This change adds longhorn, an addition to Kubernetes that adds the
ability to use distributed storage over all nodes to the cluster.
Note, that I tried that in December already but due to very high load on
the machines I rolled _everything_ back. Though, it turned out that the
high load was not because of longhorn, but instead because of bad
configuration of the server, as described in the see-also commit.
Reference: https://longhorn.io/
Reference: https://longhorn.io/docs/1.10.1/deploy/install/install-with-helm/
See-also: 4b8a3d12c4 Use etcd instead of sqlite for k3s-server
Oh damn, that was so annoying. My cluster ran on near 100% load all the
time! As it turns out that's a known issue on k3s clusters.
The solution is to add the `--cluster-init` flag to the server which let
the server use etcd instead of sqlite. And voila the cpu-usage drops to
a resonably low level in the single digit percent range.
Reference: https://github.com/k3s-io/k3s/issues/10396
Reference: https://docs.k3s.io/datastore/ha-embedded#existing-single-node-clusters
Bitnami has discontinued a lot of their container images. Alongside
these were also their images for a high availability setup of
postgresql. This change fixes that by referencing the legacy bitnami
images until a "new" approach is found.
This change is surprisingly tricky and needed some temporary
workarounds. First, there is no "official" snappass helm chart but I
found one, which does the job and looked good enough. The other problem
is the missing "official" image of snappass. The helm-chart used a
customized image which I didn't want to use, therefore I had to rebuild
a brand new image quickly. This new image is unfortunately not bound to
any repository or pipeline yet, which means that this change needs some
trust for the moment until I've set up the needed repo and CI
structures.
Reference: https://github.com/lmacka/helm-snappass/tree/main
Reference: https://github.com/pinterest/snappass
By applying this change the kubernetes cluster gets a gitea-server
setup. Note, that I use a custom-image which I have to automate in
future. The customization is necessary since I use asciidoc very often
and the default-gitea doesn't render these files, so it becomes a bit
cumbersome to read them on the web.
I'm oversaw completely, that I have to change the SSH-port for all nodes
in the cluster otherwise I cannot provide a meaningful load-balancer for
the git-ssh port in it.
Additionally this allowed me to fix some config errors which I simply
oversaw.
Since I don't have multiple terraform steps anymore it simply doesn't
make sense to me anymore to split all tasks into separate folders.
Instead I try to be as clear as possible in the README to make it easy
to follow the structure in the future without too much headache.
