines over local IPs with my ssh key for now, but I will soon be going back to my place outside of the cluster's local network. I have a public IP at my parents' house, this won't be an issue. However, how to securely expose grafana and the k8s dashboard? Stay tuned.