This is my walkthrough of everything I am self-hosting here in my HomeLab.

Glad to see you here! Watched a bunch of your videos over the last year and a half. Great content and great resources for setting up self hosted services, you helped me get my services up and available with traefik

Thank you for this. I've found your stuff very helpful while setting up my own homelab as well.

didn't see much about security

The "s" in homelab stands for security...

He has a few videos on security on his youtube channel.

Honest question (slightly unrelated): Why use k8s for this use case? In my naive understanding, k8s is useful when services need to dynamically scale up/down. But in a HomeLab there will probably never be more than one instance required, right?

For me personally, I want to get more comfortable with k8s, so I'm deploying to k8s. Till now it has been docker with a reverse proxy, all the way down.

HA. In theory, if you have several boxes, workload can move around if a box or service go down.

You need a shared storage for that.

In theory you can do the same with Docker Swarm too, except you need at least 5 nodes to sleep soundly.

Yeah shared storage is still going to be the hard part of home lab HA. Either you make a NAS with RAID or you use something redundant like Juicefs or Ceph split among your k8s nodes. Any better ideas?

> Any better ideas

Stop making a 25/7, 99.99% SLA homelab. Nobody needs HA in a homelab anyway. FO - sure, why not, especially if you have a spare hardware and spare money for the electricity.

Overall, a NAS/DAS/SAN with a RAID5/6/10 is more than enough to serve a multiple servers with the loads. You need a proper backup sooner than a HA storage. RAID10 on 4 HDDs was enough in 2012 for almost everything. RAID1 from 2 SSDs, even TLC/QLC/xLC - would be enough in 2022/2023 for almost everything, while not quite costly. Pair it with a RAID1 from a 2 4TB+ HDDs and you covered 99% of needs.

Forget Ceph and any other vSAN-like/distributed FS - they are for work, not for hosting your 3 Docker containers.

Mix of shiny toy, Standardisation and I believe Tim uses it at work anyway.

The k8s API is actually way more useful in a homelab than scaling and reliability aspects. Even a single node, as in the control plane is also a worker gets you the API. The sky is the limit after that.

The ability to throw a new pod into a cluster is great. There are lots of pre configured helm charts and whatnot making it easy to add things.