What's your approach to databases?

I just run one mariadb container via docker-compose that all my other services use as their database.

version: "2"
services:
  mariadb:
    image: lscr.io/linuxserver/mariadb:latest
    container_name: mariadb
    environment:
      - TZ=####/####
      - PUID=###
      - PGID=###
      - MYSQL_ROOT_PASSWORD==############
    volumes:
      - /docker/mariadb:/config
    ports:
      - 3306:3306
    restart: unless-stopped

Off-topic but I don't really get the appeal in running Kubernetes (or similar technologies) in a homelab. Unless it's something you want to learn for work of course.

I'm running kubernetes simply because the other options are worse.

Proxmox takes to many resources.

Docker Compose caused countless issues for me when running multiple services (especially network related).

Bare metal is annoying, because you're forced to keep all the services in lockstep, dependency wise.

I'm using kubernetes at with, the overhead is rather small (with k3s) and mostly it's working pretty great.
- Use Podman with Systemd & Quadlet. Like bare-metal but without the annoyances you mention.
- As a bonus, you can just join multiple machines to the cluster and have work spread out over them.
- I recently switched to nixos which makes dependency management and configuration itself much easier. Probably the best option to run things on bare metal IMO.
- That's funny to hear as daily for work I use k3s and RKE2 for deployments and testing and at home I use unraid specifically because of all the k3s work I do even k3s has too much overhead for updates and backups and all that IMO.
That, and you have to take into account each person’s available hardware and resources.

I have an under powered 10 year old desktop, a resonably specd 5 year old laptop with a busted screen, and 8 Raspberry Pi’s (3s and 4s). And can’t currently afford better hardware.Sometimes clustering those Pi’s makes sense.

You can use whatever you have to hand.
- That's a great point I hadn't considered tbh! And that learning new technologies even if there is no "purpose" to it can be... fun! :)
I don't like Docker as a company, the networking seems unnecessarily obtuse to me, and k3s is a smaller version of k8s, which is here to stay in my opinion (has a bigger learning curve though), and will help me in my career. Those would be my reasons, but if someone doesn't have a use for k3s I suppose there's not much of a point, considering everything is still written for docker

I'm a big fan of the zalando postgres operator. A lot of the critical features you'd want in production databases are handled and very nicely abstracted.

https://github.com/zalando/postgres-operator

Did they get it working with multi arch setups? I have a few pi’s in my cluster and last time I looked at using that it wasn’t ready for arm64
- I'm not sure, actually. My personal cluster is all x86 so I'm not usually that aware of the multiarch stuff. 😬

Are we talking database schema migrations or migrating a database between Postgres instances?

If it’s the former, the pattern is usually to run them in init containers or Jobs but I have been wanting to try out SchemaHero for a while which is a tool to orchestrate it and looks pretty neat.

ETA: Thought I was replying to your below comment but Memmy deleted it the first time for some reason, my bad.

It's about PostgreSQL upgrade.

The "pattern" there is to either dump and reinsert the entire DB or upgrade by having two installations (old and new version), which doesn't exactly work well in k8s. It's possible, but seems hacky
- I can’t think of any situation other than maybe wanting to get better indexing or changing the storage engine that I would need to re-create and re-insert that way so I’m not sure if you have a constraint that necessitates that or not but now I’m curious and I am always curious to find new or better methods so why do you do it that way?
  
  At home to upgrade Postgres I would just make a temporary copy the data directory as a backup and then just change the version of the container and if it’s needed run pg_upgrade as jobs in kubernetes.
  
  In a work environment there is more likely to be clustering involved so the upgrade path depends on that but it’s similar but there really isn’t a need to re-create the data, the new version starts with the same PVCs using whatever rollout strategy applies. Major version upgrades can sometimes require extra steps but the engine is almost always backwards compatible at least several versions.
- I've always used this docker image to do pg upgrades. It runs pg_upgrade to recreate the system tables and copy the user tables (which normally don't have any storage changes). It does require that the database isn't running during the upgrade so you're going to have a bit of downtime. Make sure you redo any changes to any configuration files, especially pg_hba.conf

https://cloudnative-pg.io

I have a single database server because I can't afford two servers with high storage. The servers that need access to it connect over wireguard VPN. This is slow as f**k don't do that.

I google why doesn't mysqld work?, then copy paste terminal commands from the first result, then google why doesn't my machine boot? then turn around 360 degrees and walk away.

then turn around 360 degrees and walk away.

And how does that work for you?
- I imagine they feel like they’re not getting anywhere.

I avoid software which requires a relational database altogether. For me that’s part of the fun of self hosting: what’s the simplest possible system I can get away with at my tiny scale?

Cautiously.

Own vm as regular daemon + acme.sh for tls.

Makes K3s more fun if db's are outside and files (when possible) are on S3 (Minio Docker on Synology).

For the rest than pvcs with longhorn as storage driver.

Never tried it but kubegres seems like a good implementation for kubernetes. I guess if you just have a single-node cluster there won't be much benefit but it seems a periodic backup to NFS is key (you can run NFS on most anything).

What currently pisses me off is the fact that it's almost impossible to do proper migrations for Postgres in k8s. I'd have to look into kubegres, but all approaches I've seen so far involve basically copying the entire PVC and the data inside into a new structure - and doing so involves hacked together scripts.

For personal use, I don't bother with databases on k8s. They are waaay easier to manage if you just let your host distribution run it as a regular service and Upgrade it through that

using mostly operator from percona for kubernetes, sometimes just a simple deployment. Running postgresql for Lemmy from docker-compose as a container.