When we talk about Kubernetes and the whole Cloud Native ecosystem, it usually goes without saying that we're also talking about high availability and redundancy. In the case of Kubernetes, if a server in the cluster fails, the pods running on it are moved to another node for continued service. So, this is a fun story of me going against the ethos of high availability in Kubernetes.
If you don't care about my storytelling, this is basically me adding a 3TB disk to Longhorn, so I can favour backups over redundancy.
Longhorn
I do storage through Longhorn, which achieves high availability by replicating data across my disks. So, I have 3 nodes, each with 512GB disks. If I create a 20GB PVC in Kubernetes, Longhorn will create 3 replicas (by default) on my nodes. Hence, if one node dies, the pods will be relocated and will have access to the 2 other replicas so that they can keep working.
The downside for some of my needs is that I may need a 100GB PVC for my Plex library. But if I do that, then I have essentially blocked out 300GB from the 1.5TB disk collection, which hurts, even though I understand it's for a good cause. In another case, I may want a 400GB PVC, in which case I need extra hard disks connected to my nodes. But still, to achieve my 3 replica goal, I need to buy 3 disks! Talk about an expensive hobby!
Adding a Disk
Since my nodes are USFF and accept only 2.5" disks, on top of the NVME slots, I had to go out and buy a hard disk enclosure for my 3.5" disk, with an external power supply and a USB 3 connection to a machine of my choice.
I have 2 disks lying around, one is a 3TB WD Purple surveillance drive, and the other is a Seagate 8TB drive. The purple drive has been lying around for a while and I didn't yet want to commit the 8TB drive to my project. So, 3TB will be fine for now, and that's what went in the enclosure.
"But Alex-", I hear you say, "there's no redundancy!". And to that I reply "then buy me more disks". The PVs going on the purple drive will have just one replica, but since everything is in one place on Longhorn, they will be backed up in the same nightly schedule as my other PVs. Which is miles better than wiring it up to my only monolith machine where I, as a professional, don't do backups. So, while I still have singular points of failure (the node, enclosure, or disk), at least it's easier for me to manage the data.
The bottom line, I would rather go through a 2 hour long recovery process than spend more money and "waste" storage just so non-"mission-critical" data can have replicas.
