Move to TrueCharts for helm charts

[Kimi450]

Luckily, TrueCharts was an inital fork of k8s-at-home so the config as I had it works here by default. Only had to change the repo.

NOTE: arm architecture doesnt seem to be supported at the moment

[Kimi450]

So as I was working on this, I saw that TrueCharts also includes a bunch of other seemingly random manifests including a prometheus operator. This prometheus operator was interferering with the operator from kube-prometheus-stack's install. Found it out because of this post.

Need to maybe see if I can disable these manifests from being installed cause I dont need them

[Kimi450]

Even after removing the starter manifests install, custom CRDs are installed (no way around this at the moment)

I am still seeing high disk and CPU usage compared to the old setup.

• Disk has gone up from ~3.5% to 8-30% range.
• CPU has gone from 15% and 7.5% (~22%) to 19% and 10% (~30%) for user and system usage respectively
• More active TCP connections too

Reproduce

• Install minikube with this automation
• install old setup (see observations)
• install new setup (from this PR) and see the higher usage on grafana 1860 dashboard
• redo step 2 and 3 as a sanity on the same minikube install and the problem persists meaning this is just related to the TrueCharts helm charts.

Logs from the helm pods dont show anything interesting

Comparing the helm manifests (with helm get all on both releases, old setup and new) show nothing concerning either. I am confused :)

[Kimi450]

Todo:

• Can use --skip-crds to skip CRD install with the helm command
• PGID == fsGroup and PUID == runAsUser dont need to be set. Src

[Kimi450]

Relevant discord thread

[Kimi450]

I tried the following (all of em and different combinations as well and saw some coincidental improvements) StatefulSet changes

• remove fsGroupChangePolic
• remove supplementalGroups list
• set enableServiceLinks=true like in k8s-at-home
• remove env vars

ingress changes

• remove traefik
• remove ingress completely

service changes

• remove ipfamilypolicy enforcement (it has to be single stack for minikube apparently anyway)

But in the end, the high disk usage I saw was being I was being an idiot :)

I had Minikube set to use 2.5GB of RAM and that cause a lot of page swaps I think (or some other related stuff) causing the high disk utilisation leading to relavtively higher CPU usage too. Increasing it to 6 gigs fixed the issue. Not related to the limits in the values of the charts (cause theyre not requests).

Because the RAM usage at the host level showed as <50% util, I didn't even think about minikube not getting enough resources...

[Kimi450]

Jellyfin doesnt seem to allow hardware acceleration. Looking into it.

[Kimi450]

The jellyfin hardware accelleration issue was linked to not running the pod as admin, so I will run it was root now. That fixed the issue.