3 ms·
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. B
by preya2k 4y ago
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?
- c0wb0yc0d3r 4y agoFor 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.
- imglorp 4y agoHA. In theory, if you have several boxes, workload can move around if a box or service go down.
- justsomehnguy 4y agoYou 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.
- imglorp 4y agoYeah 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?
- justsomehnguy 4y ago> 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.
- Havoc 4y agoMix of shiny toy, Standardisation and I believe Tim uses it at work anyway.
- moondev 4y agoThe 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.
- trilbyglens 4y agoThe 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.